# OpenAlly: product context and articles > OpenAlly lets business owners describe what they need and get an app they can use, starting from a template on their phone. The store, stock, orders and invoices share one set of records, so every new app starts with what is already there. Model and provider choice are supporting controls; messaging services are integrations, not the product definition. ## Reading notes - Product availability and platform support should be read from the current feature pages. - The optional Aster companion supplies Android screen control; OpenAlly remains a cross-platform product. - WhatsApp Business (a business number connected through Meta; free to connect, AI replies with Premium) is separate from linking a personal WhatsApp account as a chat channel, and from OpenAlly Miro, OpenAlly's own WhatsApp number for help with OpenAlly. - Web to App builds Android apps on an Android phone or a Mac; the owner installs or submits the files, and OpenAlly does not publish to Google Play. - Every article includes its review date, disclosure, and public source list. ## Product pages Hub: https://openally.ai/product lists every feature page. ### Web to App URL: https://openally.ai/product/web-to-app OpenAlly Web to App turns a website you run into a signed Android app: an APK to install or share and an AAB for your own Google Play upload, built on an Android phone or Mac. Free on every plan. Facts: - OpenAlly Web to App turns a website at a fixed https address, or an app made in the App Builder, into an Android app that is packaged and signed on the owner's Android phone or Mac. - Each build gives a signed APK to install or share, and an AAB for the owner's own Google Play upload once the app's signing key is backed up. - A build takes as little as 10 seconds, usually under a minute; the first build also creates the app's signing key. - OpenAlly reads the site's name, description, colours and icons as suggestions the owner can change, and never reads the content of its pages. - The apps it makes run on Android 7.0 and newer. Web to App builds on Android and on the Mac app; it does not make iPhone apps. - OpenAlly never installs the app or uploads it to Google Play: the owner saves or shares the APK, and uploads the AAB in their own Play Console. - Every app shows a small Made with OpenAlly line on its welcome screen, and its package id, which starts with ai.openally.app, is assigned when the app is created and never changes. - Camera, microphone, location, notifications and clipboard reading start off, and so does the newer page surface. Only the owner can switch those on; their AI can switch them off. - Addresses with a port, an IP address, a username and password, a ? or # part, a local-network name or a temporary tunnel host are refused, because the app keeps opening that address for as long as it is installed. - An App Builder app needs a fixed https address first: published from the phone at a name it keeps, with open access, or hosted in the cloud with a Premium plan, which serves static apps. - A page inside the app can reach the phone through the App2Web SDK, window.OpenAllyApp2Web: 51 methods across 16 namespaces and 9 event streams, each capability switched on by the app's owner. - The owner can also ask their AI in a chat: with the Web to App skill it checks the site, creates the app from the link, starts an Android build and hands over the files. Asked from a messaging channel such as WhatsApp or Telegram before an Android phone has the one-time app component, it replies with the steps to download it. - Web to App is free on every plan. How to: How to turn your website into an Android app with OpenAlly Paste your site's address, check its name and icon, build, then save or share the files. It all happens in the Web to App mini-app on an Android phone or Mac. 1. Paste a link: Paste the fixed https address of a site you run. A path is fine; a temporary tunnel or a link that carries a token is not. 2. Name and icon: OpenAlly brings your site's name, colours and icon with it. Change any of them, write a welcome line, or create an icon with AI. 3. Build: Build a signed APK, plus an AAB once your signing key is backed up. Building and signing stay on your Android phone or Mac. 4. Save or share: Save the APK and tap it to install, share it with anyone, or upload the AAB to your own Google Play listing. Questions: - Q: How long does a build take? A: As little as 10 seconds, usually under a minute. The first build also creates your app's signing key, and the first build on an Android phone downloads a small one-time app component. - Q: What do I get from a build? A: A signed APK you can install on Android 7.0 and newer or share with anyone and, once your signing key is backed up, an AAB, the file Google Play takes, for your own Play Console upload. Files are named after your app and version, like rekha-pottery-1.0.0-1.apk. - Q: Does OpenAlly publish my app or install it for me? A: No. OpenAlly builds and signs the files and hands them to you. You install the APK yourself by tapping it in Downloads, and you upload the AAB to your own listing in Google Play Console. - Q: Why does the AAB need a key backup? A: Uploading an AAB ties your Play listing to your app's signing key for good, and that key lives only on your device. So OpenAlly asks you to back it up once, to a file protected by a passphrase you choose, before it builds an AAB. Lose a key with no backup and no later version can be accepted as an update. - Q: Which web addresses work? A: A fixed https address you run, with or without a path, such as https://rekha-pottery.example.com/shop. Addresses with a port, an IP address, a username and password, a ? or # part, a local-network name or a temporary tunnel are refused, because the app keeps opening that address for as long as it is installed. - Q: Does the app work offline? A: It opens your live site, so it needs a connection. Without one it shows an offline screen with a Retry button, and reloads by itself when the connection comes back. - Q: Can I remove the Made with OpenAlly line? A: No. Every app made with Web to App shows a small Made with OpenAlly line on its welcome screen while your site loads. Your name, colours and icon are everywhere else. - Q: Can I choose my own package name? A: No. OpenAlly gives each app a package id that starts with ai.openally.app when the app is created, and it stays the same for the life of the app. That is what lets every later version install as an update. - Q: Is Web to App a Premium feature? A: No. Web to App is free on every plan. [Premium](https://openally.ai/premium) helps around it: it can host an App Builder app at a fixed address in the cloud, and its image credits can draw your app's icon with AI. - Q: Can the app use the camera, location or notifications? A: Yes, when you switch them on. Camera, microphone, location and notifications start off, and each switch also decides whether the built app asks for that permission at all. Only you can switch them on; your AI can switch them off. Notifications are ones your page posts on the phone. - Q: Does it work on iPhone? A: Web to App builds on your Android phone or your Mac, and the apps it makes run on Android 7.0 and newer. It does not make iPhone apps. - Q: Can my website's own code use the phone? A: Yes. With the App2Web SDK a page inside your app can pick photos, save to Downloads, read the location, share, post notifications and more: 51 methods, each capability switched on by you. [The App2Web SDK section of the Web to App page lists every method](https://openally.ai/product/web-to-app#sdk). App2Web SDK, for web developers: window.OpenAllyApp2Web, SDK 1.0.0, 51 methods across 16 namespaces, 9 event streams and 18 error codes. When the app's owner sets the Web-to-app bridge to Classic and new (the default is Classic only), the app injects it at document start, top frame only; there is nothing to install or import. Reference: https://openally.ai/product/web-to-app#sdk Namespaces: app (3), device (3), ui (13), haptics (1), navigation (3), location (2), permissions (3), media (3), files (5), share (2), clipboard (3), notifications (2), network (2), battery (2), storage (2), lifecycle (2). Rules: feature-detect with supports() every time; await every call and branch on error.code, never the message; a capability is switched on by the app's owner when the app is built; only the top frame gets the SDK; await whenReady() before the first call. Events: lifecycle, network, battery, insets, keyboard, orientation, colorScheme, location, notification. Error codes: E_UNAVAILABLE, E_DETACHED, E_FRAME_UNTRUSTED, E_SUBFRAME_UNSUPPORTED, E_UNSUPPORTED_METHOD, E_FEATURE_DISABLED, E_PERMISSION_NOT_DECLARED, E_PERMISSION_DENIED, E_PERMISSION_BLOCKED, E_CANCELLED, E_INVALID_ARG, E_TOO_LARGE, E_TIMEOUT, E_POLICY_REFUSED, E_BUSY, E_ABANDONED, E_IO, E_INTERNAL. By design there are no device identifiers, no file paths, no way to close the app, no probe for other installed apps, and no web push, background work, contacts, SMS, calendar, Bluetooth, NFC or biometrics. ### WhatsApp Business URL: https://openally.ai/product/whatsapp-business An OpenAlly AI agent on your WhatsApp Business number, connected through Meta, for sales, bookings and support; free to connect, AI replies with Premium; Meta bills messages directly; you approve billable messages and can take over any chat. Not the personal WhatsApp channel, nor OpenAlly Miro. Facts: - WhatsApp Business in OpenAlly puts a business's own agent on its WhatsApp Business number, for sales, bookings and customer support. Messaging services are integrations: this is one way a business's agent reaches its customers. - Connecting and managing a WhatsApp Business number is free on every OpenAlly account; AI replies on that number are included with every Premium plan, Lite included. OpenAlly adds no per-message fee. Meta bills the business directly for message charges, by message type and country, and Meta's invoice is final. - Three setup modes: New business number, Existing Cloud API account, and Use with the Business App, which keeps the WhatsApp Business app in use for accounts Meta finds eligible. The number connects through Meta's own signup in the browser. - Connecting a number does not turn AI on. The owner assigns an agent, reviews its access, sends a test message, then enables AI replies. - The agent runs on the owner's designated phone or Mac, which keeps the inbox history and checks for new messages about every 30 seconds while OpenAlly is running on it; background running varies by platform. Replies depend on that device being connected. - A dedicated OpenAlly delivery service receives messages from Meta, checks Meta's signature, keeps each message encrypted and deletes it once the device has it, or after seven days at most, and passes approved replies back to Meta. It never runs the agent. - The model is the owner's choice: a model on the device, the owner's own AI provider, or OpenAlly cloud AI. - The agent can send text up to 4,096 characters, up to 3 reply buttons, lists of up to 10 choices, a website button, product cards from a linked Meta catalog, a published WhatsApp form that loads no data from the business's own server, media already in the conversation, and approved templates. - Before every send: the 24-hour reply window (only a customer message opens it; outside it only an approved template goes out), recorded consent for marketing, STOP and START opt-outs, who is handling the conversation, the designated device and the price are checked. - Billable AI messages wait for the owner's approval of the exact message and its estimated Meta charge, unless a budget covers them: named templates, purposes and conversations, in one currency, up to a set total, for up to 90 days. - The owner can take over any conversation and return it to AI. The agent hands a conversation to the owner when it should not settle it alone, and after three failed attempts it moves to the owner automatically. - People messaging the number get customer-level access: no owner controls, only the business tools the owner grants. A send with an uncertain outcome is never repeated. - If Premium ends, AI replies pause; the owner keeps reading and replying themselves, and history, export, takeover and disconnect keep working. - Templates, the business profile, catalogs and messaging limits stay in the owner's Meta account. Meta decides eligibility and approves templates. - Three different WhatsApp things: WhatsApp Business (the business's own number through Meta; free to connect, AI replies with Premium), the WhatsApp channel (a personal account linked by QR code, which is unofficial access), and OpenAlly Miro (OpenAlly's own number for help with OpenAlly, which never messages first). How to: Connect your WhatsApp Business number to OpenAlly Connect a business number through Meta's signup, started from WhatsApp Business in OpenAlly's Hub. It takes a few minutes, once. 1. Install OpenAlly on your phone or Mac and sign in: Get OpenAlly from the download page and sign in to your account. 2. In Hub, open WhatsApp Business: Connecting is free on every account. AI replies are part of Premium, and you switch them on after the number is connected. 3. Pick your setup mode and tap Continue with Meta: New business number, Existing Cloud API account, or Use with the Business App to keep the WhatsApp Business app. Meta's signup opens in your browser. 4. Sign in to Meta, pick or create the business, and choose the number: A few minutes, once. Meta confirms the number and the connection lands back in OpenAlly on its own. Questions: - Q: Is WhatsApp Business in OpenAlly included with Premium? A: Connecting is free on every OpenAlly account: connect your business number through Meta's signup, name the inbox, read and reply yourself, manage the number, export and disconnect. AI replies on that number are included with every Premium plan: Lite, Plus, Pro, Ultra and Max. OpenAlly adds no per-message fee; model usage is billed by your own AI provider, or OpenAlly cloud AI uses your plan's credits. Meta's message charges are separate: Meta bills your business for them directly. - Q: Who pays for the messages? A: Meta bills your business directly for message charges, by message type and country, through the payment method in your Meta account. Before a billable AI message goes out, OpenAlly shows its estimated Meta charge for you to approve, unless an automation budget you set already covers it, and Meta's invoice is final. Model usage is separate: your own AI provider bills you, or OpenAlly cloud AI uses your plan's credits. - Q: Do I need a new number? A: No. Choose New business number to create or select a business portfolio, WhatsApp account and number through Meta. Choose Existing Cloud API account to connect numbers you already manage, or Use with the Business App to keep using the WhatsApp Business app if Meta finds your account eligible. - Q: Can I keep using the WhatsApp Business app? A: Yes, for accounts Meta finds eligible. Choose Use with the Business App: you keep using the WhatsApp Business app and review available history and consent before importing. Imported messages never start an AI reply, and a reply you send from the WhatsApp Business app hands that conversation to you. Other accounts may need to migrate first. - Q: Does the AI reply as soon as I connect? A: No. Connecting a number does not turn AI on. Assign an agent to the inbox, review its access, send a test message from another phone, then turn on AI replies when you are ready. - Q: Can I step in? A: Yes. Tap Take over on any conversation and reply yourself; AI replies pause for that conversation until you tap Return to AI. The agent also hands a conversation to you when it should not settle it alone, and after three failed attempts it moves to you automatically. - Q: Where does my agent run, and where are messages kept? A: Your agent runs on the phone or Mac you designate, and that device keeps the inbox history. Meta delivers messages to a dedicated OpenAlly delivery service, which checks Meta's signature, keeps each message encrypted and deletes it once your device has it, or after seven days at most, and passes your approved replies back to Meta. That service never runs your agent. The [privacy policy at openally.ai/privacy](https://openally.ai/privacy#whatsapp-business) has the details. - Q: How quickly are messages answered? A: Your designated device checks for new messages about every 30 seconds while OpenAlly is running on it, and your agent answers from there. Background running varies by platform, so keep that device online with OpenAlly running for timely replies. While it is offline, new messages wait, encrypted, in the delivery service for up to seven days. - Q: What happens if Premium ends? A: AI replies pause until you renew. Everything else keeps working: your number stays connected, and you can still read conversations, reply yourself, export them, take over and disconnect. - Q: Is this the same as connecting WhatsApp by QR code? A: No. WhatsApp Business connects your business number through Meta's signup; connecting is free, and AI replies come with Premium. The WhatsApp channel links your personal account by scanning a QR code; that is unofficial access, and it can lead to account limits or a ban. OpenAlly Miro is OpenAlly's own WhatsApp number for help with OpenAlly, and it never messages first. - Q: Can the agent send promotions? A: Only within Meta's rules. You create templates and Meta approves them in your Meta account; the agent sends the exact approved version, and a marketing message only to a customer with recorded consent. STOP or UNSUBSCRIBE opts a customer out and START opts them back in, checked before every send. OpenAlly does not run bulk campaigns. ## Plans and pricing ### OpenAlly Premium URL: https://openally.ai/premium OpenAlly Premium: the Lite, Plus, Pro, Ultra and Max plans, from US$3.99 or ₹249 a month, tax included. What each plan includes, how to pay and cancel, and what stays free on every plan. Facts: - OpenAlly is free. Premium is optional: the Lite, Plus, Pro, Ultra and Max plans add OpenAlly cloud AI with no provider key, which judges each request and answers it with the lightest model that will do it justice, image generation, AI replies on WhatsApp Business, device actions from ChatGPT (the rest of the connector stays free) and publishing to a hosted address. - Plus and above also add an email address for your agent, with 50 MB of mail and 200 sends a day. Lite is sold without it. - Monthly list prices start at US$3.99 internationally and ₹249 in India, tax included. Many countries pay a lower regional price than the US dollar list, and the app shows yours before you confirm. - Plans are chosen in the OpenAlly app. On Android, Google Play bills the plan. On a Mac, checkout runs through Razorpay. - Max yearly (US$1,499.99, or ₹99,999 in India) is sold in a WhatsApp chat with OpenAlly Miro rather than an app store: message +91 87792 22606 on WhatsApp, and once the buyer's own WhatsApp number is linked to their OpenAlly account, Miro sends the checkout link. - Plans have no trial period. A plan bought in the app renews each month or year until cancelled; cancelling stops the next payment and the plan keeps working to the end of the period already paid for. All payments are final; statutory rights are unaffected. - Credits refresh each billing period with a daily limit, and unused plan credits do not carry over. Scheduled runs and channel replies can use only part of that daily limit, the plan's automation credits; when they reach it they pause and chat keeps working. Images you ask for use the plan's everyday images first, then separate image credits. - A paid plan can add credit packs of 2,000, 5,500 or 12,000 credits in the app, used after your plan's allowance, spent only while you have a paid plan, and valid for twelve months. - Web to App is free on every plan. A plan adds a hosted address, one way to give an App Builder app the fixed https address packaging needs, and image credits for an AI-drawn icon. - App Builder is free on every plan: describe a site or a small app, OpenAlly writes and compiles it on your own phone or Mac, and you change it by asking. Previewing it, sharing it on your Wi-Fi and publishing it to the internet from an Android phone or a Mac, at a reserved address the app keeps, are free too; the device serves that address, so it sleeps when the device does. - A plan hands a built app to OpenAlly to build, upload and serve at a name you choose on openally.shop or mywebonline.store, so it stays online when your device is off: 1 app on Lite, 1 on Plus, 3 on Pro, 5 on Ultra, 10 on Max. One hosted app means one app however many parts it publishes, and releasing an app frees its place. - An App Builder app can be connected to the stock, outlets and payment account you keep in Business, and to your booking, payment or restaurant rules. Publishing it to a public link from your own Android phone or Mac is free on every plan: the link goes out through a tunnel and is online only while that device is serving it. A plan's hosted address keeps it up when the device is off. - The Business app also makes a website from a template for a shop, a restaurant or a salon, drafted by OpenAlly from one of your inventories. Prices and stock come from your catalogue, orders placed there show under Website orders, and on the free plan it shows a small Made with OpenAlly badge and lists up to the number of products the plan allows. - Your own model, on-device privacy, local memory, the full workspace, App Builder with preview, LAN and public device links, and business setups stay available on every plan, including Free. - Connecting and managing a WhatsApp Business number is free on every plan, including Free; AI replies on that number are included with every Premium plan, Lite included; Meta bills its messaging charges to the business directly. - When a plan ends, OpenAlly cloud AI, image generation and hosted publishing stop; the hosted address stops serving and no part of the app runs there, while the app, its files and the links served from your own device are untouched; WhatsApp Business stays connected and its AI replies pause while the owner keeps replying; the agent mailbox stays readable and erasable but stops sending. Questions: - Q: How much does OpenAlly Premium cost? A: Lite is US$3.99 a month or US$39.99 a year (₹249 a month or ₹2,499 a year in India). Plus is US$9.99 a month or US$99.99 a year (₹599 a month or ₹5,999 a year in India). Pro is US$19.99 a month or US$199.99 a year (₹1,199 a month or ₹11,999 a year in India). Ultra is US$39.99 a month or US$399.99 a year (₹2,499 a month or ₹24,999 a year in India). Max is US$149.99 a month (₹9,999 a month in India), and Max yearly, US$1,499.99 a year (₹99,999 a year in India), is sold in a WhatsApp chat with OpenAlly Miro. OpenAlly itself is free. - Q: Why might the app show a different price? A: The prices on this page are list prices, tax included: an international price in US dollars and the Indian price in rupees. Many countries pay a lower regional price than the US dollar list. The app shows yours before you confirm, from Google Play on Android or the checkout on a Mac, and that is the price you pay. - Q: Can I try Premium before I pay? A: Plans have no trial period. OpenAlly itself is free, so you can try everything except the cloud features first, with a model you bring or one that runs on your device, and add a plan when it would save you the setup. - Q: How do I pay for a plan? A: You choose a plan in the OpenAlly app; nothing is sold on this website. On Android, Google Play bills it. On a Mac, checkout runs through Razorpay. - Q: How do I get Max yearly? A: Max yearly is sold in a WhatsApp chat with OpenAlly Miro rather than an app store. [Open the chat](https://wa.me/918779222606?text=Hi%20Miro%2C%20I%20would%20like%20the%20Max%20yearly%20plan.) and press send on the message already written, or tap the contact button on the Max yearly card in the app. Once your WhatsApp number is linked to your OpenAlly account, Miro sends the checkout link. It is US$1,499.99, or ₹99,999 in India, with Max's daily limits and a full year of credits. - Q: What is a credit? A: Credits are how OpenAlly cloud AI keeps count. A reply draws credits while a model works on it; reading the reply, scrolling and deciding what to ask next draw nothing. A long conversation or a hard task draws more than a short answer. Images never touch your chat credits: every paid plan includes a daily number of everyday images that use no credits at all, and past that, pictures draw on their own image credits. - Q: What happens if I run out of credits? A: Your allowance refreshes each billing period, and a daily limit keeps one busy afternoon from emptying the month. A paid plan can add a credit pack in the app: 2,000, 5,500 or 12,000 credits, used after your plan's allowance, spent only while you have a paid plan, and valid for twelve months. You can switch to your own model at any time. - Q: Can I cancel at any time? A: Yes. A plan you buy in the app renews at the end of each month or year until you cancel. Cancelling stops the next payment. Your plan keeps working to the end of the period you paid for, then you continue on the free plan. Cancel in Google Play's subscriptions if you subscribed there, in the app's plan settings if you paid through Razorpay, or email [support@openally.ai](mailto:support@openally.ai?subject=Cancel%20my%20OpenAlly%20plan) if our team arranged your plan. - Q: Can I get a refund? A: All payments are final, including partial periods and unused credits. Your statutory rights under consumer law are unaffected. The [refund policy](https://openally.ai/refund-policy) sets out the details, including purchases made through an app store. - Q: What happens when my plan ends? A: OpenAlly cloud AI, image generation and publishing to a hosted address stop, and you keep working with your own model or one on your device. A hosted address stops serving and no part of your app runs there; everything you built, and the links served from your own device, are untouched. WhatsApp Business stays connected with your history and account controls; its AI replies pause and you keep replying yourself. Your agent's mailbox stays readable and can be erased, but stops sending. Credit-pack credits can be spent only during a paid plan and expire twelve months after purchase. Nothing you built on your device goes away. - Q: Is WhatsApp Business included? A: Connecting is free on every plan, including Free: connect your business number through Meta's signup, name the inbox, read and reply yourself, and manage the number. AI replies on that number come with every Premium plan: put an agent on it for sales, appointments and support, and take over any conversation. Meta bills its messaging charges to you directly. The [WhatsApp Business page](https://openally.ai/product/whatsapp-business) explains how it works. - Q: Is Web to App part of Premium? A: No. [Web to App](https://openally.ai/product/web-to-app) is free on every plan: it turns a website you run into a signed Android app. A plan adds two conveniences: a hosted address, which is one way to give an App Builder app the fixed https address packaging needs, and image credits for an AI-drawn icon. - Q: What does a plan add to App Builder? A: Uptime, and nothing else. Describing a site or a small app, watching it build on your own phone or Mac, changing it by asking, previewing it, sharing it on your Wi-Fi and publishing it to the internet from an Android phone or a Mac are free on every plan. That free public address is reserved and stays the same after every republish and restart, and your device serves it, so it sleeps when the device does. A plan hands the app to OpenAlly instead: we build, upload and serve it at a name you choose on openally.shop or mywebonline.store, so it stays online when your device is off — 1 app on Lite, 1 on Plus, 3 on Pro, 5 on Ultra, 10 on Max. One hosted app means one app, however many parts it publishes, and releasing an app frees its place. - Q: Can my website show what I sell? A: Yes, and it needs no plan. The [Business app](https://openally.ai/features/apps/business) makes a website from a template for a shop, a restaurant or a salon, starting from one of your inventories, and OpenAlly writes the first draft. You change it by asking and check a private preview before you publish. Prices and stock come from your catalogue, so a change in Business reaches the site with no rebuild, and orders placed there show under Website orders. On the free plan the site shows a small Made with OpenAlly badge and lists up to the number of products your plan allows. - Q: Where do Premium replies run? A: OpenAlly cloud AI replies are worked out on servers outside your device, and a reply receives the context it needs to answer, such as the record you asked about. The keys you bring are stored on your device. Your business records and the notes in the Notes app are kept in your OpenAlly account, so they reach every device you sign in on. We keep usage totals, not the text of your prompts or replies, and we do not promise which country a reply is handled in or an uptime figure. Prices (list prices, tax included; many countries pay a lower regional price, and the app shows yours before you confirm): - Lite: US$3.99 a month or US$39.99 a year; in India ₹249 a month or ₹2,499 a year. - Plus: US$9.99 a month or US$99.99 a year; in India ₹599 a month or ₹5,999 a year. - Pro: US$19.99 a month or US$199.99 a year; in India ₹1,199 a month or ₹11,999 a year. - Ultra: US$39.99 a month or US$399.99 a year; in India ₹2,499 a month or ₹24,999 a year. - Max: US$149.99 a month (₹9,999 in India). Max yearly is US$1,499.99 (₹99,999 in India), sold in a WhatsApp chat with OpenAlly Miro rather than an app store: message +91 87792 22606 on WhatsApp. Allowances (Lite / Plus / Pro / Ultra / Max): - OpenAlly cloud AI credits a month: 5,000 / 15,000 / 35,000 / 70,000 / 300,000. A yearly plan grants twelve months' credits for the year and keeps the same daily limit. - Daily credit limit: 334 / 1,000 / 2,334 / 4,667 / 20,000. - Automation credits a day, part of the daily limit rather than on top of it (scheduled runs and channel replies): 134 / 400 / 1,400 / 2,800 / 12,000. - Everyday images a day, which use no image credits (fair use; automations use image credits): 20 / 50 / 100 / 200 / 700. - Image credits a month: 2,000 / 5,000 / 12,000 / 25,000 / 100,000. - Apps online at a hosted address: 1 / 1 / 3 / 5 / 10. - Priority support: Ultra and Max. - An email address for your agent: Plus, Pro, Ultra and Max. Every paid plan includes OpenAlly cloud AI that judges each request and answers it with the lightest model that will do it justice, image generation, AI replies on WhatsApp Business (connecting a number is free on every plan; Meta bills its messaging charges directly), device actions from assistant connectors (the rest of the connector is free) and publishing to a hosted address (publishing from your own device is free on every plan). Plus and above also include an email address for your agent (50 MB of mail, 200 sends a day). ## Articles ### From one sentence to an Android app, built on a phone A home baker describes her shop to OpenAlly, publishes it at an address that stays up, then builds and signs its Android app on her phone. Canonical: https://openally.ai/blog/from-a-sentence-to-an-android-app Markdown: https://openally.ai/blog/from-a-sentence-to-an-android-app.md Published: 2026-09-11 Reviewed: 2026-09-11 Author: Matterward Labs (https://matterwardlabs.com) Category: Guide Keywords: build an Android app on your phone; AI website builder for a small business; home bakery website and app; website to Android app; APK and AAB for your own Google Play upload; switch on phone features in a website app; App Builder and Web to App; a shop website that stays up when your phone is off Audience: Home bakers and small food businesses who sell to regulars; Small-business owners who want their own Android app; Owners who run the shop from a phone between customers; People weighing a website against an app on Google Play Disclosure: Matterward Labs makes OpenAlly. Meera and Meera's Bakes are a worked example, not a customer. Screen labels, switches and limits come from the OpenAlly app and its Web to App and Premium pages, read on 11 September 2026. Google Play rules link to Google's own help pages. Meera sells sourdough on weekdays and celebration cakes to order from her home kitchen. Her order book is a notebook with flour on the cover. On one ordinary baking day she gets a website and then her own Android app, both made on her phone between trays. **App Builder** makes the site and **Web to App** packages it. Her Premium plan keeps the site online while the phone sleeps, and the Mac only comes out in the evening, for the Google Play paperwork. > [!IMPORTANT] How this story is grounded > Meera and Meera's Bakes are a worked example, not a customer, but every step, screen label, switch and limit in her day is OpenAlly's own. The painted scenes follow her day; OpenAlly's guide figure explains the rules. #### Short answer **In brief:** Describe your shop to OpenAlly's **App Builder**, shape the draft with guided edits and preview it on your phone. Once you approve, it goes online at an address that stays put: published free from your phone, or hosted in the cloud with a Premium plan so it stays up while your phone is off. **Web to App** then turns the site into a signed Android app on the same phone. A real build takes as little as 10 seconds, usually under a minute; the first build also creates your signing key. You install or share the APK yourself, choose which phone features the app may use, and after one key backup get an AAB for your own Google Play upload. Every app shows a small Made with OpenAlly line on its welcome screen and runs on Android 7.0 and newer; iPhone customers use the website. App Builder and Web to App are OpenAlly apps, both free on every plan, and Web to App builds on an Android phone or a Mac. > Interface view: **Meera's day, phone first and Mac last.** Press a step to see its screen. Everything that builds and signs happens on her phone; the Mac keeps the key backup and the Play Console paperwork. An example shop. Every phone screen uses the app's own labels. #### Describing the shop between two batches With the first loaves in the oven, Meera opens OpenAlly on her Android phone and finishes App Builder's own starter sentence: "Create a simple website for Meera's Bakes: sourdough on weekdays, celebration cakes to order, collection on Saturdays." > Figure: **Meera, a home baker, talks to her phone at a floury kitchen counter while a small glowing assistant lifts a paper storefront out of the screen.** Between two batches, she describes her shop in a sentence, App Builder drafts a styled first version and she previews it on her phone. Nothing goes public until she approves it. (light: https://openally.ai/images/blog/gouache/gouache-web-to-app-morning.webp; dark: https://openally.ai/images/blog/gouache/gouache-web-to-app-morning-dark.webp) That gets her a first draft, not a finished shop: [App Builder](https://openally.ai/features/apps/app-builder) makes a plain starter site with a weekly menu, a gallery of celebration cakes, prices and the Saturday collection details, already styled in her default look. A preview card in the chat updates as she asks for changes, and the rest of the morning is guided edits between trays, ending with her own photos of the loaves in the gallery. There is no checkout and no live stock list behind it. Meera takes orders the way she always has, by message: a customer picks a loaf or a cake size, and the page puts together an order summary to send her or keep for collection. #### An address that stays up Her AI asks before anything goes public. At lunchtime Meera publishes to the cloud with her [Premium plan](https://openally.ai/premium), at a name she chooses on openally.shop or mywebonline.store, and approves the exact version that goes live. It stays online while her phone is off, which gives Web to App the lasting address it needs. Published free from her phone instead, the site would work too, but the installed app would show its Retry screen whenever that phone was off or offline. Premium hosting serves static sites, and hers is one: pages, prices and photos, with nothing that has to answer from her phone. > Interface view: **Two ways to publish, one Web to App build.** The free route runs from the phone; Meera's Premium route hosts a static site in the cloud that stays up while the phone is off. Both lead to the same Web to App build, and addresses that change or carry a token are refused. The build gives an APK to share and, after one key backup, an AAB for your own Google Play upload. Temporary tunnel links, local-network addresses and links carrying an access token never qualify, because the app keeps opening the address it was built with for as long as it is installed. Whatever you build, give it an address you keep. > Figure: **OpenAlly's guide figure sits on a window sill while a phone lies face down on the bedside table and a soft cloud keeps a shop card glowing, mint light running to houses.** Her phone can sleep and the page stays reachable, because the address her app opens is hosted in the cloud. (light: https://openally.ai/images/blog/miro/miro-web-to-app-site-stays-up.webp; dark: https://openally.ai/images/blog/miro/miro-web-to-app-site-stays-up-dark.webp) #### Her name, colours and icon After lunch she opens the site's Android task, "Build an Android app", which leads to "Your app studio". It starts from the shop's own name, colours and icon, and the name under the icon can run to 30 characters, the length Google Play accepts. > Figure: **Meera holds her phone showing a croissant app icon and fans out colour swatches while a small glowing assistant holds up a matching tile cut from rolled dough.** Her app starts from her site's own name, colours and icon, and she can change each one, upload her own icon or have one drawn with AI. Every app shows a small Made with OpenAlly line on its welcome screen. (light: https://openally.ai/images/blog/gouache/gouache-web-to-app-name-and-icon.webp; dark: https://openally.ai/images/blog/gouache/gouache-web-to-app-name-and-icon-dark.webp) For the icon she can keep the site's, upload one or choose Create my icon with AI, using her connected image provider or her Premium plan. She asks for a croissant on soft sky blue, then writes the welcome line customers see while the site opens: "Sourdough on weekdays, cakes to order." A "Before you build" check then confirms the web address, the launcher icon and the signing key. Her hosted address passes because it is fixed. #### The phone powers she switched on Her app opens her live site, and a few switches let the page reach parts of the phone too. Each switch lets the page use that part of the phone and decides whether the built app asks for the matching permission at all. > Interface view: **Meera's switches, and what her customers get.** Choose a family of phone features, then flip the switches it has. Only the owner can switch on location, notifications, camera, microphone and clipboard reading; her AI can switch any of them off. An example app. | Family | What Meera's customers get | Who switches it on | | --- | --- | --- | | App and device | The order page spots a patchy or metered connection and loads lighter cake photos | Meera: it comes with Classic and new, which only she can choose | | Screen and navigation | Back closes the cake-size sheet, and the order summary stays on screen at collection | Screen control: her AI, once Classic and new is on | | Location and permissions | Nearest collection point, or a postcode box for a customer who says no | Location: only Meera. The microphone stays off | | Files, photos and sharing | The order summary saves to Downloads or goes to Meera through the share sheet; copying stays off, so the coupon code is shown to type in | Downloads and copying: her AI. Camera and clipboard reading: only Meera, both off | | Notifications and haptics | While the app is open, the page can post "Order summary saved" if the customer allows it, with a short buzz | Notifications: only Meera. Vibration: her AI | ##### Your AI writes it, you switch it on App Builder is taught to write pages that check a feature is on before using it, and to fall back when it is not: without location, Meera's page asks for a postcode. When a feature is missing, her AI names the switch that turns it on. The rule is set out on the Web to App page: "Only you can switch on camera, microphone, location, notifications and clipboard reading, or choose Classic and new for the newer page features. Your AI can switch any of them off, never on." New apps start on Classic only, the small set every app has had, such as the share sheet and opening links. Classic and new "adds file pickers, location, the clipboard and screen control", in the app's words. Anything left off is stripped out, so the install screen and the Play listing show only what she turned on. The preview inside OpenAlly leaves out location, the camera and screen control, so she tries those on the installed app. > Figure: **OpenAlly's guide figure presents a wooden panel of camera, microphone, map pin, bell and clipboard switches; only the map pin and bell glow, lighting up a phone beside it.** Meera switched on location and notifications, both switches only she can turn on; camera, microphone and clipboard reading stay off. Her AI can switch any of them off, never on. (light: https://openally.ai/images/blog/miro/miro-web-to-app-phone-powers.webp; dark: https://openally.ai/images/blog/miro/miro-web-to-app-phone-powers-dark.webp) ##### What stays out of her app - Web push and background work: nothing arrives while the app is closed. - Contacts, SMS, calendar, Bluetooth, NFC, biometrics and device identifiers. - A checkout. Customers send the order summary, and Meera can reply with a payment link, as [the Payments walkthrough](https://openally.ai/blog/payments-links-owner-only-refunds) describes. - An offline copy. Web developers can find every feature by its exact name in [the developer section of the Web to App page](https://openally.ai/product/web-to-app#sdk). #### Building and signing on her phone Mid-afternoon she taps Build APK. Her app's signing key is created on the phone with this first build and lives there and nowhere else, and this first build also downloads a small one-time app component. She can close the Building your app sheet while the build carries on; the APK appears under Generated files with Save, Share and Open. There is no Install button: she saves it, taps it in Downloads to install it, and shares it with her regulars, which is the one step of the day her phone leaves to her. OpenAlly never installs the app or uploads it to Google Play. | File | What it is for | When Meera gets it | | --- | --- | --- | | APK | Installing on a phone or sharing with anyone. It runs on Android 7.0 and newer | From her first build | | AAB | Her own Google Play upload. OpenAlly never uploads it | After one key backup, from Build APK and AAB | #### One key backup, then the AAB The next card is Back up your signing key. The key exists only on her phone, so losing the phone without a backup would mean the APK she shared could never be updated. In the evening she taps Back up key, sets a passphrase of at least eight characters and saves the file; Check backup confirms it restores the key. File and passphrase go somewhere that survives losing the phone, because "neither one is any use without the other". Both go to her Mac. The app gives the reason too: "Every Android app is signed, and the signature is what lets a phone accept a new version as an update to the one it already has." With the backup made, Build APK and AAB unlocks, and she gets an AAB, [the format Google Play requires for new apps](https://developer.android.com/guide/app-bundle). #### Play paperwork at the Mac Once the kitchen is clean she sits down at her Mac, where the key backup and the AAB now live. Her site's own words, the menu, the prices and the collection line, are what the listing is written from; the app, its conversation and its key stay with the phone that built them. > Figure: **Meera sits at a laptop at her kitchen table, holding open a padded envelope while a small glowing assistant lowers a brass key into it.** One key backup, to a passphrase-protected file, unlocks the AAB for her own Google Play upload. The file now lives somewhere that survives losing the phone, and the Play Console paperwork gets a bigger screen. (light: https://openally.ai/images/blog/gouache/gouache-web-to-app-key-backup.webp; dark: https://openally.ai/images/blog/gouache/gouache-web-to-app-key-backup-dark.webp) The app's guide, "Your path to Google Play", sets out four steps: "Start with your developer account", "Give your app a storefront", a testing step for the release she uploads, and "Bring your app to people". The upload is hers to make. Google charges a [US$25 one-time registration fee](https://support.google.com/googleplay/android-developer/answer/6112435), and a personal account created after 13 November 2023 must first run a [closed test with at least 12 testers opted in continuously for at least 14 days](https://support.google.com/googleplay/android-developer/answer/14151465). Fourteen days is longer than even her slowest sourdough takes to prove. Her Data safety answers must cover what her site collects and every phone feature she switched on. Google Play's [policy on apps with limited functionality](https://support.google.com/googleplay/android-developer/answer/9898783) does not allow "apps that only have limited functionality and content", so an app that only opens a website is judged on what that website does for the person using it. #### What her customers get Her regulars get an app that opens her live site, so this week's bakes and prices show next time they open it. Without a connection it shows an offline screen with a Retry button and reloads by itself when the signal returns. Because her site is hosted, whether Meera's phone is on never comes into it. The welcome screen shows her croissant, her greeting and a small Made with OpenAlly line; her name, colours and icon are everywhere else. The app runs on Android 7.0 and newer, and iPhone customers use the same site in their browser, where [OpenAlly itself](https://openally.ai/download) runs on both. An order still ends in a conversation, with the summary on screen at Saturday collection. #### Frequently asked questions ##### Do I need a website before I start? No. App Builder can make one from a description. If you already run a site at a fixed https address, Web to App can package that instead, reading the name, colours and icon the site publishes as suggestions you can change, and never the content of its pages. ##### Can I do the whole thing from a chat? Most of it. In a conversation, App Builder makes and previews the site, asking before it publishes, and the Web to App skill creates the app, starts a build and hands you the APK on a card with Save, Share and Open. The owner-only switches and the key backup live on the app's own setup screen, so the AAB waits for that backup. ##### What happens when my phone is off? It depends where the site lives. Published free from your phone, the site runs from that phone, so the installed app shows its Retry screen until the phone is back on and online. A static site hosted with [Premium](https://openally.ai/premium), like Meera's, stays up while your phone is off. ##### Do customers need a new version when I change prices or photos? No. The app opens your live site, so new prices and photos show the next time a customer opens it. A new name, icon, switch or address needs a new build, and installed copies keep the address they were built with. The package id is OpenAlly's to assign, it never changes, and the same key signs every build, so each new version installs as an update. ##### How soon can customers find the app on Google Play? After you upload the AAB to your own Play Console listing and Google's review passes; OpenAlly never uploads it. A personal developer account created after 13 November 2023 first needs a [closed test with at least 12 testers opted in continuously for at least 14 days](https://support.google.com/googleplay/android-developer/answer/14151465). Until then, share the APK, which installs on Android 7.0 and newer. #### How this was checked Every step, screen label and limit in Meera's day comes from the OpenAlly app's own setup screens and its [Web to App](https://openally.ai/product/web-to-app) and [Premium](https://openally.ai/premium) pages, read on 11 September 2026; where they disagree with this article, trust the pages. Google's rules come from the help pages linked where they appear, opened the same day. - Meera is a worked example; no customer's app or figures appear. - The build time is OpenAlly's stated range, not a stopwatch reading. - Her switch set is one sensible set for a home bakery. #### What to read next Try a sample build on the [Web to App page](https://openally.ai/product/web-to-app#try), see what [App Builder](https://openally.ai/features/apps/app-builder) can make, or [get OpenAlly](https://openally.ai/download) for your phone. The rest of a small shop's day, from the first enquiry to the invoice, is in [how an AI agent runs a small business](https://openally.ai/blog/ai-agent-runs-small-business). About the author: [Matterward Labs](https://matterwardlabs.com) makes OpenAlly. It builds AI software, the apps around it and the systems that run them. Corrections go to the [contact page](https://matterwardlabs.com/contact). #### Declared sources - [Android Developers: About Android App Bundles](https://developer.android.com/guide/app-bundle) - [Play Console Help: Functionality, Content, and User Experience](https://support.google.com/googleplay/android-developer/answer/9898783) - [Play Console Help: App testing requirements for new personal developer accounts](https://support.google.com/googleplay/android-developer/answer/14151465) - [Play Console Help: Register for a Google Play developer account](https://support.google.com/googleplay/android-developer/answer/6112435) - [Matterward Labs](https://matterwardlabs.com) - [Matterward Labs contact page](https://matterwardlabs.com/contact) ### Inventory and GST in one place OpenAlly Business keeps a product catalogue, stock levels and GST in one record in your account. An agent can restock it; inventory deletes stay owner-only. Canonical: https://openally.ai/blog/inventory-and-gst-in-one-place Markdown: https://openally.ai/blog/inventory-and-gst-in-one-place.md Published: 2026-09-04 Updated: 2026-10-01 Reviewed: 2026-09-04 Author: Matterward Labs (https://matterwardlabs.com) Category: Feature Keywords: inventory and GST in one place; product catalogue and stock levels; GST with CGST and SGST components; inclusive or exclusive tax; OpenAlly Business app; owner-only inventory deletes; quote tax on any amount Audience: Shop owners who keep products, stock, prices and GST in one record; Households that keep pantry or bulk supplies in the same product tools; Owners deciding which inventory jobs an agent may handle Disclosure: Matterward Labs makes OpenAlly. Product claims come from the OpenAlly catalogue, reviewed 2026-09-04. Quoted lines come from the sources listed below. Four notes disagree about the same shop. The pad by the till has yesterday's count. A phone note holds the two prices you changed on Saturday. A spreadsheet on a laptop still carries last quarter's tax rates, and the shelf has the only answer anyone trusts. OpenAlly **Business** is the record those four notes keep failing to be. Products, stock levels, prices and tax sit in it together, in your OpenAlly account on every device you sign in on, and your AI can work them without you opening the app. #### Short answer **In brief:** **Business** is an OpenAlly app with a matching skill. The app holds the catalogue, the stock levels and the tax setup. The skill is the leave you give one agent to drive that app. Inventories are single-currency. Products live in one shared pool and get listed per inventory with cost, MRP, sale price and stock. Tax is modelled as GST with parallel CGST and SGST components, inclusive or exclusive, and quoted on any amount you name. An agent can restock an item. Removing a whole inventory stays with you. Android comes from [Google Play](https://play.google.com/store/apps/details?id=openally.ai), and the catalogue lists Business for Android, iPhone and Mac. > Figure: **OpenAlly's guide figure holding one record card. The card shows a product row, a stock count and a tax line, with four loose paper notes stacked aside.** Four notes that disagree, replaced by one record. Products, stock and tax are held together instead of reconciled later. (light: https://openally.ai/images/blog/miro/miro-inventory-gst-short-answer.webp; dark: https://openally.ai/images/blog/miro/miro-inventory-gst-short-answer-dark.webp) #### The Business app and the Business skill **Business** is the OpenAlly app for what you sell. Its catalogue line is one sentence: manage inventory, products, prices, and taxes for your business. The app page breaks that into four parts. Inventory carries stock levels with low-stock flags, and the counts stay honest. Catalog and prices let you edit items and prices yourself, or tell the AI what to change. Tax summaries are the numbers you file, ready when you are. The fourth is the one owners feel: "Raise everything in Winter by 5%" is a sentence, not an afternoon (Business app page, read 4 September 2026). **A skill** is the leave you give one agent to use a particular tool. The Business skill describes its own reach in plain words. It runs a product catalog and stock: create inventories, add products to a shared pool, list them per inventory, organise with categories, set up taxes, and quote tax on any amount. Hand it to the agent that runs the shop. Keep it off the one that only books appointments. The skill is catalogued for Android, iPhone and Mac, and it is managed by the Business app. That pairing matters when you set an agent up. You configure the catalogue once in the app, and only agents holding the skill can reach it afterwards. > Figure: **OpenAlly's guide figure between a shelf panel of product rows and a floating grant card. The card carries a small wand and a padlock.** The app holds the catalogue, the stock and the tax setup. The skill is the leave one agent gets to drive it all. (light: https://openally.ai/images/blog/miro/miro-inventory-gst-app-and-skill.webp; dark: https://openally.ai/images/blog/miro/miro-inventory-gst-app-and-skill-dark.webp) > Figure: **Android Products screen labelled Your global catalog, with a search field and rows for physical and digital items** The global catalog on Android: one shared product pool you search and open. Draft status shows on the row when an item is not ready. (light: https://openally.ai/images/blog/ui/workspace-android-catalog.webp; dark: https://openally.ai/images/blog/ui/workspace-android-catalog-dark.webp) > Figure: **Mac Products view with the global catalog on the left and one product open on the right, including tax class and listed inventories** On Mac a product opens beside the catalog list. Category, tax class, and where it is listed sit on the same card. (light: https://openally.ai/images/blog/ui/workspace-mac-catalog.webp; dark: https://openally.ai/images/blog/ui/workspace-mac-catalog-dark.webp) #### One product pool, one currency per inventory **An inventory** is a container for what you hold in one place, and each runs on a single currency. Products live one level up, in a pool shared across every inventory you keep. You list a product into an inventory with its own cost, MRP, sale price and stock. **MRP** is the maximum retail price printed on the pack. That split earns its keep quickly. For example, a boutique with a shop and a weekend stall keeps one pool of products and two inventories over it. The same jar of pickle is one product, listed twice, at two sale prices, with two stock counts. Categories keep the listing readable once it runs past a screen. Editing the product changes the thing itself. Editing a listing changes what one inventory charges and holds. Owners who have merged two spreadsheets by hand will recognise which of those edits used to be painful. > Figure: **OpenAlly's guide figure lifting one product token out of a central pool. Two labelled trays beside it hold the same token with different price tags.** One product, two listings. The pool holds the item once, and each inventory prices and counts it on its own. (light: https://openally.ai/images/blog/miro/miro-inventory-gst-product-pool.webp; dark: https://openally.ai/images/blog/miro/miro-inventory-gst-product-pool-dark.webp) > Figure: **Android Inventories screen listing stock locations and price books, with one row marked Archived** Inventories on Android: each row is a stock location or price book. Archive sits on the row, not in a separate trash list. (light: https://openally.ai/images/blog/ui/workspace-android-inventory.webp; dark: https://openally.ai/images/blog/ui/workspace-android-inventory-dark.webp) > Figure: **Mac Inventories split view with stock locations on the left and listings for the selected store on the right** Mac Inventories keeps the location list beside the listings for the place you selected. It is the long-list surface for the same record. (light: https://openally.ai/images/blog/ui/workspace-mac-inventory.webp; dark: https://openally.ai/images/blog/ui/workspace-mac-inventory-dark.webp) #### GST, inclusive or exclusive Tax gets set up beside the products, rather than rebuilt from receipts at quarter end. Business models GST with parallel CGST and SGST components, and each tax is either inclusive or exclusive. **Inclusive tax** means the displayed price already contains the tax. **Exclusive tax** means the tax is added to the base price. One sale can be a shelf label of 118 rupees inclusive or a base price of 100 rupees exclusive, and what changes is the number you print rather than the money you take. That is arithmetic on an example rate, not a statement about what your own goods carry. The catalogue ships a starter for the setup: "Set up an 18% GST tax (CGST 9% + SGST 9%) for my catalog." India's [Central Board of Indirect Taxes and Customs](https://cbic-gst.gov.in/about-gst.html) explains why two components run side by side. It defines SGST as "State Goods and Services Tax, one of the three categories under Goods and Services Tax (CGST, IGST and SGST)". The same page says the "Centre will levy and administer CGST & IGST while respective states will levy and administer SGST". That starter is an example command in the app. It is not tax advice, and it does not claim that everything on your shelf carries one rate. You set the rates your goods actually attract, and the app keeps the components apart the way the tax is levied. > Figure: **OpenAlly's guide figure at a board where one tax card splits into two equal cards. One card is marked central, the other is marked state.** GST is modelled as parallel components instead of one flat number. The split follows the way the tax is actually levied and administered. (light: https://openally.ai/images/blog/miro/miro-inventory-gst-gst-split.webp; dark: https://openally.ai/images/blog/miro/miro-inventory-gst-gst-split-dark.webp) > Figure: **Android Edit tax screen for GST with Exclusive selected and CGST plus SGST component fields** GST setup on Android: kind, inclusive or exclusive, rounding, and parallel CGST and SGST components on one edit sheet. (light: https://openally.ai/images/blog/ui/workspace-android-gst.webp; dark: https://openally.ai/images/blog/ui/workspace-android-gst-dark.webp) > Figure: **Mac Edit tax dialog for GST with Exclusive selected, Banker's rounding, and CGST plus SGST components** The same tax edit on Mac: GST kind, exclusive or inclusive, rounding, and equal CGST and SGST components before you save. (light: https://openally.ai/images/blog/ui/workspace-mac-gst.webp; dark: https://openally.ai/images/blog/ui/workspace-mac-gst-dark.webp) #### Tax quoted on any amount, and summaries you can file Once the rates are in, tax stops being a separate calculation at the counter. You can ask for tax on any amount you name, and the answer comes back split the way your setup describes. A customer at the counter asking what something costs with tax is answered from the catalogue. **Tax summaries** are the fourth card on the app page, and the line there is deliberately short: the numbers you file, ready when you are. Reading a summary changes nothing in the record, so it is ordinary work for an agent to fetch on request. Two more reads sit in the same family. Asking what is low reads stock levels and their low-stock flags. Asking what your stock is worth totals the value you are holding now, rather than the value you held at the last stocktake. > Figure: **OpenAlly's guide figure holding a small calculator card beside a shelf label. An arrow runs from a bare amount to a figure with the tax shown underneath.** Quoting turns an amount you name into the split your setup describes. Summaries gather the same figures for filing. (light: https://openally.ai/images/blog/miro/miro-inventory-gst-tax-quote.webp; dark: https://openally.ai/images/blog/miro/miro-inventory-gst-tax-quote-dark.webp) #### An agent can restock, and inventory deletes stay with the owner The Business skill lists the everyday asks in its own words. Add a product, restock an item, find what's low, or value your stock on hand. Those are reads and small writes. An agent does them when you ask, in the conversation you were already having. Container deletes belong to a different class of action, and the skill page states the boundary in one line. > [!IMPORTANT] The inventory stays under owner control > Container deletes are owner-only. An agent can restock; it drops an inventory only when the owner asks, and refuses anyone else. | Action | What it does | Who does it | | --- | --- | --- | | Create an inventory | Opens a container that runs on one currency | Your agent, when you ask | | Add a product | Puts an item into the shared pool | Your agent, when you ask | | List a product for sale | Sets cost, MRP, sale price and stock in one inventory | Your agent, when you ask | | Restock an item | Sets the stock level on a listing | Your agent, when you ask | | Set up a GST tax | Adds parallel CGST and SGST components | Your agent, when you ask | | Quote tax on an amount | Returns the split for a figure you name | Your agent, when you ask | | Find what is low | Reads stock levels and low-stock flags | Your agent, when you ask | | Value stock on hand | Totals what you are holding now | Your agent, when you ask | | Delete a whole inventory | Removes the container and its listings | Owner only: your agent, when you ask | Money has its own guard in a different app. Capture and refund are owner-only, and Razorpay is the first provider rather than the only one there will ever be. The article on [payment links and owner-only refunds](https://openally.ai/blog/payments-links-owner-only-refunds) covers that half of the counter. > Figure: **OpenAlly's guide figure sliding boxes onto a shelf. A gold padlock rests on the tall cabinet that holds the whole inventory.** Restocking is ordinary agent work. Removing an entire inventory sits behind a lock that only the owner opens. (light: https://openally.ai/images/blog/miro/miro-inventory-gst-restock-lock.webp; dark: https://openally.ai/images/blog/miro/miro-inventory-gst-restock-lock-dark.webp) #### Shop shelf and household cupboard, same product tools The product tools do not ask what kind of building you are standing in. A household can hold nappies or washing powder in Business, count them, and ask what is low, and that is the same shipped catalogue rather than a second system with a friendlier name. [The workspace guide](https://openally.ai/blog/one-workspace-for-shop-and-home) covers the wider arrangement, where household tools sit beside shop ones in a single account. #### The phone in the aisle and the Mac for the long list Stock gets counted where the stock is, which is rarely at a desk. Android comes from [Google Play](https://play.google.com/store/apps/details?id=openally.ai). [The Mac page](https://openally.ai/download/mac) carries the Mac build, and [the download page](https://openally.ai/download) covers the rest. The catalogue lists Business for Android, iPhone and Mac, and that listing is a platform badge rather than an App Store launch. A phone suits a shelf walk. A Mac earns its place when you read a whole inventory at once, check a long list of listings against what the shelves say, or set rates across a catalogue in one sitting. The account, the agents and the conversation carry over between them. > Figure: **OpenAlly's guide figure walking a teal path from a phone held up in a shop aisle to a desk where a long list fills a monitor.** A shelf walk fits in one hand. Reviewing a whole listing is easier on a bigger screen, in the same account. (light: https://openally.ai/images/blog/miro/miro-inventory-gst-android-mac.webp; dark: https://openally.ai/images/blog/miro/miro-inventory-gst-android-mac-dark.webp) #### What the owner still does - Choose the currency when you create an inventory. Once it holds listings, changing it is an owner-only step. - Decide which agents get the Business skill, such as the one that runs the shop. - Set the tax rates your goods actually attract. - Delete a whole inventory, if it ever comes to that. - Walk the shelves now and then, because a count is only as good as the last person who looked. #### How this was checked The methodology is a catalogue read. Every product line above comes from the OpenAlly catalogue entries for the Business app and the Business skill, read on 4 September 2026. Where this page and the catalogue disagree, the catalogue is right. Those product claims are published as structured data on the catalogue pages themselves, because the blog compiler does not accept it inside an article's Markdown. Both quotations from the Central Board of Indirect Taxes and Customs were checked against the linked page on the same day. - The action table was built for this page by lining the Business skill's own starters up against its container-delete boundary. No single catalogue card shows that grid. - Treating the shared product pool and the per-inventory listing as two separate records is a reading of the catalogue description rather than a quote from it. About the author: [Matterward Labs](https://matterwardlabs.com) makes OpenAlly. It builds AI, hardware and software for the edge, and the platform around both. This article was reviewed by Matterward Labs on 4 September 2026. For press, a correction, or other contact, use the inbox on that site. #### What to read next The [Business app](https://openally.ai/features/apps/business) page lists what the app does and which skills drive it. The [Business skill](https://openally.ai/features/skills/business) page shows the access it asks for before you grant it, together with its starters. The [full catalogue](https://openally.ai/features) lists every app and skill with the platforms each one supports. #### Declared sources - [CBIC, Government of India: Know About GST](https://cbic-gst.gov.in/about-gst.html) - [OpenAlly on Google Play](https://play.google.com/store/apps/details?id=openally.ai) - [Matterward Labs](https://matterwardlabs.com) ### What OpenAlly Companion Face does Companion Face is an optional OpenAlly face for gestures, moods and lip-sync on Android and Mac, with focus, wellness, now playing and system moments. Canonical: https://openally.ai/blog/what-is-openally-companion-face Markdown: https://openally.ai/blog/what-is-openally-companion-face.md Published: 2026-09-03 Updated: 2026-09-11 Reviewed: 2026-09-03 Author: Matterward Labs (https://matterwardlabs.com) Category: Feature Keywords: what is OpenAlly Companion Face; OpenAlly companion face; optional AI face Android Mac; lip-sync companion face; focus wellness now playing face; OpenAlly Slap companion hook Audience: You want an optional face for an agent you already use; You are deciding whether Companion Face belongs on Android and Mac; You use Slap and want to know whether the face is required Disclosure: Matterward Labs makes OpenAlly. Product claims come from OpenAlly's own catalogue and the public Companion Face feature page, read on 3 September 2026. The catalogue gives Companion Face one line: "A companion face that reacts to your assistant and helpful system moments." The job under that line is optional and easy to picture. You grant an approved agent a face. The face waves, changes mood, lip-syncs a short line, and idles when the day is quiet. #### Short answer > Figure: **OpenAlly's guide figure stands between two devices. A small living face sits in a Mac notch and at a phone corner, with no camera lens** The opening pose. Optional face on the two catalogue surfaces, Android and Mac. (light: https://openally.ai/images/blog/miro/miro-companion-face-short-answer.webp; dark: https://openally.ai/images/blog/miro/miro-companion-face-short-answer-dark.webp) **In brief:** **Companion Face** is an OpenAlly app that gives an approved agent an expressive face. The face does gestures, moods, lip-sync and idle cycles. The public catalogue lists it for Android and Mac. There is no iPhone listing for this app. You can leave the face off. [Slap](https://openally.ai/blog/what-is-openally-slap) still works without it. [Matterward Labs](https://matterwardlabs.com) publishes Companion Face as part of OpenAlly. The [Companion Face feature page](https://openally.ai/features/apps/companion-face) is the public receipt for every claim below. #### What the face actually does > Figure: **OpenAlly's guide figure stands by four mint tiles. A timer, a glass, a music note, and a battery feed a small reacting face** Useful daily moments. Focus, wellness, now playing, and system pulse. (light: https://openally.ai/images/blog/miro/miro-companion-face-powers.webp; dark: https://openally.ai/images/blog/miro/miro-companion-face-powers-dark.webp) **Powers** are the jobs you can give the face. They are not generic app utilities. The feature page groups them as focus, wellness, now playing, system pulse, an optional Slap hook, and idle cycles with Studio. | What you switch on | What the face does | | --- | --- | | Focus timer | Shows a countdown at a glance and a small celebration when the session ends | | Wellness nudges | Reminds you to stretch, drink water, or look away, with quiet hours you set | | Now playing | Reacts when a track starts or changes, on Mac and Android | | System pulse | Turns volume, charging, headphones, unlock, appearance, heat, network, and low power into short face reactions | | Slap hook | Adds desk-tap reactions if you connect Slap. You still control Slap from the Slap app | | Idle cycles and Studio | Plays playlists and expression sets when the day is quiet | For example, a focus session is a work block, a short break, or a long break. The face is a glanceable clock, not a second calendar. Wellness nudges can remind you to look away from a screen. The [National Eye Institute](https://www.nei.nih.gov/learn-about-eye-health/healthy-vision/keep-your-eyes-healthy) puts that rest in plain terms: take a break every 20 minutes and look at something about 20 feet away for 20 seconds (National Eye Institute, 2025). Companion Face does not claim that rule as a medical feature. It is a reminder you schedule. Quiet hours stay respected. Now playing follows a live media session. System pulse is the rest of the day: mute, a full battery, headphones that put their own pair on the face, a lid lift on a supported MacBook, heat, network, unlock, and a change of appearance. > [!IMPORTANT] The screens on the feature page are a demonstration > Panel titles and sample sessions on the public Companion Face page show the shape of the flow. They are samples, not measurements. No figure on that page reports usage, speed, or how many faces anyone has switched on. #### Sensors, not a camera > Figure: **OpenAlly's guide figure covers a camera lens. The other hand points at volume, battery, and headphone marks feeding an eyeless face** Sensors and system moments. Not a camera. (light: https://openally.ai/images/blog/miro/miro-companion-face-sensors.webp; dark: https://openally.ai/images/blog/miro/miro-companion-face-sensors-dark.webp) **A sensor moment** is an operating-system event the face can react to. Volume and mute. Charging, unplugging, low power, and a full battery. Headphones that connect, with the device name. Lid angle and lift on a supported MacBook. Heat, network, unlock, and appearance. On Android, motion, pocket, light, and selected accessibility moments. The companion does not keep raw sensor readings or typed text. It is deliberately without a camera. The feature page says that in one line: no camera. Voice features are opt-in. On Mac, the face lives by the notch. On Android, you use it in OpenAlly, or you choose the optional camera-edge view. That camera-edge view is a placement, not the whole product. It is not an overlay story for every platform, and it is not Aster. Aster is a separate, sideloaded Android companion for screen control. Companion Face is a face. #### Voice with lip-sync > Figure: **OpenAlly's guide figure holds a phone. The small face talks, with a local-voice mark and a separate key mark beside it** A short spoken line. The mouth follows on-device voice or a key you bring. (light: https://openally.ai/images/blog/miro/miro-companion-face-voice.webp; dark: https://openally.ai/images/blog/miro/miro-companion-face-voice-dark.webp) **On-device voice** is a voice already on your device. The face lip-syncs. No speech request is sent away. You can also bring your own [ElevenLabs](https://elevenlabs.io) key. The feature page lists that path for lifelike voices, Scribe, sound effects, music, and voice design. Generated audio is saved for playback. The companion can still lip-sync what you play. Short conversational lines are the fit. Longer text belongs in text. For instance, "Wave at me from your companion face and say hello out loud" is a starter on the [Companion Face skill](https://openally.ai/features/skills/companion-face). The skill is how an approved agent controls expressions, speech, and idle moments. Grant it only to the agents that should have a face. Claude Code, Codex, and your own automations can also drive reactions. Progress, a permission request, attention, and done can show on the face. You choose whether that also notifies, speaks, or summarises. Those hooks are optional. They are not the definition of the app. #### Slap is an optional hook > Figure: **OpenAlly's guide figure taps a desk. A phone already plays a soundbite, and a tiny optional face recoils on a dotted unpluggable line** Slap already plays the clip. The face is a hook you can leave unplugged. (light: https://openally.ai/images/blog/miro/miro-companion-face-slap-hook.webp; dark: https://openally.ai/images/blog/miro/miro-companion-face-slap-hook-dark.webp) **Slap** is the app that turns a desk tap, a device shake, or the charger going in into a meme soundbite. It has its own [feature page](https://openally.ai/features/apps/slap) and its own [explainer](https://openally.ai/blog/what-is-openally-slap). Slap is listed for Android, iPhone and Mac. Companion Face is listed for Android and Mac. Connect Slap and the face can recoil while the meme clip owns the mouth. Leave the hook off and Slap still barks. The face is an optional power hook. You control it from the Slap app, not from a buried switch on the face. > [!IMPORTANT] The face is optional > OpenAlly and Slap keep working if you never switch Companion Face on. The hook is a connection you choose. #### Studio and agent authoring > Figure: **OpenAlly's guide figure sits at a studio desk. Expression cards are arranged, and one agent-suggestion card waits to be kept or dropped** Studio authoring. Inspect, reuse, or delete idle cycles and expressions. (light: https://openally.ai/images/blog/miro/miro-companion-face-studio.webp; dark: https://openally.ai/images/blog/miro/miro-companion-face-studio-dark.webp) **Studio** is where idle cycles and expression sets live. Playlists run when the day is quiet. Agents can suggest new expressions. Those suggestions are labelled so you can inspect them, reuse them, or delete them. The skill starters on the catalogue card are the same idea in chat form. Wave hello. Invent a playful expression and add it. Create a calm morning idle cycle and switch to it. None of that is a hidden animation pack. You can see what was added. #### Companion Face on Android and Mac > Figure: **OpenAlly's guide figure stands between two machines. An Android phone at chest height and an open Mac each carry a small companion face** Two catalogue platforms. Android and Mac, not a third device. (light: https://openally.ai/images/blog/miro/miro-companion-face-android-mac.webp; dark: https://openally.ai/images/blog/miro/miro-companion-face-android-mac-dark.webp) The public catalogue lists Companion Face for Android and Mac. OpenAlly itself remains a cross-platform product. This app is one of the focused apps whose platform list is narrower than the whole suite. Android is the OpenAlly app from [the download page](https://openally.ai/download). Mac is a direct download from [the Mac download page](https://openally.ai/download/mac) rather than a Mac App Store listing. The Mac side is not a separate account. Your agents and your chat history continue there. > Figure: **Android Companion screen labelled IN APP, with a large face card and a Listening status pill** Companion Face inside the OpenAlly Android app: the in-app face card with Listening on. This is the in-app surface, not a camera-edge overlay. (light: https://openally.ai/images/blog/ui/companion-face-android-inapp.webp; dark: https://openally.ai/images/blog/ui/companion-face-android-inapp-dark.webp) > Figure: **Mac Companion screen with a face in the notch and Idle Listening Thinking Speaking tiles, Thinking selected** Companion Face by the Mac notch: Idle, Listening, Thinking, and Speaking on the Companion face card, with Thinking selected. The face sits in the notch. (light: https://openally.ai/images/blog/ui/companion-face-mac-notch.webp; dark: https://openally.ai/images/blog/ui/companion-face-mac-notch-dark.webp) On Mac, the face lives by the notch. On Android, use it in OpenAlly or choose the optional camera-edge view. Approved agents can control expressions and speech. That is the placement copy on the live feature page. It is not a claim that Companion Face sits over every other app on every device. #### Where Companion Face sits in the catalogue The [app catalogue](https://openally.ai/features/apps) lists Companion Face beside the other focused apps. The [full catalogue](https://openally.ai/features) puts it next to the skills, tools and agents that share a workspace. The matching skill uses the same name and the same two platforms. If you already read the Slap explainer, the line you will recognise is the optional one. The face can recoil. Slap does not need it. #### How this was checked > Figure: **OpenAlly's guide figure holds a printed page. The other hand checks it against a short list** A catalogue check. Claims are read against the public Companion Face page. (light: https://openally.ai/images/blog/miro/miro-companion-face-sources.webp; dark: https://openally.ai/images/blog/miro/miro-companion-face-sources-dark.webp) The methodology is a catalogue read. Every product line above comes from OpenAlly's own catalogue and the public [Companion Face feature page](https://openally.ai/features/apps/companion-face), read on 3 September 2026 (Matterward Labs, 2026). Where this page and the catalogue disagree, trust the catalogue. The look-away rest is the National Eye Institute's wording, not a Companion Face clinical claim. ElevenLabs is named because the feature page names that bring-your-own path. Powers, sensors, hooks and voice paths on that feature page are structured data. They are the receipt this article is checked against. - Every product claim was opened against the live catalogue and the public Companion Face page on 3 September 2026. Each of those pages is linked at the point of use. - The public catalogue lists Companion Face on exactly two platforms, Android and Mac. That count is a first-party catalogue read for this article, not a market survey. - Reading Slap as an optional hook, rather than as a requirement of the face, is an editorial framing of the feature-page line "optional power hook". Slap's own page already says the face is optional. About the author: [Matterward Labs](https://matterwardlabs.com) makes OpenAlly. It builds AI software, the apps around it, and the systems that run them. This article was reviewed by Matterward Labs on 3 September 2026. Corrections go to the company [contact page](https://matterwardlabs.com/contact). #### Declared sources - [National Eye Institute, Keep Your Eyes Healthy](https://www.nei.nih.gov/learn-about-eye-health/healthy-vision/keep-your-eyes-healthy) - [ElevenLabs](https://elevenlabs.io) - [Matterward Labs](https://matterwardlabs.com) - [Matterward Labs contact page](https://matterwardlabs.com/contact) ### OpenAlly is not another chat app OpenAlly installs on your phone and your Mac, does the work you granted it, and thinks with a model you already pay for. Canonical: https://openally.ai/blog/openally-is-not-another-chat-app Markdown: https://openally.ai/blog/openally-is-not-another-chat-app.md Published: 2026-09-02 Updated: 2026-10-09 Reviewed: 2026-09-02 Author: Matterward Labs (https://matterwardlabs.com) Category: Feature Keywords: assistant that does things on your phone; OpenAlly is not a chat app; bring your own AI model; on-device AI assistant; AI agent for a small business; assistant that asks before paying Audience: Owners running a shop or a service business from a phone; People who want errands finished rather than conversations extended; Anyone choosing between a chat app and an assistant that can act Disclosure: Matterward Labs makes OpenAlly. Product claims come from the OpenAlly catalogue and its public product pages, read on 2 September 2026. OpenAlly is software you install on a phone and on a Mac. It does the work you granted it, thinking with the model you chose. Talking to it is the least interesting thing it does. A chat app clocks off the moment the answer lands on screen. The restock is still unordered. The payment link is still unsent. The customer on WhatsApp is still waiting, and the Mac on the far side of the room is still asleep. Tuesday does not grade prose. The category has a name and a literature behind it. MIT Sloan's explainer on [agentic AI](https://mitsloan.mit.edu/ideas-made-to-matter/agentic-ai-explained) separates "the now familiar chatbots that field questions and solve problems" from software that "integrates with other software systems to complete tasks independently or with minimal human supervision". Matterward Labs builds OpenAlly around the second half of that sentence. Two boxes of stock to reorder. A link to send while the buyer is still standing at the counter. An order to log. A file to open on a machine nobody wants to walk over to. #### Installed, not visited **In brief:** **OpenAlly** is an assistant you install rather than a site you visit, and you hand it work a skill at a time. Android comes from [Google Play](https://play.google.com/store/apps/details?id=openally.ai). See [the Mac download page](https://openally.ai/download/mac) for the current Mac release. Your chats and agents are written into a database file on the device in your hand, and your products, orders and invoices are kept in your OpenAlly account. Capturing or refunding a payment stays with you, an email reply waits as a draft until you press send, and on an Android screen driven through the optional Aster companion, a send, a payment or a delete waits for your tap. > Figure: **OpenAlly's guide figure turning away from a talking speech bubble. A phone in the other hand shows a task ticking over.** An answer arriving on a screen. Set against it, an errand that actually finished. (light: https://openally.ai/images/blog/miro/miro-not-chat-phone-work.webp; dark: https://openally.ai/images/blog/miro/miro-not-chat-phone-work-dark.webp) #### Your device is the computer Open the app and the work sits where you are standing. Chats and agents go into a database file on the phone or on the Mac. Products, orders, bookings and invoices are kept in your OpenAlly account, so the phone, the Mac and anyone you invite work from the same records. Writing in IEEE Spectrum, engineer Dwith Chenna defines [edge AI](https://spectrum.ieee.org/edge-ai) as "deploying AI algorithms directly on devices at the edge of the network, rather than relying on centralized cloud data centers". That is where OpenAlly keeps your chats and agents by default. Where the thinking happens is a separate decision, and it stays yours. A provider you already pay for. A key you paste. A small model on the handset itself. Paste a provider key and the device talks to that provider directly. Nothing of ours sits in the middle to fall over or to read what you sent. Your account and [Premium](https://openally.ai/premium) do live in the cloud, and only if you use them. So does the delivery service behind WhatsApp Business: it keeps a business's Meta authorisation encrypted, and holds each message, encrypted, until the business's phone or Mac collects it. Whichever model you pick still reads what you send it, wherever that model happens to run. That is a narrower promise than nothing ever leaving the phone, and it is the true one. **An agent** is an assistant you set up for a particular kind of work. The one that talks to buyers is separate from the one that files notes, and separate again from the one that watches the calendar. Each gets its own leave to act. None of them inherits another's access. Keeping a single everyday assistant is perfectly reasonable, and most people start there. Specialists earn their keep later, when support, research, sales and operations begin needing different reach. At that point the separation stops being tidiness and becomes the whole safety story. Work you hand to a specialist comes back into the conversation you were already having, so delegating does not mean chasing an answer across three screens. > Figure: **OpenAlly's guide figure holding a phone open like a small workshop. Record cards and a brain token sit stacked inside the handset.** Working records live on the device. The brain is a part you slot in, and a part you can swap out later. (light: https://openally.ai/images/blog/miro/miro-not-chat-records-stay.webp; dark: https://openally.ai/images/blog/miro/miro-not-chat-records-stay-dark.webp) #### Nothing to rent, nothing to keep awake Setup instructions for phone assistants tend to end with a second machine. Rent a small server. Leave a box humming in a cupboard at home. Keep a laptop awake so the handset has something to phone. OpenClaw's official Android app, for one, requires another computer running macOS, Linux or Windows, as [our OpenClaw on Android notes](https://openally.ai/blog/ai-agent-on-your-phone-without-a-pc) set out from its documentation. OpenAlly's setup ends with the phone. There is no OpenAlly server for you to host, no container to babysit and no home lab to build first. The app is the whole installation. A laptop is welcome if you want one in the picture. Point the app at an OpenAI-compatible server on your own machine. That is how a local runner such as Ollama joins the story. The [Local LLM tool](https://openally.ai/features/tools/local-llm) runs a small model fully on the device, offline, on Android, iPhone and Mac. Both are switches you may flip. Neither is a condition you have to satisfy before the app earns its space. Setup pages for phone assistants keep asking for a spare box, and community projects exist to squeeze that box onto the handset itself. OpenAlly is built around that complaint. > Figure: **OpenAlly's guide figure walking past an unplugged rack and an idle laptop. Only the phone in hand is doing any work.** No rented box, and nothing humming at home. A laptop is an option in this picture, never a prerequisite. (light: https://openally.ai/images/blog/miro/miro-not-chat-no-host.webp; dark: https://openally.ai/images/blog/miro/miro-not-chat-no-host-dark.webp) #### The model is a setting > Figure: **Guide figure slots a brain token into a phone. Spare tokens rest on the table.** The model is a setting you can swap. The rest of the workspace stays put. (light: https://openally.ai/images/blog/miro/miro-not-chat-bring-brain.webp; dark: https://openally.ai/images/blog/miro/miro-not-chat-bring-brain-dark.webp) The brain is yours to choose and yours to change. There are three ways in. Sign in with a subscription you already pay for, and pay nothing further for it: ChatGPT, GitHub Copilot, Antigravity. Beside them sit plan rows that take a plan key instead of a sign-in. Those include the Z.AI GLM Coding Plan, Kimi For Coding, the MiniMax Token Plan, Synthetic and OpenCode Go. Moonshot AI, the Xiaomi MiMo Token Plan and the Alibaba Token Plan sit on that same shelf. OpenAlly's own route is there too, and it is the one route that runs through us, if you pick it. Paste a raw API key and go straight to the provider instead. Anthropic, OpenAI and Google AI are the front three. Azure OpenAI, Vertex AI and Amazon Bedrock cover a work account. Groq, Cerebras, Mistral and xAI cover speed or price. OpenRouter, NVIDIA and Vercel AI cover shopping through a single bill. Treat that as a sample rather than an inventory. The live list, with the platforms each entry supports, sits in [the catalogue](https://openally.ai/features). Or run the thing yourself. An OpenAI-compatible server on your own hardware is a first-class option, and so is a small model on the handset. Switch next month because something cheaper or sharper appeared, and your agents, your granted skills and your records stay exactly where they were. The model is the part that moves. ChatGPT, Claude, Gemini and Copilot are brains, and mostly they are brains you visit: a tab, a window, a place you go to think out loud. Grok and Perplexity are a different shape again, and good ones. You go, you ask, you read. OpenAlly sits on the other side of that trade. It is the surface on your own device that goes and does, and it is glad to borrow whichever brain you rate this quarter. #### Skills are the leave you grant **A skill** is the leave you give one agent to use a particular tool, written in ordinary language rather than as a screen full of switches. The agent that handles buyers gets Sales and Payments. The one that only files notes gets neither, and cannot be talked into either halfway through a conversation. Every grant is separate. One agent's access tells you nothing about the next agent's, which is dull to explain and restful to live with. That split is what lets one account hold a shop and a household without mixing their reach. The specialist that raises a payment link never inherits the Health skill. The everyday assistant that files a shopping list never inherits refunds. Consequences stop the machine. On an Android screen driven through Aster, paying, sending, deleting and changing a setting all wait for a tap. Email works the same way everywhere: the agent reads the conversation, drafts the reply in your voice, tidies it, and puts it in front of you. You press send. | What you ask for | Who finishes it | | --- | --- | | A reply to a customer email | The agent drafts it, you press send | | A payment link and the order behind it | The agent, as soon as you ask | | Capturing or refunding a payment | Owner only: your AI, with your sign-off | | Deleting an inventory, a product or a tax | The agent, only on the owner's own request | | A send, payment or delete on an Android screen | Nobody, until you tap | If none of the built-in skills fits the way you work, Skill Creator lets you teach the app a new one. The rest of the grant list is work, not a brochure. Web Search, Weather and Summarise cover the everyday lookup. Invoicing sits next to Payments when the customer needs a document, not only a link. Image Generation and Video Generation run once you have actually granted them. GitHub is there if you ship code. Agent Admin, Skill Creator and Session History live in the app for people who want to inspect or extend the setup. They are not why you install it. ##### The OpenAlly apps an agent can drive The apps are the other half of the arrangement. Each opens as its own screen, the way any app on your phone does, and each can also be driven by an agent you granted the matching skill. That duality is the trick, and it is why a phone full of OpenAlly apps does not turn into a phone full of chores. **Payments** connects a payment provider, raises links and orders, and reconciles the money when it lands. **Business** holds the catalogue, the stock levels and the tax. **Email** labels, drafts and replies in your voice. **Remote Deck** turns the phone into a controller for a Mac you approved. **OpenLink** does something similar for a desktop coding session over your own private link. **Image Providers** connects an image model, so an agent can actually produce the picture, on your plan or your key. Not all of it is commerce, and it would be a duller product if it were. **App Builder** walks you from a description to a real app you can preview and publish, in a guided flow on the device. [**Web to App**](https://openally.ai/product/web-to-app) turns a site you run into a signed Android app, built on your Android phone or Mac, which you install, share or upload to Google Play yourself. [A home baker's day with both](https://openally.ai/blog/from-a-sentence-to-an-android-app) runs from a sentence typed between batches to an app on her regulars' phones. **Vibe Pages** turns a mood into a shareable page with its own background, fonts and playlist. **Companion Face** gives the assistant a face beside your camera, on Android and on Mac. **Slap** exists so you can slap your machine and have it bark a meme soundbite back at you. **Health** keeps records and medicines on the device. You can ask what is due tonight, or look up a measurement from the spring. It organises facts and stops there. It is not a tracker and it is not a medical device. It does not diagnose or treat. Anything clinical belongs with a clinician. Four of them ask for a part of Android that other platforms do not hand out, so App Automations, the SMS Analyser, Flare and Voice Notes are Android-only. The [full catalogue](https://openally.ai/features) lists every app, skill and tool with the platforms each one supports. > Figure: **OpenAlly's guide figure holding a short lead. A friendly helper stands beside three labelled tool cards.** Skills are granted per agent, in plain words. One agent's grant tells you nothing about the next one's. (light: https://openally.ai/images/blog/miro/miro-not-chat-skills-apps.webp; dark: https://openally.ai/images/blog/miro/miro-not-chat-skills-apps-dark.webp) #### What OpenAlly will tap, and what it refuses to **Aster** is an optional Android companion you install yourself, from the project's [GitHub Releases page](https://github.com/satyajiit/aster-mcp/releases/latest). It sits outside the Play listing, and OpenAlly works perfectly well without it. With it, the [App Automations skill](https://openally.ai/features/skills/app-automations) lets an agent drive the apps already on the handset: open the delivery app, fill the form, walk to the last screen, and stand aside. The interesting part is where it stops. You can halt a run in the middle of a step. Logins and payments hand straight back to you, and so do CAPTCHAs and two-factor prompts when an agent is working in a web page, because software that guesses at those is software you should uninstall. A send, a purchase, a delete or a settings change on screen waits for your approval in the app. Banking and payment apps are blocked from screen automation unless you go in and say otherwise. > [!IMPORTANT] The stop button is the feature > Screen automation is where an assistant can do real damage, so it is where OpenAlly is deliberately most boring. A run halts at whatever step it reached, leaves the app exactly where it stands, and hands the phone back to you. > Figure: **OpenAlly's guide figure with a raised palm. A phone beside it shows a paused automation and a locked bank card tile.** Screen control on Android with the brakes drawn in. A halt you can call, and a card the automation cannot reach. (light: https://openally.ai/images/blog/miro/miro-not-chat-aster-stop.webp; dark: https://openally.ai/images/blog/miro/miro-not-chat-aster-stop-dark.webp) #### The shop and the household share one workspace One account carries the till and the kitchen. The assistant that works out which order never paid can sort the shopping list afterwards, and neither errand needs a separate subscription or a second app to live in. For a household that is convenience. For a shop run out of a phone, it is the difference between an assistant and one more app to check. **Payments** is where a payment provider gets connected. Razorpay is the first one wired in, rather than the only one there will ever be. Razorpay's own documentation describes [payment links](https://razorpay.com/docs/payments/payment-links/) as a way to accept payments without a website or app. That is the situation when the sale happened inside a chat. An agent can raise the link and the order behind it. It can list what has arrived as structured data you can read against your own orders. Capture and refund stay owner-only, in the app, in your hands. [The Payments walkthrough](https://openally.ai/blog/payments-links-owner-only-refunds) sets out the full division of labour. WhatsApp is a door into that workspace rather than the product behind it. Contacts sit in tiers, so the assistant knows who it is talking to: Owner, Trusted, Member, Customer, Blocked. Group chats stay quiet until somebody says the assistant's name out loud. For example, a supplier chat can run for a week without the assistant saying a word, then answer the moment you call it in. A business that wants its own number answered uses [WhatsApp Business](https://openally.ai/product/whatsapp-business) instead: it connects through Meta, is free to connect with AI replies on Premium, and holds each billable AI message for the owner's approval unless it fits a budget the owner set. Telegram, Slack, Discord, Signal, Google Chat and iMessage are doors along the same corridor. They are useful when the work already lives there, and beside the point when it does not. LINE, Mattermost, Nextcloud Talk, Twitch and Zalo sit in that same public set. None of them is the product. [The shop and home guide](https://openally.ai/blog/one-workspace-for-shop-and-home) tells the longer version of this story. > Figure: **OpenAlly's guide figure between a shop counter tile and a kitchen shelf tile. One account marker joins the two.** Two halves of a life on one account: the till and the household, with no second subscription between them. (light: https://openally.ai/images/blog/miro/miro-not-chat-till-kitchen.webp; dark: https://openally.ai/images/blog/miro/miro-not-chat-till-kitchen-dark.webp) > Figure: **Android Connect Razorpay unlock sheet. It lists what the assistant may do, with a 500 rupee prompt chip and Refund a payment.** Owner sign-off sits in front of the merchant connection, and refunds stay owner-only. This is not a live ledger. (light: https://openally.ai/images/blog/ui/workspace-android-payments.webp; dark: https://openally.ai/images/blog/ui/workspace-android-payments-dark.webp) > Figure: **Mac Payments Features tab. Cards for Payment Links, Orders, Payments and Refunds, each carrying an Ask AI action.** The same money work on Mac, including owner approval on refunds. These are feature cards, not real amounts. (light: https://openally.ai/images/blog/ui/workspace-mac-payments.webp; dark: https://openally.ai/images/blog/ui/workspace-mac-payments-dark.webp) #### Across the room, on the Mac The Mac is continuity rather than a second product. Same account, same agents, same records, more screen. A conversation you started on the bus is open on the desk when you sit down, with the order it created still attached. **Remote Deck** is the app that turns the phone into a control deck for a Mac you have approved, reached over a private, verified link. Approve the machine once through your own account, and the phone runs the launchers and controls you set up on it: the next slide, the volume, the app you want in front. [The Remote Deck explainer](https://openally.ai/blog/what-is-openally-remote-deck) covers what the link does and does not reach. Mobile-first is the honest description of the whole product. Most of a working day happens between customers, on a handset, with one thumb. The Mac is where you sit down afterwards and read a month of orders instead of a single line. > Figure: **OpenAlly's guide figure between a phone held at chest height and an open laptop on a desk. A short mint line joins the two screens.** Continuity rather than a second product: the phone you started on, and the Mac that holds the same records. (light: https://openally.ai/images/blog/miro/miro-not-chat-mac-same.webp; dark: https://openally.ai/images/blog/miro/miro-not-chat-mac-same-dark.webp) > Figure: **Mac Remote Deck Overview. Host availability reads Ready, a phone-controller card reads Not connected, and it is waiting for your phone.** The Mac side of the link before a phone arrives. Reachable only by devices on the same account. (light: https://openally.ai/images/blog/ui/remote-deck-mac-approval.webp; dark: https://openally.ai/images/blog/ui/remote-deck-mac-approval-dark.webp) > Figure: **Android Remote Deck with Authenticating status, a Local network badge, and a line reading Controls are paused.** The deck is authenticating a pinned device on the local network. Controls stay paused until the private session connects. (light: https://openally.ai/images/blog/ui/remote-deck-android-pairing.webp; dark: https://openally.ai/images/blog/ui/remote-deck-android-pairing-dark.webp) > Figure: **Android Remote Deck open in the foreground. A dark grid of eight large tiles, with Refresh deck, Switch Mac and Disconnect in the menu.** The live deck on Android. It works while you look at it, and it stops when you leave. (light: https://openally.ai/images/blog/ui/remote-deck-android-foreground.webp; dark: https://openally.ai/images/blog/ui/remote-deck-android-foreground-dark.webp) #### A better Tuesday An assistant belongs on the hardware you already carry, because that is where the work turns up. The brain behind it swaps out, because the good one changes every few months and nobody should have to move house each time. The shop and the household run in one workspace. Banking and payment apps stay out of the automation's reach. That is a worse elevator pitch than "ChatGPT for India". It is a better Tuesday. Search will keep trying to turn OpenAlly into other products, so a few corrections in advance. A bundle identifier turning up somewhere is not an iPhone launch. Time-saved figures on this site are worked arithmetic from a stated workload rather than customer results. Install it on Android from [Google Play](https://play.google.com/store/apps/details?id=openally.ai), take the Mac build from [the Mac download page](https://openally.ai/download/mac), or start at [the download page](https://openally.ai/download) and pick your platform. Then grant it exactly one errand. The restock, probably. > Figure: **OpenAlly's guide figure ticking off the last item on a short Tuesday list. A phone and a Mac sit finished behind.** The whole promise, unglamorously: a shorter list at the end of an ordinary day. (light: https://openally.ai/images/blog/miro/miro-not-chat-tuesday.webp; dark: https://openally.ai/images/blog/miro/miro-not-chat-tuesday-dark.webp) #### How this was checked The methodology is a catalogue read. Every product line above comes from the OpenAlly catalogue and its public product pages, read on 2 September 2026. Where this page and the catalogue disagree, trust the catalogue. Razorpay wording was checked the same day against its payment links documentation. The MIT Sloan and IEEE Spectrum pages were opened and quoted directly. - Every external quotation above was opened and checked against the live page on 2 September 2026, and each of those pages is linked at the point where it is quoted. - On 8 October 2026 the generated public catalogue listed 29 apps, 44 skills, 8 tools, 12 channels and 46 model-provider rows. Those counts are a snapshot of that file, not a market survey. - Reading a skill as leave granted per agent, rather than as a feature on a list, is an editorial framing of the per-agent grant and the owner-only money rules. No catalogue card says it that way. About the author: [Matterward Labs](https://matterwardlabs.com) makes OpenAlly. It builds AI software, the apps around it, and the systems that run them. This article was reviewed by Matterward Labs on 2 September 2026, and corrections go to the [contact page](https://matterwardlabs.com/contact). #### Declared sources - [OpenAlly on Google Play](https://play.google.com/store/apps/details?id=openally.ai) - [MIT Sloan Ideas Made to Matter: Agentic AI, explained](https://mitsloan.mit.edu/ideas-made-to-matter/agentic-ai-explained) - [IEEE Spectrum: Edge AI and hardware constraints](https://spectrum.ieee.org/edge-ai) - [Razorpay Docs: Payment Links](https://razorpay.com/docs/payments/payment-links/) - [Matterward Labs](https://matterwardlabs.com) - [Matterward Labs contact page](https://matterwardlabs.com/contact) - [Aster companion releases on GitHub](https://github.com/satyajiit/aster-mcp/releases/latest) ### OpenAlly Payments: send a link, track the order, keep refunds owner-only OpenAlly Payments creates shareable payment links and orders, lists and reconciles incoming money, and keeps capture and refunds under owner control. Canonical: https://openally.ai/blog/payments-links-owner-only-refunds Markdown: https://openally.ai/blog/payments-links-owner-only-refunds.md Published: 2026-09-02 Updated: 2026-09-11 Reviewed: 2026-09-02 Author: Matterward Labs (https://matterwardlabs.com/) Category: Feature Keywords: payment links for a small business; track orders and reconcile payments; owner-only refunds; Razorpay payment provider; OpenAlly Payments app; reconcile incoming money Audience: Shop owners who send payment links and chase what has not arrived; People who are paid over chat and want a record of it; Owners who keep capture and refunds in their own hands Disclosure: Matterward Labs makes OpenAlly. Product claims come from the OpenAlly catalogue, reviewed 2026-09-02. Quoted lines come from the sources listed below. When a customer says yes in a chat, you still need a way to get paid, a durable record of the order, and a way to match the money when it lands. OpenAlly **Payments** covers that administrative stretch. An assistant can generate shareable links, raise orders, and reconcile arrivals. Capture and refund remain with you. #### Short answer **In brief:** **Payments** is an OpenAlly app and a matching skill. The app holds the provider keys. The skill is the leave you give one agent to use that app. Together they generate shareable payment links and orders, track incoming payments, and leave capture and refunds under exclusive owner control. Razorpay is the first provider, not the only one there will be. The Android app is on [Google Play](https://play.google.com/store/apps/details?id=openally.ai). The catalogue lists the Payments skill for Android, iPhone and Mac. > Figure: **OpenAlly's guide figure in gold glasses pointing at a phone that shows a chain-link icon, with an arrow to a card holding a green check and a gold coin** The short shape of Payments: a shareable link on a phone, then a settled payment. Capture and refund stay off this picture. (light: https://openally.ai/images/blog/miro/miro-payments-short-answer.webp; dark: https://openally.ai/images/blog/miro/miro-payments-short-answer-dark.webp) #### The Payments app and the Payments skill **Payments** is the OpenAlly app that connects a payment provider. The public tagline is: create payment links, track orders, and reconcile incoming money. **The Payments skill** is the leave you give one agent. Its public line is: create payment links and orders, track incoming payments, and keep capture or refunds under owner control. **An agent** is an assistant you set up for a job. Provider configuration lives in the Payments app. You do not type keys at an agent during a conversation. Give the skill to the agent that talks to buyers. Keep it off the agent that only files notes. > Figure: **OpenAlly's guide figure between a phone list of link, cart, and bag rows and a floating card with a wand and a small gear** The Payments app holds the provider connection. The matching skill is the leave you give one agent to use that app. (light: https://openally.ai/images/blog/miro/miro-payments-app-and-skill.webp; dark: https://openally.ai/images/blog/miro/miro-payments-app-and-skill-dark.webp) #### Send a shareable payment link **A payment link** is a web address you send so someone can pay you. You do not have to build a checkout page. Razorpay says [payment links](https://razorpay.com/docs/payments/payment-links/) "let you accept payments without a website or app". The same page says you "Generate a unique link from your Dashboard or via APIs and share it with customers via email, SMS, or social media". In OpenAlly you ask in plain words after the keys are in the app. For example, a repair shop can finish a visit in chat, ask for a link, and send it while the buyer is still there. Catalogue starters include "Create a ₹500 payment link for" and "List my recent payment links". > Figure: **OpenAlly's guide figure pointing at a phone while a hand holds a payment card with a rupee mark, a blank amount field, and a green confirm bar, joined by a chain-link** A payment link is an address you send so someone can pay. You do not have to build a checkout page first. (light: https://openally.ai/images/blog/miro/miro-payments-shareable-link.webp; dark: https://openally.ai/images/blog/miro/miro-payments-shareable-link-dark.webp) #### List orders and reconcile incoming money **An order** is the record of what they bought and what they owed, next to the link that should collect it. **Reconcile** means match money that arrived with the orders you expected. Your assistant can list recent payments and links as structured data (amount, status, order), then tell you which orders are still open. The Payments skill's catalogue entry puts it as "list and reconcile payments", and one of its starters is "Show my last 10 payments." For example, a stall can take orders all weekend. On Monday you ask for the last 10 payments and see which order never paid. > Figure: **OpenAlly's guide figure kneeling to place a check token on one of three order tiles, a shopping bag, a shirt, and a gift, joined in a dashed cycle** Reconcile means match money that arrived with the orders you expected. The tiles stand for those records, not live shop totals. (light: https://openally.ai/images/blog/miro/miro-payments-orders-reconcile.webp; dark: https://openally.ai/images/blog/miro/miro-payments-orders-reconcile-dark.webp) #### Capture and refund stay with the owner **Capture** takes the money. **Refund** sends it back. OpenAlly treats both as owner-only. Making a link, making an order, listing, and matching are normal work for an agent. > [!IMPORTANT] Money movement stays with the owner > An agent can create payment links and orders, list recent payments and links, and reconcile them against your orders. Capture and refund are owner-only: your AI can do them only with your owner sign-off, and refuses when a customer or anyone else asks. | Action | What it does | Who does it | | --- | --- | --- | | Create a payment link | Makes a shareable link the buyer pays on | Your agent, when you ask | | Create an order | Records what the payment is for | Your agent, when you ask | | List payments and links | Shows recent payments and links | Your agent, when you ask | | Reconcile against orders | Matches money to the orders you expected | Your agent, when you ask | | Capture a payment | Moves money in | Owner only: your AI, with your sign-off | | Refund a payment | Sends money back | Owner only: your AI, with your sign-off | > Figure: **OpenAlly's guide figure holding a gold key in front of a barred gate, with a checkmark control and a reverse-arrow control visible behind the bars** Capture takes the money. Refund sends it back. Both stay owner-only, behind a gate that only the owner's sign-off opens. (light: https://openally.ai/images/blog/miro/miro-payments-owner-only-capture-refund.webp; dark: https://openally.ai/images/blog/miro/miro-payments-owner-only-capture-refund-dark.webp) #### Shop till and household, same workspace Payments sits on the shop side of one OpenAlly workspace. Home tools can live in the same account. [The workspace guide](https://openally.ai/blog/one-workspace-for-shop-and-home) tells that wider story. Here the point is smaller. You configure the payment provider once inside Payments. Only agents granted the Payments skill can afterwards reach that merchant connection. The [Payments skill page](https://openally.ai/features/skills/payments) shows the access it asks for before you grant it. > Figure: **OpenAlly's guide figure seating a bank-building token at the start of a winding path that continues past a link, a document, and a refund mark** Razorpay is the first provider on the path, not a claim that it will be the only one. Keys go in the Payments app, not into a chat. (light: https://openally.ai/images/blog/miro/miro-payments-razorpay-first.webp; dark: https://openally.ai/images/blog/miro/miro-payments-razorpay-first-dark.webp) #### Android first, Mac for the long list Most payment links originate on a telephone, inside the conversation that requested one. Get Android from [Google Play](https://play.google.com/store/apps/details?id=openally.ai). Use [the Mac page](https://openally.ai/download/mac) for the Mac app, and [the download page](https://openally.ai/download) for the rest. The catalogue lists iPhone beside Android and Mac; that listing is not an App Store launch announcement. A larger Mac screen helps when you compare a month of orders against a month of payments, which is awkward to do in one hand. > Figure: **OpenAlly's guide figure walking a teal path with a chain-link disc, a phone panel on the left and a desk with a list on a monitor on the right** Most links start on a telephone. A larger Mac screen helps when you compare a month of orders against a month of payments. (light: https://openally.ai/images/blog/miro/miro-payments-android-mac.webp; dark: https://openally.ai/images/blog/miro/miro-payments-android-mac-dark.webp) > Figure: **Android Connect Razorpay unlock sheet listing what the assistant may do, with a 500 rupee prompt chip and Refund a payment** The Connect Razorpay unlock sheet on Android: owner sign-off, UPI, and a 500 rupee prompt. This is not a live orders ledger. (light: https://openally.ai/images/blog/ui/workspace-android-payments.webp; dark: https://openally.ai/images/blog/ui/workspace-android-payments-dark.webp) > Figure: **Mac Payments Features tab with cards for Payment Links, Orders, Payments and Refunds, each with Ask AI** Features cards on Mac, including owner approval on refunds. This is not the Connect Razorpay unlock sheet and not a live ledger of amounts. (light: https://openally.ai/images/blog/ui/workspace-mac-payments.webp; dark: https://openally.ai/images/blog/ui/workspace-mac-payments-dark.webp) #### What the owner still does - Put provider keys in the Payments app. - Choose which agents get the Payments skill. - Sign off a capture. - Sign off a refund. - Talk to the provider if their side fails. #### How this was checked The methodology is a catalogue read. Product lines above come from the OpenAlly catalogue for the Payments app and skill, read on 2 September 2026. If this page and the catalogue disagree, trust the catalogue. Razorpay wording was checked on the same day against its Payment Links docs. - The table was built for this page by lining up Payments with Sales and Invoicing. No one catalogue card shows that grid. - Treating money in and money out as separate leaves is a reading of the owner-only line, not a quote. #### What to read next See the [Payments app](https://openally.ai/features/apps/payments) and the [Payments skill](https://openally.ai/features/skills/payments). For the longer sale, [Sales](https://openally.ai/features/skills/sales) quotes a cart, takes an order, sends a payment option, issues an invoice, and reconciles when money lands. About the author: [Matterward Labs](https://matterwardlabs.com/) makes OpenAlly. It builds AI software, apps, and the systems that run them. This article was reviewed by Matterward Labs on 2 September 2026. For press, a bug, or other contact, use the inbox on that site. #### Declared sources - [OpenAlly on Google Play](https://play.google.com/store/apps/details?id=openally.ai) - [Razorpay Docs: Payment Links](https://razorpay.com/docs/payments/payment-links/) - [Matterward Labs](https://matterwardlabs.com/) ### What OpenAlly Remote Deck does: control your Mac from your phone Remote Deck turns your phone into a foreground controller for the Mac actions you approved, reached over a private, verified link. Canonical: https://openally.ai/blog/what-is-openally-remote-deck Markdown: https://openally.ai/blog/what-is-openally-remote-deck.md Published: 2026-09-02 Updated: 2026-09-11 Reviewed: 2026-09-02 Author: Matterward Labs (https://matterwardlabs.com) Category: Feature Keywords: what is OpenAlly Remote Deck; control Mac from phone; OpenAlly Remote Deck app; phone as Mac controller; private verified link Mac; Android Mac remote control Audience: You want to reach your Mac from your phone without walking back to the desk; You are deciding whether Remote Deck belongs on your phone and your Mac; You use OpenAlly already and want to know what the Mac link controls Disclosure: Matterward Labs makes OpenAlly. Product claims come from OpenAlly's own catalogue and the public Remote Deck feature page. The catalogue gives Remote Deck a single line: "Control your Mac from your phone over a private, verified link." The job underneath it is narrow and easy to picture. You are away from the desk, something on the Mac still needs doing, and the phone already in your hand becomes the surface you do it from. #### Short answer > Figure: **OpenAlly's guide figure holding a phone flat like a control pad while a Mac laptop sits open on the desk behind** The pose that opens the article: the phone in your hand, the Mac still doing the work. (light: https://openally.ai/images/blog/miro/miro-remote-deck-short-answer.webp; dark: https://openally.ai/images/blog/miro/miro-remote-deck-short-answer-dark.webp) **In brief:** **Remote Deck** is an OpenAlly app that turns your phone into a control deck for a Mac you have approved. You design the deck on the Mac, placing app launchers and controls such as media, volume, slides, windows and a screenshot, then run them from the phone. Pairing goes through your own account, and only devices you approve can connect. The public catalogue lists Remote Deck for Android, iPhone and Mac. The Android app comes from [Google Play](https://play.google.com/store/apps/details?id=openally.ai), and the Mac app is a direct download from [the Mac download page](https://openally.ai/download/mac) rather than a Mac App Store listing. #### What Remote Deck actually controls > Figure: **OpenAlly's guide figure pointing at a phone as a tick appears beside an action on an approved Mac** Approval is the boundary: the Mac you linked, and the actions that Mac has approved. (light: https://openally.ai/images/blog/miro/miro-remote-deck-approved-actions.webp; dark: https://openally.ai/images/blog/miro/miro-remote-deck-approved-actions-dark.webp) The catalogue copy is deliberately narrow: run the launchers and controls you set up on your approved Mac, from your phone. Approval is the word carrying the weight. Remote Deck reaches the Mac you linked and the actions that Mac has approved. It is a control deck for that machine, not a general remote-access product: it does not show the Mac's screen, and it does not pass your typing through. The [Remote Deck feature page](https://openally.ai/features/apps/remote-deck) puts what you get in three lines: - **Your Mac, from your phone.** Run the app launchers and controls you set up on your Mac from a deck on your phone, over a private, verified link. - **Verified pairing.** Only devices you approve can connect through your account. - **A deck you design.** Place launchers and controls in Deck Studio on the Mac. Each layout is one page you swipe on the phone, and a layout can follow the app in front. The Mac is where the deck is made. The app's own controls include play or pause, the next and previous track, volume and mute, the next and previous slide, minimising, closing or fullscreening the focused window, and a screenshot saved to the folder set on the Mac. An app launcher opens a Mac app or brings it to the front. The phone runs what the Mac offers and cannot create or change an action. For example, a presentation can move on from the other side of the room, with the laptop still on the lectern. | What you do | What Remote Deck does | | --- | --- | | Open the app on the phone | The phone becomes the deck while it stays in the foreground | | Pair through your account | Only a Mac you approve can connect | | Tap a control on the phone | The approved Mac carries out that one action | | Leave the app | The deck stops, rather than driving the Mac behind other apps | > [!IMPORTANT] The screens on the feature page are a demonstration > The machine name and the controls shown on the public Remote Deck page exist to show the shape of the flow. They are samples, not measurements. No figure on that page reports speed, usage, or how many devices anyone has linked. #### Pairing through a private, verified link > Figure: **OpenAlly's guide figure holding up a phone and a Mac joined by a short marked line between the two screens** Pairing runs once through the account you already sign in to, which is what verified means here. (light: https://openally.ai/images/blog/miro/miro-remote-deck-verified-link.webp; dark: https://openally.ai/images/blog/miro/miro-remote-deck-verified-link-dark.webp) > Figure: **Mac Remote Deck Overview: Host availability Ready, a phone-controller card marked Not connected, and Waiting for your phone** The Mac host approval surface: this Mac is Ready for a phone, controller access is same-account mesh devices, and no phone is connected yet. (light: https://openally.ai/images/blog/ui/remote-deck-mac-approval.webp; dark: https://openally.ai/images/blog/ui/remote-deck-mac-approval-dark.webp) > Figure: **Android Remote Deck with Authenticating status, a Local network badge, and Controls are paused** The deck is authenticating a pinned device on the local network. Controls stay paused until the private session is connected. This is not a QR pairing screen. (light: https://openally.ai/images/blog/ui/remote-deck-android-pairing.webp; dark: https://openally.ai/images/blog/ui/remote-deck-android-pairing-dark.webp) You pair once, through the account you already sign in to. The Mac you approve becomes a linked device; a Mac you have not approved is simply not reachable from the phone. The phone lists only verified Macs on your local network or your Tailscale network, and controls one of them at a time. That is the practical meaning of the tagline: the link is private between your own devices, and it is verified because approval is an act you perform rather than a default you inherit. The [Remote Deck feature page](https://openally.ai/features/apps/remote-deck) states the rule in one line: "Only devices you approve can connect through your account." The app keeps the other half too: a Mac you withdraw shows as Revoked, and it has to be enrolled again before the phone can drive it. Approval also stays yours to withdraw. A device you approved is a device you can stop approving, and the phone loses the deck the moment the Mac stops being a linked device. #### The phone stays a foreground controller > Figure: **OpenAlly's guide figure holding a phone out in front with the screen facing the reader and both hands on it** A controller you hold and watch, rather than something arranging itself quietly behind your other apps. (light: https://openally.ai/images/blog/miro/miro-remote-deck-foreground.webp; dark: https://openally.ai/images/blog/miro/miro-remote-deck-foreground-dark.webp) > Figure: **Android Remote Deck open in the foreground: a dark grid of eight large tiles, with Refresh deck, Switch Mac — This Mac, and Disconnect in the menu** The live Android deck while it is on screen: control tiles for the session, plus Refresh, Switch Mac, and Disconnect. This is the deck itself, not a pairing screen. (light: https://openally.ai/images/blog/ui/remote-deck-android-foreground.webp; dark: https://openally.ai/images/blog/ui/remote-deck-android-foreground-dark.webp) Remote Deck is a foreground controller. You open it, you use it while it is in front of you, and it stops when you leave. The phone is the surface you tap; the Mac is where the work carries on happening. Nothing about the app is arranged to sit behind your other apps and reach the Mac while you are looking elsewhere. When you leave, the app shows "Controller paused" and says why in its own words: "Return to OpenAlly to reconnect. Commands are never queued in the background." That shape is why the app is worth having on a phone at all. A controller you hold is one you can watch, and watching it is how you notice that the tap landed on the thing you meant. #### Remote Deck on Android and Mac > Figure: **OpenAlly's guide figure standing between an Android phone held at chest height and an open Mac laptop on a desk** The two surfaces this section covers: the phone you carry, and the Mac you approved. (light: https://openally.ai/images/blog/miro/miro-remote-deck-android-mac.webp; dark: https://openally.ai/images/blog/miro/miro-remote-deck-android-mac-dark.webp) The Android app is the OpenAlly app on [Google Play](https://play.google.com/store/apps/details?id=openally.ai). The Mac app is a direct download from [the Mac download page](https://openally.ai/download/mac) rather than a Mac App Store listing. The public catalogue lists Remote Deck for iPhone as well, alongside Android and Mac. The Mac side is not a separate account or a separate body of work. Your account, your agents and your chat history continue on the Mac, so the machine you drive from the phone is the same machine that holds the conversation you started earlier on the phone. #### How it differs from Slap > Figure: **OpenAlly's guide figure raising one hand beside a phone while a Mac sits undisturbed on the desk alongside** A quick contrast: Slap answers a knock, and Remote Deck drives the Mac you approved. (light: https://openally.ai/images/blog/miro/miro-remote-deck-not-slap.webp; dark: https://openally.ai/images/blog/miro/miro-remote-deck-not-slap-dark.webp) [Slap](https://openally.ai/blog/what-is-openally-slap) is the app that turns a desk tap, a device shake or the charger going in into a meme soundbite, while Remote Deck is the one that drives Mac-approved actions. Both open as their own screen rather than living inside a conversation, which is the family they belong to. #### Where Remote Deck sits in the catalogue > Figure: **OpenAlly's guide figure beside a grid of app cards with one card lifted forward from the others** Where Remote Deck sits among the focused apps in OpenAlly's public catalogue. (light: https://openally.ai/images/blog/miro/miro-remote-deck-catalogue.webp; dark: https://openally.ai/images/blog/miro/miro-remote-deck-catalogue-dark.webp) Remote Deck is one of the focused apps that open as their own screen. The [app catalogue](https://openally.ai/features/apps) lists it for Android, iPhone and Mac; the [full catalogue](https://openally.ai/features) puts it beside the skills, tools and agents it shares a workspace with; and the [FoneClaw comparison](https://openally.ai/blog/openally-vs-foneclaw) walks the same collection of focused apps from a different angle. The [download page](https://openally.ai/download) covers each platform in turn. If you read that comparison first, the phrase you will recognise here is the one it uses for this app: the phone becomes a custom control deck for actions the Mac has approved. #### Who makes it, and where these claims come from > Figure: **OpenAlly's guide figure holding a printed page in one hand and checking it against a short list held in the other** Every claim above is checked against OpenAlly's own catalogue and the public Remote Deck page. (light: https://openally.ai/images/blog/miro/miro-remote-deck-sources.webp; dark: https://openally.ai/images/blog/miro/miro-remote-deck-sources-dark.webp) About the author: [Matterward Labs](https://matterwardlabs.com) makes OpenAlly. We checked OpenAlly's own catalogue and the public Remote Deck feature page on 2 September 2026 (Matterward Labs, 2026). On 11 September 2026 we corrected what the deck does against the app's own screens and labels: the earlier text said the phone shows the Mac's screen and types on it, and it does neither. The tagline, the three feature lines and the platform list are the ones the live pages display. If something here reads wrong, corrections go to the company [contact page](https://matterwardlabs.com/contact). #### Declared sources - [OpenAlly on Google Play](https://play.google.com/store/apps/details?id=openally.ai) - [Matterward Labs](https://matterwardlabs.com) - [Matterward Labs contact page](https://matterwardlabs.com/contact) ### How an AI agent gets things done on your phone, and what Google now allows What lets an AI agent tap, type and scroll inside your other apps, what Google Play permits, and where OpenAlly hands control back to you. Canonical: https://openally.ai/blog/how-ai-agents-do-things-on-your-phone Markdown: https://openally.ai/blog/how-ai-agents-do-things-on-your-phone.md Published: 2026-08-31 Updated: 2026-09-11 Reviewed: 2026-08-31 Author: Matterward Labs (https://matterwardlabs.com) Category: Guide Keywords: how AI agents control your phone; AI agent that uses other apps; phone agent screen control; Android accessibility automation policy; App Automations; AI agent that taps and types for you Audience: People who want an AI agent on their phone that can use other apps; Android users deciding whether to allow screen automation; Anyone comparing what a phone agent is allowed to touch Disclosure: Matterward Labs makes OpenAlly. Product claims come from OpenAlly's own catalogue and its Google Play listing. Google's policy is quoted word for word from Play Console Help and Android Help, reviewed by Matterward Labs on 31 August 2026. An assistant that only talks needs nothing else on your phone. Ask it to post an update or book a ride, and that changes. It now needs a way into an app you already use. On Android the phone grants that reach, and Google Play policy sets the limits. The policy is clear about what an app may decide for you. > [!IMPORTANT] What this article does and does not claim > About the author: Matterward Labs makes OpenAlly. Google's policy below is quoted word for word from Play Console Help and Android Help, reviewed by Matterward Labs on 31 August 2026. OpenAlly's behaviour is quoted from its own catalogue and its Google Play listing. Nothing here assesses any app against Google's policy, including ours. #### Short answer **In brief:** an AI agent can act inside your other apps only where the phone and the store allow it. Google Play bars an app from acting on its own through the Accessibility API. It allows a fixed script that a person wrote. In OpenAlly, two features work inside the screens of your other apps, App Automations for saved flows and Aster screen control for one-off tasks, and both drive the screen through Aster, a second app you install yourself on Android. [Google Play's accessibility policy](https://support.google.com/googleplay/android-developer/answer/10964491) says that **"Any use of the Accessibility API that enables an app to autonomously initiate, plan, and execute actions or decisions is strictly prohibited."** The next sentence carves out an exception. It reads: **"This does not prohibit deterministic, rule-based automation, where behavior follows a static, human-defined script."** So the line sits between deciding and following. **App Automations** is the OpenAlly feature that opens your other apps and works them for you, and it is listed for Android. Its catalogue entry says your AI "opens an app and taps, types, and scrolls through a flow for you". It also says the feature "hands off to you at logins and payments", and that "Every trigger has a daily ceiling you set". The [Google Play listing](https://play.google.com/store/apps/details?id=openally.ai) states the boundary in one line: "It never pays, and never sees your password." Payments and passwords stay with you. Driving the screen itself happens in that separate, optional install. Aster is a companion app you put on the same Android phone yourself, from outside the Play Store, and grant on its own. App Automations needs it too: its catalogue page says it "reaches the screen through the Aster companion". Where the agent runs, and whether it works without a signal, are different questions from this one. The table near the end of this guide points to both answers. #### How touching another app actually works There are three kinds of reach. They hand over very different amounts. - **Its own chat.** The agent reads what you typed and writes back. It can think, summarise and draft. It cannot see or change anything in Instagram, your bank or your calendar. - **A connected account.** You sign in once and let the agent use a service through the door that service provides. The scope is whatever the service chose to expose. You can revoke it there. - **Working the screen of another app.** The agent reads what is shown, then taps, types and scrolls the way you would. On Android that reach usually comes through the Accessibility API, the interface that lets an app read what is on screen and act on it. It needs no help from the app's owner. That is why it reaches furthest, and why it is watched most closely. This third kind is what people mean by an AI agent that controls their phone. It is the one with a written policy attached: Google Play's [policy on the AccessibilityService API](https://support.google.com/googleplay/android-developer/answer/10964491), quoted in the next section. #### What Google Play allows Google draws its line around autonomy. The policy does not ban automation as such. **The Accessibility API** is the Android interface that lets an app read what is on screen and act on it. Google's rule is about what an app decides to do with that reach. Play Console Help names the banned behaviour directly: **"Any use of the Accessibility API that enables an app to autonomously initiate, plan, and execute actions or decisions is strictly prohibited."** That sentence is about deciding, not about tapping. The permitted behaviour comes next: **"This does not prohibit deterministic, rule-based automation, where behavior follows a static, human-defined script (for example, 'If Trigger X occurs, perform Action Y')."** So a fixed script is allowed. The same page adds a scope test. It says that **"Apps using the Accessibility API for automation purposes must ensure that all actions performed on a user's behalf are for a narrow and clearly understood purpose."** A [second Play policy page](https://support.google.com/googleplay/android-developer/answer/9888170) covers the paperwork. It says the Accessibility API **"is not designed and cannot be requested for"** remote call audio recording. The same bar applies to an app that autonomously initiates, plans and executes actions or decisions. The page also requires that **"The use of the Accessibility API must be documented in the Google Play listing"**. An app that has not declared itself an accessibility tool has more to do. It must complete a declaration in Play Console. It must also show a clear in-app disclosure and ask for consent. This is a rule about a behaviour, not about a kind of company. An app that picks its own course of action across your apps is out. An app that follows steps a person set out is in. The policy says both things in the same paragraph. The declaration language on that page still highlights 3 November 2021. That was when apps targeting API level 31 with an accessibility service first had to file a policy declaration. #### What OpenAlly App Automations does The catalogue entry for App Automations names the flows it ships with: post to LinkedIn, search and play on YouTube, summarise Instagram activity, book an Uber up to payment. For example, you give the LinkedIn flow a topic and it writes the post, and you pick from a few drafts before it publishes. The catalogue describes the first run in one sentence: **"Describe the flow once and your AI walks it on the real screen to learn the steps, then replays them fast and deterministically."** The Automations screen offers two ways to create one. **Record with AI** has your assistant open the app and demonstrate the flow; when you have described a goal instead, **Build it from your goal** has it work toward that goal step by step and record what it does. **Build it step by step** has you add each action yourself in the editor, with no recording. A recorded flow then waits for you: the app lists the captured steps and says "Nothing replays until you save." After that first run: - Replay is deterministic, and it self-heals a step if the app changes. - It hands off to you at logins and payments. - You can schedule one. You can also have it run when something happens on your phone: "a shake, a plug-in, an app opening, a track changing, a notification your Flare rule matched". - "Every trigger has a daily ceiling you set, and a triggered run can only get things ready for you: it never buys anything." - It runs on-device, with your confirmation. The Play listing says the same thing in customer language: "It self-heals when the screen changes and hands control back to you at logins and payments. It never pays, and never sees your password." Those are the two hand-off points. The [feature catalogue](https://openally.ai/features) puts it in one line: "Turn familiar Android app flows into repeatable shortcuts, with your review at logins, payments, and other sensitive steps." Before you set one up, check where it stops. For example, the Uber flow gets a ride ready up to payment and then waits for you. A flow that finished the booking would be doing something else. That boundary is set by the design, not by how the flow is worded. #### Optional Aster screen control Screen control on Android does not arrive with the main app. **Aster** is a separate companion that you install yourself, from outside the Play Store, on the same phone. OpenAlly's catalogue describes what it adds once you allow it: "drive your screen to complete multi-step tasks for you, all on-device with your confirmation." Grant that access and you grant Aster alone. You can stop Aster on its own without disturbing anything else. Google's policy language above applies to any app that uses the Accessibility API. Assessing a particular app against it is Google's job. What we can state is the product shape. Aster is optional. It is installed on its own, granted on its own and stopped on its own. Screen control is scoped to Android rather than to OpenAlly as a whole. OpenAlly itself is a cross-platform product, and [the download page](https://openally.ai/download) is the source of truth for where it runs. > Interface view: **Screen control is a separate grant.** An approved Android screen task moving from OpenAlly to the optional Aster companion. #### What Google ships on some flagship phones Google has its own path on a few phones. Its [Gemini Intelligence support page](https://support.google.com/android/answer/17499451) says Gemini "can help you automate multi-step tasks across select Android apps, like ordering groceries, making reservations or shopping online." The list of phones is short. As of 31 August 2026 that page lists Samsung Galaxy Z Fold8, Samsung Galaxy Z Fold8 Ultra, Samsung Galaxy Z Flip8, Pixel 11, Pixel 11 Pro, Pixel 11 Pro XL and Pixel 11 Pro Fold. The stated requirement is 12GB+ RAM. The page also notes that individual features "may vary by country and language" and that "Features may also vary by device type." If you carry one of those phones, that route is open to you without installing anything. #### Three different questions about a phone agent These three get collapsed into one. They come apart the moment you put a phone agent to work. | The question | What it decides | Where we answer it | | --- | --- | --- | | Where does the agent run? | Whether a second machine has to stay switched on | [An AI agent on your phone without a PC](https://openally.ai/blog/ai-agent-on-your-phone-without-a-pc) | | Does it work without a connection? | Which requests survive a dead signal | [What a phone can do offline](https://openally.ai/blog/ai-on-your-phone-offline) and [running a model on an Android phone](https://openally.ai/blog/gemma-4-android-litert-lm) | | May it touch your other apps? | Whether it can act outside its own chat | This article | An agent can run entirely on your phone and still have no permission to open Instagram. It can work Instagram well and still need a signal for every reply. Two comparisons walk the same three questions against named products, [OpenAlly and CellClaw](https://openally.ai/blog/openally-vs-cellclaw) and [OpenAlly and FoneClaw](https://openally.ai/blog/openally-vs-foneclaw). The [small-business guide](https://openally.ai/blog/ai-agent-runs-small-business) shows what this reach is for once you have it. #### Frequently asked questions ##### Can an AI agent control my phone without approving each step? Google Play's accessibility policy prohibits an app that uses the Accessibility API to "autonomously initiate, plan, and execute actions or decisions". It also requires that automated actions be "for a narrow and clearly understood purpose". OpenAlly's App Automations entry describes itself as running "On-device, with your confirmation". It hands off at logins and payments, and it gives every trigger a daily ceiling you set. ##### Do I need the Aster companion to use OpenAlly? No. Aster is optional and separate, but App Automations and Android screen control both run through it. You install it yourself on the same Android phone, from outside the Play Store, and grant it screen access on its own. Without it you still get the rest of OpenAlly. ##### Can an automation buy something while I am not looking? The catalogue is explicit that a triggered run "can only get things ready for you: it never buys anything". The Google Play listing says "It never pays, and never sees your password." Payments and logins hand control back to you. Every trigger has a daily ceiling you set. ##### What exactly does Google Play prohibit? Autonomy, in the narrow sense of an app that initiates, plans and executes actions or decisions by itself through the Accessibility API. Deterministic automation that follows a static, human-defined script is carved out in the [same policy paragraph](https://support.google.com/googleplay/android-developer/answer/10964491). A [second page](https://support.google.com/googleplay/android-developer/answer/9888170) requires that any use of the Accessibility API be documented in the Google Play listing. Apps that are not accessibility tools also need a Play Console declaration and an in-app disclosure. #### Declared sources - [Google Play: Use of the AccessibilityService API](https://support.google.com/googleplay/android-developer/answer/10964491) - [Google Play: Permissions and APIs that Access Sensitive Information](https://support.google.com/googleplay/android-developer/answer/9888170) - [Android Help: Gemini Intelligence on your phone](https://support.google.com/android/answer/17499451) - [OpenAlly on Google Play](https://play.google.com/store/apps/details?id=openally.ai) ### One OpenAlly workspace for the shop and the household One OpenAlly workspace where your AI runs a shop's stock, GST, payments and invoices beside a household's everyday admin, on your phone and on a Mac. Canonical: https://openally.ai/blog/one-workspace-for-shop-and-home Markdown: https://openally.ai/blog/one-workspace-for-shop-and-home.md Published: 2026-08-31 Updated: 2026-10-06 Reviewed: 2026-08-31 Author: Matterward Labs (https://matterwardlabs.com) Category: Guide Keywords: one workspace for business and home; AI assistant for a small shop; AI for household admin; GST and stock on your phone; AI invoicing for a small business; OpenAlly on Android and Mac Audience: Small-shop owners who run stock, prices and payments themselves; People running a household who want daily admin in one place; Anyone using the same device for work and for home Disclosure: Matterward Labs makes OpenAlly. Product claims come from OpenAlly's own catalogue. External figures, if any, are linked to their original publishers. A shop and a household run on the same small errands. Something is low and needs ordering, a price needs checking, a bill needs sending, and somebody wants an answer before the day is out. OpenAlly keeps both sets in one workspace. The tools that price a product, take a payment and issue an invoice sit beside the ones that settle what is for dinner and keep the medicines organised. Start on your phone or at the Mac, and the same account picks the work up on the other. #### Short answer **In brief:** one OpenAlly workspace covers a small shop and a household, because the same assistant can reach both sets of tools. On the shop side that means a product catalogue, stock levels, tax, payment links, orders and invoices. At home it means everyday administration, group choices, health records and a private knowledge base. The Android app is [on Google Play](https://play.google.com/store/apps/details?id=openally.ai), and the Mac app continues the same work. You decide which agent gets which of those tools. An assistant that runs the shop has no reason to hold your medicine list, and the one that helps at home has no reason to read your orders. > Figure: **OpenAlly's guide figure holding a phone in one hand and pointing at a floating product card that has just been ticked green** Shop tools and household tools can live in one workspace. You choose which agent is allowed which set. (light: https://openally.ai/images/blog/miro/miro-workspace-phone.webp; dark: https://openally.ai/images/blog/miro/miro-workspace-phone-dark.webp) **An agent** is an assistant you configure and point at the work you assign it. Keep a single one for everything, or hand focused work to specialists, such as separate agents for the shop and for the household. You control what each one may use, and delegated work comes back to the same conversation. Knowledge bases work the same way. Add PDFs, web pages and text, such as a supplier's terms or a tenancy agreement, and an agent answers from that material rather than from its own general knowledge. Knowledge stays organised on the device, and reaches an outside model only when you choose a request that needs it. | What you ask for | What it uses | Who has to approve it | | --- | --- | --- | | Show me what is low on stock | The Business catalogue and its stock levels | Nobody. Reading changes nothing | | Create a payment link for that order | Payments, through the provider you connected | You, when you connected the provider | | Refund that payment | Payments | Only you. Money leaving is a separate permission from money arriving | | Put dinner to a vote for the four of us | Group Decisions, on the device | Everyone votes, then you review the plan | | What medicines are due tonight | The health records you allowed | Nobody. It is a read | #### The shop side: stock and prices > Figure: **OpenAlly's guide figure at a mint shop counter beside a board of products, each row with a price tag and stock dots** This half holds what you sell and what it cost you. Stock is counted with it. (light: https://openally.ai/images/blog/miro/miro-workspace-shop.webp; dark: https://openally.ai/images/blog/miro/miro-workspace-shop-dark.webp) **Business** is the OpenAlly app that holds what you sell: a product catalogue and its stock, kept in your OpenAlly account. You create inventories, each with one currency, add products to a shared pool, and list them per inventory with cost, MRP, sale price and stock. Categories keep the listing usable once it grows past a screen. Once the catalogue is in, you can stop opening the app for ordinary jobs. Ask your AI instead to: - add a product, or restock one that has run down; - find what is running low, before a customer finds it for you; - value the stock you are holding right now. Deleting an entire inventory is yours alone: an assistant does it only when you ask, and refuses when anyone else does. The [Business entry in the catalogue](https://openally.ai/features/apps/business) shows what the app does and which skills your AI uses to drive it. #### Tax belongs with the catalogue > Figure: **OpenAlly's guide figure pointing at a diagram where one box branches into cards labelled CENTRAL GST and STATE GST** GST is set up next to the products, not left for the end of the month. Quotes follow that setup. (light: https://openally.ai/images/blog/miro/miro-workspace-tax.webp; dark: https://openally.ai/images/blog/miro/miro-workspace-tax-dark.webp) You set GST up alongside the products: parallel CGST and SGST components, inclusive or exclusive, and tax quoted on any amount you name. India's Central Board of Indirect Taxes and Customs [defines SGST](https://cbic-gst.gov.in/about-gst.html) as "State Goods and Services Tax, one of the three categories under Goods and Services Tax (CGST, IGST and SGST)". The same CBIC page sets out who administers each part: the "Centre will levy and administer CGST & IGST while respective states will levy and administer SGST". Match your rates to that split once, and a customer asking what something costs with tax is answered from the catalogue instead of from arithmetic at the counter. #### Getting paid, and who is allowed to move money > Figure: **OpenAlly's guide figure holding a card marked with a chain-link icon beside a panel of five rows, each ticked green** Your assistant raises the link and reads back what has settled. Capture and refund stay with you. (light: https://openally.ai/images/blog/miro/miro-workspace-pay.webp; dark: https://openally.ai/images/blog/miro/miro-workspace-pay-dark.webp) **Payments** is the app that connects a payment provider and then lets your assistant work with it. Razorpay is the first provider it supports. You add your provider keys in [Payments](https://openally.ai/features/apps/payments), then ask for a link or an order in ordinary language. Razorpay's own documentation puts the idea plainly: [payment links](https://razorpay.com/docs/payments/payment-links/) "let you accept payments without a website or app". From there your assistant can create shareable payment links and orders, list payments, and reconcile them against the orders you were expecting. > [!IMPORTANT] Money movement stays with the owner > Capture and refund are owner-only. An agent can raise a payment link, list what has arrived, and reconcile it against your orders. Money leaving is a separate permission from money arriving, so taking a payment or sending one back waits for you. Deleting a whole inventory is owner-only in the same way. [Invoicing](https://openally.ai/features/skills/invoicing) closes the loop. An order's invoice arrives as a card in the chat that you can print, share, save or send. You can restyle it with a built-in template, classic, modern, minimal, bold or elegant, and a curated theme colour, then deliver it to the buyer on their own channel. Reads carry no customer details. Invoicing is a per-agent skill you hand to the agents that should have it, and the Executive Assistant always has it. #### The household side: daily admin, choices and records > Figure: **OpenAlly's guide figure at a kitchen table with a dinner vote card and a medicines checklist** The household half is the same workspace with a different set of tools switched on. Nothing is duplicated. (light: https://openally.ai/images/blog/miro/miro-workspace-household.webp; dark: https://openally.ai/images/blog/miro/miro-workspace-household-dark.webp) At home the same agents and the same conversation do the work, and only the tools they may reach change. **Group Decisions** is the skill for questions that need more than one opinion, such as dinner tonight, which film to watch, or where to meet. Your AI writes the options down, records each person's answer as they reply, and tells you where the vote stands. One person, one vote, and changing your mind replaces your earlier answer. A tie does not get broken for you. Closing a decision writes a plan for you to review, and closing it can never order anything, change a cart or spend money, however the vote went. Decisions stay on the device and expire after six hours. Read the closed plan yourself before anything is bought or booked; the [Group Decisions entry](https://openally.ai/features/skills/group-decisions) sets out the rest. [Health](https://openally.ai/features/apps/health) keeps records and medicines on the device, so you can ask what is due tonight or look up how a measurement has moved since the spring. It never diagnoses or treats, and it is not a medical device. For anything clinical, contact a clinician. A household can use the shop tools too. The Business catalogue will hold nappies and washing powder as happily as it holds saleable stock. #### From the phone to the Mac > Figure: **OpenAlly's guide figure seated at a mint desk, laptop open on a split view of status dots and a ticked checklist** Bigger screen, same conversation. The work you started standing at the counter is still open here. (light: https://openally.ai/images/blog/miro/miro-workspace-mac.webp; dark: https://openally.ai/images/blog/miro/miro-workspace-mac-dark.webp) OpenAlly is for people and for businesses, on a phone and on a Mac. A stock check fits in a pocket, and a month of invoices is easier on a desk. The Mac app carries the same account, agents and chat history. Business, Payments, Invoicing, Sales, Group Decisions and Health are all catalogued for Android, iPhone and Mac. [The Mac page](https://openally.ai/download/mac) has the current build, [the download page](https://openally.ai/download) covers the rest, and the Android app sits on [Google Play](https://play.google.com/store/apps/details?id=openally.ai). Selling through a conversation has a skill of its own. [Sales](https://openally.ai/features/skills/sales) quotes a cart from your catalogue, takes an order, and sends a payment link and QR or takes cash on delivery. It then issues a numbered invoice and reconciles when the money lands. You set which shop and which channel it may use before any of that runs. The [feature catalogue](https://openally.ai/features) shows the access each skill asks for before you grant it, and [the small-business guide](https://openally.ai/blog/ai-agent-runs-small-business) is the longer version of this story, following one customer from a search through a booking to a paid invoice. About the author: [Matterward Labs](https://matterwardlabs.com) makes OpenAlly. Product claims here come from OpenAlly's own catalogue, and this guide was reviewed by Matterward Labs before publication. The Razorpay and CBIC wording is quoted from the pages linked above (both checked on 31 August 2026). #### Declared sources - [OpenAlly on Google Play](https://play.google.com/store/apps/details?id=openally.ai) - [Razorpay Docs: Payment Links](https://razorpay.com/docs/payments/payment-links/) - [CBIC, Government of India: Know About GST](https://cbic-gst.gov.in/about-gst.html) - [Matterward Labs](https://matterwardlabs.com) ### What OpenAlly Slap does: tap, shake, meme soundbites What OpenAlly Slap does: a desk tap, a device shake or the charger going in plays a meme soundbite, with sound packs, plug sounds and a leaderboard. Canonical: https://openally.ai/blog/what-is-openally-slap Markdown: https://openally.ai/blog/what-is-openally-slap.md Published: 2026-08-31 Updated: 2026-08-31 Reviewed: 2026-08-31 Author: Matterward Labs (https://matterwardlabs.com) Category: Feature Keywords: what is OpenAlly Slap; OpenAlly Slap app; meme soundbite app; slap detection sound app; phone shake sound effect; sound packs for slap; slap leaderboard; Mac desk tap sound Audience: People who want a playful reaction from their phone or Mac; Anyone deciding whether to switch on Slap and its sound packs; OpenAlly users curious about plug sounds and the leaderboard Disclosure: Matterward Labs makes OpenAlly. Product claims come from OpenAlly's own catalogue, the public Slap feature page, and the app's own Slap screens. Hearing figures are linked to their publisher. Tap the desk your phone is sitting on, shake the device, or plug in the charger, and a meme soundbite plays straight back at you. The catalogue tagline is "Slap your machine and it barks back a meme soundbite", and none of it is a metaphor. #### Short answer **In brief:** **Slap** is an OpenAlly app that picks up a real desk tap, a device shake, or a charger connection. It is a playful reaction, not a security feature, and it does not guard the device. It makes a noise and keeps a count. > Figure: **OpenAlly's guide figure tapping a finger on a round mint table where a phone stands, with sound arcs and a music note rising from it** One tap on the surface the phone is resting on, and the clip comes back. This is the pose the whole feature is built around. (light: https://openally.ai/images/blog/miro/miro-slap-tap.webp; dark: https://openally.ai/images/blog/miro/miro-slap-tap-dark.webp) Each impact plays a meme soundbite from the sound pack you turned on. The public [app catalogue](https://openally.ai/features) lists Slap for Android, iPhone and Mac. The screens below are the Android app. > Figure: **Slap tab home on Android, with lifetime slaps at 0 and rank Fresh Hands** The Slap home on a fresh install: nothing hit yet, so nothing counted. It is a settings surface, not a busy activity board. (light: https://openally.ai/images/blog/ui/slap-android-home.webp; dark: https://openally.ai/images/blog/ui/slap-android-home-dark.webp) React to slaps is off by default, so nothing happens until you turn it on. Lifetime slaps reads 0 and the rank is Fresh Hands, which the app annotates in its own deadpan: "Zero slaps. Your device is either spared or deeply neglected." The leaderboard opens from here. The rest of the tab is settings. #### What sets Slap off Four things set Slap off: a knock on the surface the device is resting on, a shake of the device itself, a charger going in or out, and a USB device connecting on a Mac. A Try it control fires the same reaction on demand. Any of them only counts when the device really feels it, and Sensitivity is where you set how hard that has to be. > Figure: **OpenAlly's guide figure shaking a phone in both hands as teal and purple lightning bolts and motion arcs fly off it** A shake runs at the same sensitivity you set for a desk tap. Same threshold, different gesture. (light: https://openally.ai/images/blog/miro/miro-slap-shake.webp; dark: https://openally.ai/images/blog/miro/miro-slap-shake-dark.webp) On a Mac the bar is higher. It wants a real hit on the chassis or the desk under it, not a soft trackpad tap. > Figure: **Slap tab on Android scrolled to Sensitivity, Playback, and an Anti-Theft row** The same Slap tab holds Sensitivity, Playback, and Anti-Theft as separate controls. Slap itself is still only a reaction. (light: https://openally.ai/images/blog/ui/slap-android-trigger.webp; dark: https://openally.ai/images/blog/ui/slap-android-trigger-dark.webp) Sensitivity offers Low for hard slaps only, Medium as the balanced default, and High where light taps count. Under Playback, Harder slaps play louder scales the volume with how hard the hit lands, and Try it fires one canned reaction without you touching the device. Anti-Theft sits further down the same tab and does an unrelated job. That volume scaling is worth a thought if the device lives near your head. The US National Institute on Deafness and Other Communication Disorders [puts the safe ceiling at 70 A-weighted decibels](https://www.nidcd.nih.gov/health/noise-induced-hearing-loss). Below that, even long exposure is unlikely to cause hearing loss. At or above 85 dBA, long or repeated exposure can cause it. Consider leaving Harder slaps play louder switched off, or dropping the system volume, before an afternoon of thumping the desk. | What you do | What Slap does | | --- | --- | | Tap the desk your phone sits on | Plays a clip from your active sound pack and adds one to your count | | Shake the phone | The same reaction, at the sensitivity you set | | Knock the Mac chassis, or the desk under it | Plays a clip, the notch face can recoil, and the hit counts for your rank | | Plug the charger in, or pull it out | Plays a clip from your plug pack, on its own switch | | Attach or detach a USB device on a Mac | Counts as a plug event, debounced for 1.5 seconds | | Press Try it in the Android app | Fires one canned reaction without you hitting anything | Everything wrapped around the sound is optional. Turn the extras off and you still get the clip. #### The sound packs **Sound packs** are the themed clip libraries Slap plays from. You download whichever ones interest you. > Figure: **OpenAlly's guide figure lifting one pack cover from a tiered mint rack of sound pack covers** Pick a library first, then go and hit the desk. The pack you switch on decides what plays back. (light: https://openally.ai/images/blog/miro/miro-slap-packs.webp; dark: https://openally.ai/images/blog/miro/miro-slap-packs-dark.webp) The library is split into India, International, Games and Adult sections. A random pack picks a different clip on every impact. **Escalating packs** are ordered instead, so the reaction gets worse the more you hit. Desi Pain, for example, runs mild to wild. - **India.** Desi Pain is the flagship escalation pack, described as ordered mild to wild desi protest. Moye Moye and Just Looking Like a Wow play at random. - **International.** Ow! Classic escalates through its ows. Vine Boom, Bruh, Emotional Damage, Metal Pipe, Oof, Augh, Bonk, Airhorn, Fart with Reverb and Sad Violin all fire at random. - **Games.** Halo Deaths, which is exactly what it sounds like. - **Adult.** Its own category chip, filtered out of the other three, and nothing from it reaches your device unless you go and download it. > Figure: **Sounds tab on Android with pack covers, Get buttons, and All India International Games Adult chips** The live Sounds tab, where each cover carries its own Get button. Adult is one chip in the category row. (light: https://openally.ai/images/blog/ui/slap-android-packs.webp; dark: https://openally.ai/images/blog/ui/slap-android-packs-dark.webp) The [Slap feature page](https://openally.ai/features/apps/slap) shows the live catalogue. Every cover, regional tab and pack mode is visible there. #### Plug sounds and extras **Plug sounds** are the trigger that needs no impact at all. A charger going in or coming out plays a clip from your assigned plug pack, and on a Mac a USB device attaching or detaching counts as well. > Figure: **OpenAlly's guide figure plugging a charging cable into a phone as mint sound marks the connection** A plug reaction is its own trigger. It does not depend on slap detection being on. (light: https://openally.ai/images/blog/miro/miro-slap-plug.webp; dark: https://openally.ai/images/blog/miro/miro-slap-plug-dark.webp) The switch is separate. Plug reactions keep working whether or not slap detection is enabled. Every event is debounced for 1.5 seconds, which stops a fumbled cable turning your desk into a soundboard. The demonstration on the feature page uses Ow! Classic. > Figure: **Effects tab on Android showing screen shake, impact flash, haptic buzz, and Plug sounds off** Shake, flash and haptics live on Effects. Plug sounds is an extra on this tab and is switched off. (light: https://openally.ai/images/blog/ui/slap-android-plug.webp; dark: https://openally.ai/images/blog/ui/slap-android-plug-dark.webp) That Effects tab is also where the optional theatre lives: screen shake, impact flash and haptic buzz, each with its own strength. Slap counter can hide the card without stopping the count. Plug sounds sits under Extras on the same surface. #### Slap on Mac Slap on a Mac wants a real hit on the chassis, or on the desk the Mac sits on, and answers it with a clip from the active pack. The notch face can recoil, and the hit counts towards your rank under the Mac filter. > Figure: **OpenAlly's guide figure touching the top edge of an open laptop lid as a small face in the notch reacts with wide eyes** A chassis or desk impact is what the Mac build listens for. A soft trackpad tap is not enough. (light: https://openally.ai/images/blog/miro/miro-slap-mac.webp; dark: https://openally.ai/images/blog/miro/miro-slap-mac-dark.webp) A Try Now control fires the reaction without you touching the machine. Whether you hit a laptop you paid for is between you and the laptop. The extras land a little differently on desktop, where a lid creak joins the optional screen shake and confetti on MacBooks. The [companion face](https://openally.ai/features/apps/companion-face) is optional on every platform that has it. Slap runs perfectly well with the face never activated, and switching it on adds a recoil while the meme clip takes over the mouth. The Android build is on [Google Play](https://play.google.com/store/apps/details?id=openally.ai). The Mac build is a disk image from [the Mac download page](https://openally.ai/download/mac), which asks for Apple silicon and macOS 14 or later. #### The leaderboard **The leaderboard** is login-gated, so you sign in to appear on it. Ranks are filtered by All, Android, iOS and Mac. > Figure: **OpenAlly's guide figure holding a small trophy beside a ranked list with tiny avatars** The board is a rank, not a security log. Phone and Mac report to the same list. (light: https://openally.ai/images/blog/miro/miro-slap-leaderboard.webp; dark: https://openally.ai/images/blog/miro/miro-slap-leaderboard-dark.webp) Your AI is the avatar beside your name, and every row carries a streak tag. Phone and Mac report to the same board. Signing in on both leaves you with a single rank rather than two separate ones. > [!IMPORTANT] The board on the feature page is a sample, not usage data > The names, tiers and totals displayed on the public Slap page exist to demonstrate the layout. They are not usage statistics, and no number on that page reports how many people run Slap or how forcefully they hit. Slap opens as its own screen rather than living inside a chat, which is how the focused apps in the catalogue work. The [FoneClaw comparison](https://openally.ai/blog/openally-vs-foneclaw) covers that collection, and the [download page](https://openally.ai/download) has every platform. About the author: [Matterward Labs](https://matterwardlabs.com) makes OpenAlly. The pack names, modes and section titles come from OpenAlly's own catalogue and the public [Slap feature page](https://openally.ai/features/apps/slap). The screen labels quoted here, such as React to slaps, Fresh Hands and Harder slaps play louder, are read off the app's own Slap, Sounds and Effects tabs in the captures above. All three surfaces were checked the same day (31 August 2026, reviewed by Matterward Labs on that date). The listening levels are the NIDCD's, not ours. If something here reads wrong, send a note to the company [contact page](https://matterwardlabs.com/contact). #### Declared sources - [OpenAlly on Google Play](https://play.google.com/store/apps/details?id=openally.ai) - [NIDCD, Noise-Induced Hearing Loss](https://www.nidcd.nih.gov/health/noise-induced-hearing-loss) - [Matterward Labs](https://matterwardlabs.com) - [Matterward Labs contact page](https://matterwardlabs.com/contact) ### AI on your phone in 2026: what runs offline, and what it takes to get real work done What offline AI on a phone can and cannot do in 2026, what small local models are good at, and why acting on your accounts is the much harder half. Canonical: https://openally.ai/blog/ai-on-your-phone-offline Markdown: https://openally.ai/blog/ai-on-your-phone-offline.md Published: 2026-08-27 Updated: 2026-09-11 Reviewed: 2026-08-27 Author: Matterward Labs (https://matterwardlabs.com) Category: Guide Keywords: run AI on your phone; offline AI app; local AI on phone; offline AI iPhone; local LLM Android; AI that works without internet; on-device AI models Audience: People who want AI that works without internet; Android and iPhone users comparing offline AI apps; Small business owners deciding what to run on a phone Disclosure: Matterward Labs makes OpenAlly. Third-party capabilities in this article come from each project's own public documentation and repository, checked on 27 August 2026. Industry figures come from the primary sources linked throughout. A phone in 2026 can run a real language model with the aeroplane mode switch on. That part is settled. What happens next, when you want the thing to do something other than talk back, is where it gets harder. #### Short answer A modern phone runs a small language model offline, and it is genuinely useful for private text work: drafting, summarising, sorting, transcribing. Making that same model **act on your accounts**, so it can answer a customer, raise an invoice or move a booking, is a separate and much harder layer. The binding limit is the space the model has left to think in once you have described everything it is allowed to do. > [!IMPORTANT] Where these claims come from > Third-party capabilities are taken from each project's own documentation and repository, read on 27 August 2026. OpenAlly device figures come from our own dated field notes on [running Gemma 4 on Android](https://openally.ai/blog/gemma-4-android-litert-lm). Product capabilities come from the [OpenAlly feature catalogue](https://openally.ai/features). > Interface view: **Two useful lanes, one deliberate boundary.** Private text work can remain on the phone. Reading or changing a live account is a separate route that needs a connection and access you granted. #### What a phone can actually run today Offline AI on a phone means the model file sits in the phone's storage and the arithmetic happens on the phone's own chip. No request leaves the handset, so it works on a plane, in a basement, and on a prepaid data plan that ran out on the 20th. The size band runs from a few hundred megabytes to a few gigabytes. At the small end, Google's AI Edge team documents [Gemma 3 1B at 529MB, reading up to 2,585 tokens per second on a mobile GPU](https://developers.googleblog.com/google-ai-edge-small-language-models-multimodality-rag-function-calling/), which it describes as processing about a page of content in under a second. That figure is the speed of reading a prompt rather than the speed of writing an answer, and the two are far apart on a phone. At the larger end, the two on-device packages in OpenAlly's Android catalogue weigh 2.59GB and 3.66GB, behind device-memory floors of 6GiB and 8GiB (about 6.4GB and 8.6GB), as recorded in our [Gemma 4 field notes](https://openally.ai/blog/gemma-4-android-litert-lm). Storage is the easy half. Installed memory is not available memory: in one load attempt in the same field notes, a 12GB handset reported roughly 5.2GB actually free, and the load failed under memory pressure. That is one observation on one device, not a minimum-RAM rule, and it is why the floors sit so far above the file sizes. Running a model and letting it call something are also two different pieces of software, shipped on different timetables. Google's on-device function calling library, the piece that lets a local model decide to invoke something, ships as its own component with its [own documentation](https://developers.googleblog.com/google-ai-edge-small-language-models-multimodality-rag-function-calling/). Hardware matters too. MLC LLM, the engine behind one of the better-known local chat apps, states in its [Android deployment documentation](https://llm.mlc.ai/docs/deploy/android.html) that it needs an actual mobile GPU, and names the Samsung S23 with Snapdragon 8 Gen 2 as the handset its demo build targets. That is a note about a reference build and not a published minimum specification, but it is a fair signal of what these engines are tuned against. Assume a recent phone, and treat an older or cheaper one as something to test. > Interface view: **A model file can fit before the runtime does.** Published download sizes sit beside one OpenAlly field-note observation: installed memory, actually free memory and supported mobile hardware are different constraints. #### What a local model is genuinely good at [PocketPal AI](https://github.com/a-ghorbani/pocketpal-ai) runs on Android, iOS and iPadOS, downloads models from Hugging Face, benchmarks them, and works entirely offline once a model is on the device. MLC Chat does the same job on a different engine, compiling models ahead of time for the phone's GPU. For one class of work they are the right tool: - **Drafting.** A reply, a description, a first pass at a message you will edit anyway. - **Summarising.** A long note, a transcript, a document you do not want to paste into a website. - **Sorting and labelling.** Deciding which of five buckets a piece of text belongs in. - **Transcription.** Turning a recording into text without the audio leaving the handset. OpenAlly ships this as the [Whisper tool](https://openally.ai/features/tools/whisper) on Android, and the [Local LLM tool](https://openally.ai/features/tools/local-llm) covers the text side on Android and iPhone. Every one of those jobs has the same shape. Text goes in, text comes out, and nothing outside the phone needs to change as a result. That shape is what small models are good at, and privacy here is structural, because the request genuinely has nowhere else to go. > Interface view: **The local sweet spot is a closed text loop.** A note or voice memo becomes a draft, summary, labels or transcript without an account changing elsewhere; OpenAlly Whisper transcription is Android-specific. #### Where local tool use stops Several local models can use tools already. PocketPal's repository documents a tool-use loop where the model calls built-in engines mid-conversation: `calculate` for arithmetic, `datetime` for the current date and time, and `render_html` for displaying generated markup. Each of those runs inside the conversation. One evaluates an expression, one reads the clock, one draws markup the model has just written. None of them opens a door to something that belongs to you. The jobs that give a person hours back are the other kind. Look at the customer message that came in overnight. Check whether that item is in stock. Send the payment link. Move the two o'clock appointment and tell the customer it moved. Each of those has to reach a real account, read the current state of something, and change it. That is a different category of work, and self-contained tools do not reach it. If what you want is a private assistant that drafts and summarises with no account and no connection, a good local chat app is the correct answer, and you can stop here with the right tool already chosen. > Interface view: **A self-contained tool stops at the account boundary.** PocketPal's documented calculate, datetime and render_html tools stay inside the conversation. Messages, stock, payments and bookings need another layer. #### Why doing is harder than chatting Before a model can choose an action, it has to be told what actions exist. Each one arrives as a schema: a name, a description, the arguments it takes, the allowed values. The model reads all of that before it decides anything. On a phone, those descriptions compete for the same small context window as your actual message. In our own testing on 21 July 2026, described in the [Gemma 4 field notes](https://openally.ai/blog/gemma-4-android-litert-lm), the schemas for fifteen tools alone were estimated at **2,825 tokens**. Against that build's 4,096-token window, the menu took 69% of the room (2,825 of 4,096 tokens) before the conversation started. > Interface view: **A 4,096-token window before the conversation starts.** The measured Gemma 4 field-note budget: 2,825 estimated tokens for fifteen tool schemas, leaving the conversation, instructions, history, and answer to share what remains. The squeeze is not ours alone, and it does not ease at scale. Four groups have published measurements on large tool catalogues, from [Microsoft Research](https://www.microsoft.com/en-us/research/blog/tool-space-interference-in-the-mcp-era-designing-for-agent-compatibility-at-scale/), [GitHub](https://github.blog/ai-and-ml/github-copilot/how-were-making-github-copilot-smarter-with-fewer-tools/), [Anthropic](https://www.anthropic.com/engineering/advanced-tool-use) and a [chance-corrected study from Meta](https://arxiv.org/abs/2605.24660): | Finding | Reported as | Source | | --- | --- | --- | | Large tool spaces can lower model performance for some models | Up to 85% lower | Microsoft Research, 11 September 2025 | | Public servers already publish catalogues far past what a small window holds | 256 tools in the largest one catalogued | Microsoft Research, 11 September 2025 | | Tools across those servers collide on names, so the model cannot tell them apart by name alone | 775 tools carrying colliding names | Microsoft Research, 11 September 2025 | | A 16,000-token model runs out of window once enough tools are described to it at once | 93 tools | Microsoft Research, 11 September 2025 | | Cutting a default toolset improved task success and cut latency | 40 tools down to 13, worth 2 to 5 percentage points on SWE-bench Verified and SWE-Lancer with GPT-5 and Sonnet 4.5, and roughly 400ms | GitHub, 19 November 2025 | | Selecting tools by relevance beat handing over a fixed list | 94.5% tool-use coverage against 69.0% | GitHub, 19 November 2025 | | Loading tool descriptions on demand cut token use and raised tool-use accuracy | 85% fewer tokens; accuracy 49% to 74% on Opus 4 and 79.5% to 88.1% on Opus 4.5 | Anthropic, 24 November 2025 | | Showing roughly seven tools matched the coverage of a fixed list of fifty | 90.3% against 90.8% | Meta preprint, 23 May 2026 | | Choosing about two tools per question instead of a fixed five raised how often Claude Sonnet 4.6 picked the right tool when it was among those shown | 87.1% to 93.1% | Meta preprint, 23 May 2026 | The results point the same way. Past a certain number, extra options stop paying for themselves: they cost window, they cost latency, and on some models they cost accuracy. Each team reached the same fix: work out which few actions this particular request needs, describe only those, and leave the rest out of the conversation. That fix is cheap on a server, where you can afford a second model call to do the working out. On a phone, with a small window and a battery, you have to do it without one. That was the part of building a phone agent that took us longest. #### What that means for your work Split the jobs you want done into the two categories above. **Jobs a local model handles well on its own.** Draft the reply. Summarise the meeting. Tidy the note. Turn the voice memo into text. These stay on the phone, cost nothing per use, and work with no signal. **Jobs that need to reach your accounts.** These are the four from earlier: the customer thread, the stock check, the payment link, the booking. Each needs an account connection you have granted, and a model chosen for the job rather than for the phone it runs on. They return hours where the first group returns minutes. That is the story in [how an AI agent runs a small business](https://openally.ai/blog/ai-agent-runs-small-business). Most people want both, which is why the sensible design leaves the choice with you. OpenAlly keeps the private text work on the device and lets you route the rest to a provider you already trust, a plan you already pay for, a server you run, or the OpenAlly service. The [trust and control guide](https://openally.ai/under-the-hood) sets out where each route sends a request. "It runs locally" and "nothing ever leaves the phone" are different promises. The first is about where the model runs. The second is about every other request the app makes, and it is much harder to keep. > Interface view: **Choose the route that fits this job.** An OpenAlly interface view of the current processing choices: on-device work, a provider key, a supported subscription, your own compatible server, or OpenAlly Cloud. #### Choosing what to run where Match the job to the tool. | What you want | What to use | | --- | --- | | Private drafting and summarising, no account, no connection | A dedicated local chat app such as PocketPal AI or MLC Chat, which is simpler than anything else here | | The same, plus transcription and document work, inside a wider app | OpenAlly's [Local LLM tool](https://openally.ai/features/tools/local-llm), with [Whisper](https://openally.ai/features/tools/whisper) transcription on Android, and the option to switch to a larger model when a job needs it | | Anything that reads or changes a real account, order, invoice or booking | An agent with permissions you granted, such as OpenAlly, and a model chosen per job | | Work that starts on a phone and continues on a laptop | A product that treats both as the same body of work | > Interface view: **Start with the phone; continue with more room.** A source-aligned interface view of OpenAlly continuity: the same account, agents, and chat history are available on mobile and Mac, while Mac remains optional. Write down whatever must keep working with no signal. That short list decides your local model. Everything else can afford a network request. #### Frequently asked questions > Interface view: **Five checks before you choose an offline model.** The FAQ in one field card: platform support, working-memory headroom, privacy scope, tool reach, and the storage, battery and time costs of running locally. ##### Does offline AI work on iPhone as well as Android? Both platforms run local models. [PocketPal AI](https://github.com/a-ghorbani/pocketpal-ai) documents support for Android, iOS and iPadOS, and OpenAlly's [Local LLM tool](https://openally.ai/features/tools/local-llm) is built for both. Where OpenAlly differs today is transcription: the [Whisper tool](https://openally.ai/features/tools/whisper) is an Android capability. The [download page](https://openally.ai/download) lists where OpenAlly stands on each platform. ##### How much memory does my phone need? More than the model file. Google's mobile-optimised [Gemma 3 1B is a 529MB download](https://developers.googleblog.com/google-ai-edge-small-language-models-multimodality-rag-function-calling/), and a running model needs working space on top of that beside the operating system and everything else open. In our own testing a 12GB handset had roughly 5.2GB genuinely free when a model tried to load. OpenAlly's device-memory floors of 6GiB and 8GiB are safety gates we chose, not hardware requirements anyone has published. ##### Is a local model automatically private? The model request is, because it has nowhere to go. An app around it can still send other things: analytics, crash reports, backups, or a fallback request to a service when the local model cannot handle a job. Read what the app says it sends. The word "local" does not tell you. OpenAlly's [privacy policy](https://openally.ai/privacy) lists what leaves the device and when. ##### Can a local model actually do things, or only chat? Some can call tools, and the tools tend to be self-contained ones such as PocketPal's `calculate` and `datetime`. Reaching your messages, orders or calendar is a different layer that needs granted permissions and enough context space to describe the available actions. The published evidence, from [Microsoft Research](https://www.microsoft.com/en-us/research/blog/tool-space-interference-in-the-mcp-era-designing-for-agent-compatibility-at-scale/) through [Anthropic](https://www.anthropic.com/engineering/advanced-tool-use), [GitHub](https://github.blog/ai-and-ml/github-copilot/how-were-making-github-copilot-smarter-with-fewer-tools/) and [Meta's chance-corrected study](https://arxiv.org/abs/2605.24660), is that the answer is to show the model a small, relevant set of actions rather than all of them. ##### Does running a model locally cost anything? The download is one-off and the arithmetic is free after that. The cost shows up as storage, battery and speed. In our own [June 2026 field test on a Galaxy S26 Ultra](https://openally.ai/blog/gemma-4-android-litert-lm), the first response took roughly 10 to 15 seconds. That is a fair trade for private work and a poor one when a customer is waiting. #### Declared sources - [Google AI Edge: small language models with multimodality, RAG and function calling](https://developers.googleblog.com/google-ai-edge-small-language-models-multimodality-rag-function-calling/) - [PocketPal AI repository](https://github.com/a-ghorbani/pocketpal-ai) - [MLC LLM Android deployment documentation](https://llm.mlc.ai/docs/deploy/android.html) - [Anthropic engineering: advanced tool use](https://www.anthropic.com/engineering/advanced-tool-use) - [Microsoft Research: tool-space interference in the MCP era](https://www.microsoft.com/en-us/research/blog/tool-space-interference-in-the-mcp-era-designing-for-agent-compatibility-at-scale/) - [GitHub: making Copilot smarter with fewer tools](https://github.blog/ai-and-ml/github-copilot/how-were-making-github-copilot-smarter-with-fewer-tools/) - [How Many Tools Should an LLM Agent See? A Chance-Corrected Answer (Meta Platforms, preprint)](https://arxiv.org/abs/2605.24660) ### OpenClaw on Android: Can an AI Agent Run on Your Phone Alone? OpenClaw's Android app connects to another host that runs the agent. Here is what phone AI agents need, and how OpenAlly keeps the agent on the phone. Canonical: https://openally.ai/blog/ai-agent-on-your-phone-without-a-pc Markdown: https://openally.ai/blog/ai-agent-on-your-phone-without-a-pc.md Published: 2026-08-25 Updated: 2026-10-06 Reviewed: 2026-08-25 Author: Matterward Labs (https://matterwardlabs.com) Category: Guide Keywords: OpenClaw on Android; AI agent on your phone; phone AI agent without a PC; agent that runs on the device; Android AI agent without a server Audience: Android users who want an AI agent that can act; People trying to run OpenClaw on a phone; Anyone comparing phone AI agents; Individuals and businesses choosing a mobile AI agent Disclosure: Matterward Labs makes OpenAlly. OpenClaw's Android design and setup come from its documentation, reviewed on 25 August 2026, with the requirements line quoted word for word. Its foundation history comes from OpenClaw and its founder; the mobile launch date comes from TechCrunch. Every source is linked. In OpenClaw's official mobile design, the Android app is a client. The agent and its working state live on a second host you have to keep available: a laptop, desktop, home server or hosted server. That is a deliberate design. It decides what still has to be switched on when you leave the desk. > [!IMPORTANT] How this comparison was made > Matterward Labs makes OpenAlly. OpenClaw's documented Android and remote-host requirements were reviewed on 25 August 2026, and every outside fact below links to its source. The diagrams explain those public documents; they are not screenshots of OpenClaw. #### Short answer **OpenClaw's official Android app needs another host.** Its documentation lists an OpenClaw Gateway on macOS, Linux or Windows as a requirement. The phone is a client; the other host runs the agent and owns its state. OpenClaw's own remote guide says that host can be a laptop, desktop, home server or hosted server. Where the AI model runs is a separate choice. **OpenAlly runs the agent and keeps its working records on the phone itself.** There is no second OpenAlly host to install, pair or keep reachable. The product spans Android and iPhone with [Mac](https://openally.ai/download/mac) continuity on the same OpenAlly account, while the [download page](https://openally.ai/download) remains the source of truth for current platform availability. A cloud model or connected service still needs that service and an internet connection. Both designs are deliberate, and the rest of this article explains what each one buys you. #### What a phone agent has to do before it can help you An agent is not a single thing. Underneath, several different jobs have to happen, and products place them in different locations. **The interface** listens, shows, taps and types. It is the part you can see on the phone, but seeing it there does not tell you where the rest lives. **The agent** keeps the task moving. It decides which record, tool or model to use next, checks what happened, and carries the result into the following step. **The AI model** reasons over each request and proposes an answer or next action. It can run on the phone, on a computer you own, or at a provider you choose. That location is a separate decision from where the agent runs. **The memory** holds what the work depends on: your history, notes, customers, files and other records. Every product in this category has AI. Ask instead where the interface, agent, model and memory live, because those four answers determine when the product works and what it can reach. > Interface view: **The parts behind one chat box.** A location map separating the visible phone interface from the agent, memory and AI model route. #### What OpenClaw's Android app is built to do [OpenClaw](https://openclaw.ai/blog/introducing-openclaw-foundation/) is an open-source project in this category. Its creator [joined OpenAI in February 2026](https://steipete.me/posts/2026/openclaw), and the project is now stewarded by an independent non-profit foundation. Android and iPhone apps [arrived on 30 June 2026](https://techcrunch.com/2026/06/30/openclaw-is-finally-available-on-android-and-ios/). The Android app is capable. Reading OpenClaw's own [Android documentation](https://docs.openclaw.ai/platforms/android), it handles chat with full history, camera photos and video, voice notes and a continuous talk mode, questions about the phone's own status and content, read-only file browsing, approval prompts, and a Wear OS companion. Messages you send while out of signal are saved on the phone first and delivered later. What it does not do is run your agent on Android. The documentation calls the app a companion and tells you to **"run it on macOS, Linux, or Windows via WSL2"**. OpenClaw's [remote-host guide](https://docs.openclaw.ai/gateway/remote) is more precise: the other host owns sessions, channels and state, and it runs the agent. The phone connects as a client. The AI model may still be on that host or at a provider elsewhere. That arrangement has real advantages. An agent on an always-available host can reach that host's files and software, avoids making a phone battery the centre of the system, and keeps working after you put the phone down. The official remote guide explicitly covers persistent hosted servers, home servers, desktops and laptops. If you already maintain one of those hosts, keeping the agent there is a sensible way to stay close to the files and software you want it to use. It is one reasonable design among several. #### The setup that design asks for The consequence shows up during installation. Following OpenClaw's documented steps, you install and start the OpenClaw Gateway on the chosen host, then open the Connect tab on the phone, approve the pairing from that host's command line, and check the connection there again. None of that is hidden. It is all in the documentation, and it gives the operator direct control over pairing. It also asks more than installing a phone app by itself. The second consequence matters more day to day. OpenClaw documents remote access to a home server, hosted server, desktop or laptop, so the host does not have to be in your house. The dependency remains: if the chosen host is asleep or unreachable, the Android client does not take over the agent's work. > Interface view: **Setup shows which host owns the agent.** OpenClaw's documented Android path starts, pairs and verifies a Gateway on another host; OpenAlly keeps the agent and working records on the phone. #### Why people try to run it on the phone anyway Community projects show that people do want the whole OpenClaw installation on an Android phone. One documented route installs OpenClaw through Termux or a standalone community app. Its author reports a 3–10 minute automated installation and includes guidance for keeping the work available under Android's battery management. That is a legitimate route for people who want OpenClaw itself on the handset and are comfortable relying on a community-maintained package. It is different from the official companion app, so support, updates and device constraints are part of the choice. The same project's [documentation](https://github.com/AidanPark/openclaw-android) says local AI inference is technically functional, then lists the practical limits: free memory, model storage and CPU speed. It recommends a cloud provider for production use. That is why “the agent is on the phone” and “the AI model is on the phone” are separate questions. #### What changes when the agent runs on the phone itself OpenAlly starts from the other end. The agent and its working records sit on the device in your hand, and it works whenever you, a schedule or an enabled event trigger it. So the dependency gets smaller. There is no second OpenAlly host to install, no host pairing to redo and no home network to return to. The [download page](https://openally.ai/download) shows current availability across Android, iPhone and Mac. > [!IMPORTANT] What OpenAlly still depends on > No second OpenAlly host is required for the agent and its working records. A cloud AI route, connected account, messaging channel, web search or cross-device feature still needs an internet connection and the service you chose. You also sign in to an OpenAlly account when you set the app up. The [privacy page](https://openally.ai/privacy) lists those paths precisely. Because the agent lives with your records rather than beside them, a customer message can become a quote, an order that reserves stock, a payment link and a numbered invoice without you stitching the steps together. The [practical guide to running a small business this way](https://openally.ai/blog/ai-agent-runs-small-business) follows that journey end to end, and the [feature catalogue](https://openally.ai/features) shows what you can put to work today. A computer still helps when you want one. Sign in to the [Mac app](https://openally.ai/download/mac) with the same account and your agents and chat history continue there, where there is more room to review what each agent is allowed to do. The Mac extends the work rather than hosting the phone's agent. Close the laptop and the agent on your phone carries on with whatever does not depend on an online service. > Interface view: **The computer is an option, not a requirement.** A phone work queue connects to a Mac review surface while the phone remains the agent host. On the specific job of driving the phone's screen for you, OpenAlly keeps that separate. Screen control, calls and messages come through [Aster](https://github.com/satyajiit/aster-mcp), an optional open-source companion you install on the same Android phone and can stop on its own. It is a second app rather than a second machine, and it is scoped to that Android device rather than to the whole product. You grant Android screen access to Aster, separately from everything else OpenAlly does. > Interface view: **Screen control stays a separate choice.** An approved Android screen task moving from OpenAlly to the optional Aster companion. #### Where the model runs is a separate question "The agent runs on your phone" and "the model inference also happens on your phone" are different claims. OpenAlly makes the first one. The second depends on a supported route you choose. Phone agents often split those two locations: the interface sits on the handset while the model request travels over the internet to a provider. That is common across the category. The interface, agent and working memory are on your device. The AI model can be wherever you choose to put it: bring your own provider key, sign in with a subscription you already pay for, point OpenAlly at your own server, or pick a model that runs directly on the phone. Our [engineering notes on running a model on an Android phone](https://openally.ai/blog/gemma-4-android-litert-lm) cover what that last route can and cannot do today, measured rather than promised. You sign in to an OpenAlly account when you set the app up. The two change independently. You can switch which model answers you without rebuilding any of your work, and the [trust and control guide](https://openally.ai/under-the-hood) sets out where your information sits in each case. > Interface view: **Change the route, not the work.** On-device, provider-key, subscription, self-hosted and OpenAlly service routes connect to the same agent work. #### Which setup fits you | What you want | The better fit | Why | | --- | --- | --- | | An agent that works on a bus, in a shop, or away from home | OpenAlly | The agent is on the phone, so there is no second OpenAlly host to keep reachable. Online models and services still need a connection. | | An agent that reaches the files and software on another host | OpenClaw | Its documented design puts the agent on that host, which is where those things already are. | | To run customer records, stock, orders, invoices and payments | OpenAlly | Those records are part of the product rather than apps an agent has to drive. | | To self-host the agent and contribute to its code | OpenClaw | It is open source with an independent foundation behind it. | | To avoid pairing your phone with another agent host | OpenAlly | The agent and working records are already on the phone. | | To move between supported phone and desktop surfaces | Both | OpenAlly continues the signed-in account, agents and chat history from phone to Mac. OpenClaw connects its clients to the same agent host. | If you already run a home server or hosted server and want the agent close to it, OpenClaw's approach is the point. If your phone is where your work actually happens, a second agent host may be a dependency you do not need. #### What has to stay reachable When the agent stops working, what turned off? With OpenAlly, the core answer is the phone. The agent and its working records are on the device you are carrying. A model provider, connected service or cross-device feature may still be online, but there is no second OpenAlly host running the agent. With OpenClaw, the answer includes the chosen host: a laptop, desktop, home server or hosted server. That is the right trade for people who want an open-source agent close to that host's files and software. If you would rather not maintain a second agent host, [start with OpenAlly](https://openally.ai/download). #### Frequently asked questions ##### Do I need a computer to use an AI agent on my phone? It depends on the product. OpenClaw's Android documentation lists an OpenClaw Gateway on macOS, Linux or Windows as required, so the official app needs another reachable host. That host can be your computer, a home server or a hosted server. OpenAlly runs its agent on the phone, so there is no second OpenAlly host to pair or keep reachable. ##### Does OpenAlly work on iPhone as well as Android? OpenAlly's product scope includes iPhone alongside Android and Mac. The [download page](https://openally.ai/download) is the source of truth for current platform availability. Android screen control is the exception: it needs the separate Aster companion, so it stays scoped to Android. ##### Does an AI agent on a phone work without internet? Partly. The agent is on the device, but offline work also needs an on-device model and on-device data or tools. Cloud models, web search, connected accounts, messaging channels and sync still need a connection. The [on-device model field notes](https://openally.ai/blog/gemma-4-android-litert-lm) show the current limits of the fully local route. ##### Is my information leaving the phone? Messages and knowledge stay on the device. Business records and the notes you write in the Notes app are kept in your OpenAlly account, so the same records and notes appear on your other devices. Feature data may leave when a request needs an outside service: for example an AI provider, connected account, messaging channel, search service or sync feature. Separately, diagnostics, analytics, and push delivery have limited data paths that depend on the platform and your settings. The [privacy page](https://openally.ai/privacy) lists every recipient and what each one gets, and the [trust and control guide](https://openally.ai/under-the-hood) explains the AI routes. ##### Is OpenAlly related to OpenClaw? They are separate products with different aims and ownership. This article compares the shape of their mobile designs rather than claiming a formal relationship: OpenClaw centres on an open-source agent you host, and OpenAlly centres on personal and business work running on the device you carry. #### Declared sources - [OpenClaw Android platform documentation](https://docs.openclaw.ai/platforms/android) - [OpenClaw remote-host documentation](https://docs.openclaw.ai/gateway/remote) - [OpenClaw Foundation announcement](https://openclaw.ai/blog/introducing-openclaw-foundation/) - [OpenClaw founder's OpenAI announcement](https://steipete.me/posts/2026/openclaw) - [OpenClaw mobile launch coverage](https://techcrunch.com/2026/06/30/openclaw-is-finally-available-on-android-and-ios/) - [Aster open-source Android companion](https://github.com/satyajiit/aster-mcp) - [A community project putting OpenClaw on the phone](https://github.com/AidanPark/openclaw-android) ### OpenAlly vs CellClaw: Which Android AI Agent Does More? Compare OpenAlly and CellClaw across Android control, privacy, AI models, business workflows and platforms, and see why OpenAlly is the stronger choice. Canonical: https://openally.ai/blog/openally-vs-cellclaw Markdown: https://openally.ai/blog/openally-vs-cellclaw.md Published: 2026-08-20 Updated: 2026-10-06 Reviewed: 2026-08-25 Author: Matterward Labs (https://matterwardlabs.com) Category: Comparison Keywords: OpenAlly vs CellClaw; CellClaw alternative; AI agent for Android; CellClaw vs OpenAlly; private AI assistant Audience: People comparing Android AI agents; Small businesses choosing an AI assistant; Android users who want phone automation; People comparing private and on-device AI choices Disclosure: Matterward Labs makes OpenAlly. CellClaw facts in this comparison come from CellClaw's official public pages, with the evidence and review date shown so readers can check the claims themselves. **OpenAlly is the stronger overall choice** for most people and businesses comparing Android AI agents. CellClaw has a sharp strength: it is built to operate one Android phone through chat, an overlay, notifications, a Quick Settings tile, voice, and device tools. OpenAlly can also put Android screen actions to work through Aster, then goes much further: personal knowledge, customer conversations, products, stock, orders, bookings, payment links, invoices, websites, apps, broader AI choice, iPhone, and Mac. The two products have different centres of gravity. **CellClaw is built to operate an Android phone; OpenAlly is built to operate the work around it.** If you want the broader agent system, OpenAlly wins. > [!IMPORTANT] A fair comparison > Matterward Labs makes OpenAlly. We reviewed CellClaw's official website, privacy policy, and linked public source on 20 August 2026, and rechecked them on 25 August 2026. CellClaw receives full credit here for the Android capabilities and privacy controls its own pages document. #### Who each product suits Choose **OpenAlly** when your goal extends beyond opening apps and tapping through screens. It connects conversations to personal knowledge and durable work: a customer, a product, a booking, an order, a payment, an invoice, a website, or an app. That work can continue across Android, iPhone, and Mac, while Android screen control remains an optional capability through Aster. Choose **CellClaw** when the job begins and ends on one Android phone and you value direct access from a floating overlay, notification reply, Quick Settings, or wake word. CellClaw's [official product page](https://cellclaw.com/) describes an autonomous Android agent with SMS, calls, contacts, calendar, camera, files, sensors, notifications, screen reading, and app control. That is a capable phone agent, and a narrower product definition than OpenAlly's. OpenAlly has another practical advantage: its Android app is a normal Google Play install from the [download page](https://openally.ai/download). CellClaw's official site still tells readers to clone and build it from GitHub, and both the repository it links and the licence file linked beside it returned a public 404 when we rechecked on 25 August 2026. That site's footer now points readers who would rather skip the key setup to CellBreak, described on [CellClaw's own page](https://cellclaw.com/) as giving “the full CellClaw experience with optional managed credits”, with no keys required. CellBreak's own site did not return a readable page to us during this review, so this article repeats only what CellClaw publishes about it. None of this is our main argument, but it matters to somebody deciding what they can install and use today. > Interface view: **From controlling the phone to completing the work.** Android actions sit inside a wider OpenAlly workspace for knowledge, customers, sales, payments, creation, and cross-device continuity. #### Two products, two finish lines ##### CellClaw finishes at the screen CellClaw gives the Android phone an agent that can see the screen, select an action, and keep working until the phone task is complete. Its published feature list covers SMS, calls, email, contacts, calendars, app launching, screen capture, cross-app messaging, files, clipboard, camera, location, browser actions, notifications, schedules, and memory. It also offers five useful ways to reach the agent: the main chat, a floating overlay, a notification reply, a Quick Settings tile, and a wake word. For someone who wants fast access to device control without changing surfaces, this is CellClaw's strongest case. ##### OpenAlly finishes at the completed work OpenAlly's [feature catalogue](https://openally.ai/features) begins with what a person or business is trying to finish. The agent can work over personal knowledge, build a site or web app, manage a catalogue and stock, remember customers, coordinate bookings, create orders, send payment links, issue invoices, answer customers on a [WhatsApp Business](https://openally.ai/product/whatsapp-business) number through Meta, and use focused tools for useful or fun projects. Android control belongs inside that larger system rather than bounding it. A screen action can help obtain information or operate an app, while OpenAlly keeps the resulting work connected to the records, approvals, and next steps that matter afterwards. #### Seven jobs and who finishes them A raw tool count is a weak way to compare agents. A tool can tap a button without knowing what business event the tap should create. These seven jobs expose more of the product. | Job | OpenAlly | CellClaw | Stronger fit | | --- | --- | --- | --- | | 1. Hold a useful conversation and remember context | Personal agents work with conversations, files, knowledge, tasks, and the apps assigned to them. | Stores conversations and memory facts locally and recalls them across sessions. | **OpenAlly** for a broader personal workspace; both provide persistent context. | | 2. Operate an Android app | Sends approved Android screen tasks to the separate Aster companion. | Makes screen reading, taps, typing, swipes, app launching, overlays, and notification actions central to the Android app. | **CellClaw** for direct one-app convenience; **OpenAlly** for the clearer permission separation. | | 3. Turn a customer message into a sale | Connects supported conversations to catalogue, stock, customer, order, payment, and invoice work. | Can open, read, and reply in messaging apps; its public material does not document a connected sales system of record. | **OpenAlly**. | | 4. Run everyday business operations | Includes customer, product, inventory, booking, order, payment, and invoice workflows, plus WhatsApp Business on the business's own number. | Publishes phone tools for messages, email, contacts, calendar, files, and app operation. | **OpenAlly**. | | 5. Choose how AI is provided | Offers a broad catalogue across your own provider keys, supported subscriptions, your own server, on-device choices, and OpenAlly Cloud. | Publishes four provider routes: Anthropic, OpenAI, Google Gemini, and OpenRouter, plus a footer pointer to CellBreak for “optional managed credits” with no keys required. | **OpenAlly** for provider and deployment breadth. | | 6. Continue the work on a computer | The same account, agents, and chat history continue from Android or iPhone to the [Mac app](https://openally.ai/download/mac), where you choose the model per agent and assign the skills and tools each agent may use. | CellClaw's reviewed public product material documents Android. | **OpenAlly**. | | 7. Extend and repeat the work | Agents, skills, apps, channels, schedules, and automations share the same working context. | Supports scheduled tasks, a long-running heartbeat, and reusable Markdown skills triggered by keywords. | **OpenAlly** for breadth; CellClaw remains useful for repeated phone tasks. | Across the seven jobs, CellClaw is most compelling when direct Android operation is the destination. OpenAlly is stronger when Android operation is one step in a larger piece of personal or business work. > Interface view: **Seven jobs, two different centres.** CellClaw concentrates on the Android handset. OpenAlly carries the task through knowledge, business records, AI choices, customer outcomes, and other devices. #### One customer message shows what “more” really means What follows is a worked example rather than a measurement: one message, traced through both products using only what each one's public material documents. Consider a small clothing shop. A customer sends a WhatsApp message: **“Do you have the blue linen shirt in medium? I can pay now if it arrives tomorrow.”** CellClaw documents the phone-facing parts of this job, as its pages read on the 25 August 2026 recheck. It can open a messaging app, read the screen, reply, search the web, use files, and work with notifications. If the shop already has separate stock, order, and payment apps, CellClaw can help navigate the Android interfaces it is allowed to use. OpenAlly connects the conversation to the sale itself: | Customer journey | OpenAlly action | What remains connected | | --- | --- | --- | | The customer asks for a size and colour | Checks the product variant, current stock, price, tax, and fulfilment information | Customer enquiry and accurate catalogue context | | The item is available | Prepares a stock-aware quote and delivery choice | Product, variant, quantity, and quoted total | | The customer confirms | Creates the order and reserves the relevant stock | Customer, order, and inventory movement | | The customer wants to pay | Creates the payment link or QR and sends it through the approved channel | Order and payment request | | The payment lands | Reconciles the payment against the order | Payment state and order state | | The sale completes | Prepares the numbered invoice or receipt for review and sharing | Customer history, order, payment, and invoice | The [Sales skill](https://openally.ai/features/skills/sales), [Business app](https://openally.ai/features/apps/business), [Payments app](https://openally.ai/features/apps/payments), and [Invoicing skill](https://openally.ai/features/skills/invoicing) make this a connected journey rather than six unrelated phone actions. CellClaw can help operate the phone where the sale is discussed. OpenAlly can operate the work that turns the discussion into a paid, recorded customer outcome. For a deeper walkthrough covering discovery, messages, booking, payments, reviews, and repeat customers, read [how an AI agent runs a small business](https://openally.ai/blog/ai-agent-runs-small-business). #### Android screen control: one app or a separate permission boundary Both products can help with Android screen actions, but they place that power differently. CellClaw puts screen reading, taps, typing, swipes, a floating overlay, notification access, messaging, and other phone capabilities inside its Android product. Its site says Accessibility operates separately for stability. It also documents three autonomy profiles, per-app allow and block lists, encrypted credential storage, and optional biometric approval for sensitive operations. OpenAlly uses [Aster](https://github.com/satyajiit/aster-mcp) as an optional, separately installed Android companion. Aster receives the Android screen permissions needed for App Automations; the main OpenAlly app does not take overlay or notification-listener access. Aster has its own approval, stop, and connection controls. The separation lets someone use OpenAlly's agents, knowledge, business apps, channels, and model choices without automatically placing screen-level Android access inside the main app. When screen control is useful, the user adds and approves Aster for that specific Android job. CellClaw's all-in-one approach is convenient. OpenAlly's separation gives the stronger default permission posture. How a package is distributed changes what it is permitted to do at all: [OpenAlly vs FoneClaw](https://openally.ai/blog/openally-vs-foneclaw) works through a case where a sideloaded build and a Google Play edition carry different permission boundaries. > Interface view: **Android screen power remains a separate choice.** The main OpenAlly app sends an approved Android screen task to Aster, which keeps screen-level access inside the optional companion. #### Three meanings of the word “local” “Local” can describe where an app runs, where records are stored, or where an AI model processes a prompt. Those are different questions. ##### How CellClaw describes the path CellClaw's [privacy policy](https://cellclaw.com/privacy) says its conversations, memory, preferences, and cached content are stored on the Android device. It says CellClaw does not operate backend servers or include analytics, advertising, or tracking services. Those are meaningful privacy strengths. The same policy says prompts and relevant context are sent directly from the device to the chosen AI provider, and that each provider applies its own data practices. CellClaw's official page names Anthropic, OpenAI, Google Gemini, and OpenRouter. Its explicitly local AI feature is wake-word detection; the reviewed public material does not document an on-device generative model for normal agent conversations. So “no CellClaw server” is accurate, while “every AI request remains on the phone” would not match the documented provider path. ##### How OpenAlly separates the choices OpenAlly keeps working messages and knowledge on the device, and business records in your OpenAlly account. The processing path then follows the option selected for the request: - A supported on-device model processes the model request on the device. - Your own provider key sends the request directly to that provider. - Your own server sends the request to the server you control. - OpenAlly Cloud processes requests sent through that service. - A connected service receives the information needed for the action the user asked it to perform. The [trust and control guide](https://openally.ai/under-the-hood) explains these choices in plain language, while the [privacy policy](https://openally.ai/privacy) carries the detailed disclosures. OpenAlly's advantage is the ability to choose among more processing paths without rebuilding the agent, business records, or workflow around a single provider. The field notes on [Gemma 4 and on-device Android AI](https://openally.ai/blog/gemma-4-android-litert-lm) also explain the practical difference between running a local conversational model and asking a small local model to coordinate many tools. > Interface view: **Local records and clearly chosen AI routes.** Both products keep important state on the phone. OpenAlly adds a wider set of processing choices, including on-device models, provider keys, a user-controlled server, and OpenAlly Cloud. #### Five reasons to choose CellClaw anyway A useful comparison should preserve the case for the competitor, so here is CellClaw's, in five areas: - **Immediate Android access.** The overlay, notification reply, Quick Settings tile, wake word, and main chat put the agent close to whatever is happening on the phone. - **Broad phone hardware access.** Its public list includes calls, SMS, contacts, calendar, camera, location, sensors, files, clipboard, and screen actions. - **Local working state.** Conversations, memory, and schedules stay on the device according to its privacy policy. - **Direct provider communication.** Model requests go from the phone to the selected provider rather than through CellClaw infrastructure. - **Visible autonomy controls.** Three profiles, per-app lists, optional biometric checks, and approval choices help users decide how freely the agent acts. For an Android enthusiast who primarily wants a phone operator, these are substantial strengths. CellClaw does not need to become a business platform to be good at that job. #### What OpenAlly adds after the tap OpenAlly is stronger overall because it joins capabilities that CellClaw presents mainly as phone actions. Each line below names the public catalogue page it comes from, as those pages read when we rechecked them on 25 August 2026. - **Agents that hold the work, not just the reply.** [Personal and business agents](https://openally.ai/features/agents) operate over conversations, files, knowledge and tasks, and each agent is given its own role, knowledge, model, and access to the [apps](https://openally.ai/features/apps) and [skills](https://openally.ai/features/skills) it may use. - **A customer message that ends as a record.** The [Business app](https://openally.ai/features/apps/business), [Sales skill](https://openally.ai/features/skills/sales), [Payments app](https://openally.ai/features/apps/payments) and [Invoicing skill](https://openally.ai/features/skills/invoicing) carry one conversation through catalogue, stock, order, payment link and invoice. - **Screen automation kept outside the main app.** The [App Automations skill](https://openally.ai/features/skills/app-automations) turns a familiar Android app flow into a repeatable shortcut, with your review at logins, payments, and other sensitive steps, while the screen-level permissions stay inside the separately installed [Aster companion](https://github.com/satyajiit/aster-mcp). - **A wider AI choice than one provider list.** OpenAlly's public catalogue holds 46 current provider entries, spanning your own keys, supported subscriptions, your own server, supported on-device choices and OpenAlly Cloud, with the route selectable per agent; the [trust and control guide](https://openally.ai/under-the-hood) sets out how that choice is made. [CellClaw's official page](https://cellclaw.com/) listed four provider routes when this comparison was checked on 25 August 2026: Anthropic, OpenAI, Google Gemini and OpenRouter. - **Tools for making, alongside tools for operating.** [App Builder](https://openally.ai/features/apps/app-builder) takes a described website or mini-app through guided edits and an on-device preview before you publish it, and [Vibe Pages](https://openally.ai/features/apps/vibe-pages) turns a mood into a shareable page with visuals, type, and music. The rest of the [app catalogue](https://openally.ai/features/apps) covers the useful and the playful alike. - **Work that outlives the handset.** The account, agents and chat history that start on Android or iPhone continue on the [Mac app](https://openally.ai/download/mac), where you connect your own provider keys, choose the model per agent or run one locally, and assign the skills and tools each agent may use. Install options for every device sit on the [download page](https://openally.ai/download). What that buys is a wider unit of completion. CellClaw can finish a sequence on an Android screen. OpenAlly can carry the intention before that screen action and the records, payment, creation, or follow-up that come after it. Making does not stop at a web page, either. [Web to App](https://openally.ai/product/web-to-app) turns a site you run, including one made in App Builder, into a signed Android app, built on an Android phone or a Mac. You install it, share it, or upload it to your own Google Play listing. #### What we checked, and what we could not We reviewed CellClaw's official homepage and privacy policy on 20 August 2026 and rechecked both on 25 August 2026. We used its own published descriptions for Android tools, interaction methods, AI providers, local storage, security controls, skills, schedules, and privacy. We treated the advertised “33 tools” as CellClaw's own marketing count rather than an independently tested benchmark. The same rule applies to CellBreak: we could not verify that product's own published material to the standard used everywhere else here, so nothing in this article should be read as a description of what CellBreak actually is. The homepage links to an [official CellClaw GitHub repository](https://github.com/jordanthejet/cellclaw). That URL, and the licence file linked beside it, returned a public 404 when we rechecked on 25 August 2026, so source history, releases, tests, licence files, contributor activity, and an install package could not be independently assessed from that repository. We do not treat unavailable public material as proof that a capability does not exist. Where CellClaw's pages do not describe a business system, cross-platform app, or local generative model, this article says exactly that: **CellClaw's public material does not document it.** OpenAlly statements were checked against the current public [feature catalogue](https://openally.ai/features), [trust and control guide](https://openally.ai/under-the-hood), [privacy policy](https://openally.ai/privacy), and [download options](https://openally.ai/download), plus Aster's public source. The comparison uses jobs and customer outcomes rather than treating differently named tools as equal units. The WhatsApp Business, Premium and Web to App lines were added on 10 September 2026 from their product pages: [WhatsApp Business](https://openally.ai/product/whatsapp-business), [Premium](https://openally.ai/premium) and [Web to App](https://openally.ai/product/web-to-app). #### Frequently asked questions ##### Is OpenAlly a CellClaw alternative? Yes. OpenAlly is the stronger CellClaw alternative when you want Android screen work plus personal knowledge, business records, customer conversations, sales, payments, invoices, creation, broader AI choices, iPhone, and Mac. CellClaw remains a good fit when direct control of one Android phone is the primary requirement. ##### Can OpenAlly control Android apps like CellClaw? Yes. OpenAlly uses Aster, its separate open-source Android companion, for App Automations such as reading an approved screen and performing approved taps, typing, swipes, or navigation. This Android-specific capability remains separate from the main OpenAlly app. ##### Is CellClaw fully offline? CellClaw's public material documents local app state and an on-device wake-word detector. Its privacy policy also says prompts and relevant context go directly to the selected Anthropic, OpenAI, Google Gemini, or OpenRouter service. The reviewed public pages do not document a local generative model for normal CellClaw conversations. ##### Which supports more AI providers and model routes? OpenAlly provides the broader set of routes: your own provider keys, supported subscriptions, your own server, supported on-device choices, and OpenAlly Cloud. CellClaw publicly lists Anthropic, OpenAI, Google Gemini, and OpenRouter, with OpenRouter providing access to models available through that service. ##### Which is better for a small business? OpenAlly is stronger for a small business because customer conversations can connect to products, inventory, bookings, orders, payment links, reconciliation, and invoices. CellClaw can operate Android business apps, but its reviewed public material does not document a built-in business system of record. On the business's own number, [WhatsApp Business](https://openally.ai/product/whatsapp-business) connects through Meta and is included with [Premium](https://openally.ai/premium). Meta bills its message charges to the business directly, and each billable AI message waits for the owner's approval unless it fits a budget the owner set. ##### Does OpenAlly put a floating overlay or notification listener on Android? No. The main OpenAlly app does not take overlay or notification-listener access, and that is a deliberate trade: those permissions stay inside the optional Aster companion, which you install and approve separately when you want Android screen control. CellClaw is genuinely faster to reach from whatever is already on screen. OpenAlly is reached from the app itself, from an approved channel such as a WhatsApp or Telegram conversation, or from a hook or schedule that lets an agent act on an incoming message or event without anyone opening anything. ##### Does CellClaw work on iPhone or Mac? CellClaw's reviewed public material documents an Android product. It does not document an official iPhone or Mac app. OpenAlly spans Android, iPhone, and Mac: the account, agents, and chat history that start on a phone continue on the [Mac app](https://openally.ai/download/mac), which is a direct download from this site rather than a store listing, and channels and agent hooks can act on incoming messages and events there too. Aster and App Automations stay scoped to Android screen control. ##### Is CellClaw open source? CellClaw's website describes the project as Apache 2.0 open source. At the time of this review, the GitHub URL linked by that website, including its licence file link, returned 404 publicly, so we could not independently inspect the source or licence file. This article preserves CellClaw's published claim and separately states what was verifiable on the review date. #### The verdict CellClaw is a capable Android phone agent. It offers convenient interaction surfaces, deep device actions, local working state, direct provider communication, and useful autonomy controls. If the goal is “operate this Android phone,” CellClaw deserves consideration. OpenAlly wins the broader comparison. It can bring Android screen control into a task through Aster, while also handling the knowledge, customer relationship, inventory, booking, order, payment, invoice, website, app, and cross-device work surrounding that task. Its permission separation and wider model choices strengthen that advantage. For anyone who wants more than phone control, that is the stronger Android AI agent. Explore [OpenAlly's features](https://openally.ai/features), meet its [personal and business agents](https://openally.ai/features/agents), or choose the current install for your device on the [download page](https://openally.ai/download). #### Declared sources - [CellClaw official website](https://cellclaw.com/) - [CellClaw privacy policy](https://cellclaw.com/privacy) - [CellClaw official GitHub link](https://github.com/jordanthejet/cellclaw) - [Aster open-source Android companion](https://github.com/satyajiit/aster-mcp) ### From Discovery to Repeat Customer: How an AI Agent Runs a Small Business See how an AI agent connects WhatsApp Business, bookings, customer records, inventory, payments, invoices, reviews, and weekly insights for a small business. Canonical: https://openally.ai/blog/ai-agent-runs-small-business Markdown: https://openally.ai/blog/ai-agent-runs-small-business.md Published: 2026-08-19 Updated: 2026-10-06 Reviewed: 2026-08-19 Author: Matterward Labs (https://matterwardlabs.com) Category: Guide Keywords: AI agent for small business; WhatsApp Business automation; AI review replies for small business; salon booking automation; small business workflow automation; AI inventory and invoicing Audience: Small-business owners; Retailers selling through messaging; Salon and service-business owners; Teams comparing AI business agents Disclosure: Matterward Labs makes OpenAlly. Market figures are linked to their original publishers, and the time-savings ledger shows its workload inputs and arithmetic. A customer discovers a salon on Google, checks its reviews, and sends a WhatsApp message asking whether there is a slot after work. A few minutes later, the appointment is booked. After the visit, the customer pays, receives an invoice, and gets a useful follow-up. For the customer, that is one smooth journey. For a small-business owner, it can mean copying information between search, messaging, a diary, customer notes, payments, invoices, stock sheets, and review tools. An **AI agent for small business** becomes valuable when it connects those steps. OpenAlly™ brings [WhatsApp Business](https://openally.ai/product/whatsapp-business) and Telegram conversations, customer records, products and services, inventory, bookings, orders, payments, invoices, review replies, and weekly insights into one working loop. > [!IMPORTANT] How this guide is grounded > Product capabilities come from the [OpenAlly feature catalogue](https://openally.ai/features). Industry figures come from the official sources linked in the article. The salon time ledger shows every workload input and calculation used in its daily, weekly, and monthly totals. #### Short answer An AI agent runs a small business when it connects the whole customer journey instead of answering one message at a time. That means the customer who found you on Google, the WhatsApp Business or Telegram conversation that follows, customer records, catalogue and stock, bookings and orders, payments and invoices, reviews, and a weekly view of what to do next, all holding on to the same context. The owner still sets prices and policies, approves public replies, and decides how money moves. What that returns depends on the workload. In the six-day salon ledger below, which handles 15 repeated enquiries, 10 booking events, and 8 completed visits per day, OpenAlly returns 1 hour 40 minutes per open day, 10 hours 1 minute per week, and 43 hours 24 minutes per month. Every workload input and every calculation behind those totals is shown, so you can substitute your own numbers and rerun the arithmetic. #### What is an AI agent for small business? A chatbot answers a question. An automation follows a predefined trigger. An AI agent can understand the request, use the business information it has permission to access, prepare or complete the next step, and keep the resulting record connected to the rest of the business. Answering “Yes, we have a slot at 5:30” is only the beginning. The useful work is checking the right staff member, reserving the slot, identifying the customer, sending the confirmation, collecting payment, preparing the invoice, and knowing what should happen next. | Tool | What it usually does | Where the work stops | | --- | --- | --- | | Chatbot | Answers questions from a prompt or knowledge source | The owner still transfers the answer into bookings, orders, payments, and records | | Rule-based automation | Runs a fixed action after a trigger | Exceptions and cross-tool context usually require manual handling | | AI business agent | Understands intent, uses approved business records, and moves the work forward | The owner keeps control of policies, exceptions, public replies, and consequential money actions | The market is moving quickly in this direction. Meta says more than one million businesses use Meta Business Agent across WhatsApp and Messenger, while its messaging products carry more than one billion active business conversations each day. Its agent can answer questions, recommend catalogue items, book appointments, qualify leads, hand conversations to staff, and close sales. [Meta's announcement](https://about.fb.com/news/2026/06/meta-business-agent/) describes customer messaging as an operating surface rather than an inbox. Adoption does not automatically solve fragmentation. The [QuickBooks Small Business Insights survey](https://quickbooks.intuit.com/r/small-business-data/small-business-insights/) polls approximately 5,000 small and midsize businesses in the United States, Canada, the United Kingdom, and Australia every quarter. In its July 2026 wave, 80% said they now use AI regularly. Yet 36% still reported difficulty with the lack of integration between the different digital tools and systems they use. The next advantage is therefore connected work rather than one more isolated AI feature. #### The complete discovery-to-repeat-customer workflow A useful small-business agent follows the customer journey rather than the boundaries between apps. | Stage | What the customer experiences | What OpenAlly keeps connected | | --- | --- | --- | | Discovery | Finds accurate information, posts, offers, photos, and reviews on Google | The hours, services, and prices the owner keeps current, plus review replies and listing updates drafted for the owner to publish | | Conversation | Asks a question on WhatsApp or Telegram | Customer intent, conversation context, products, services, prices, and availability | | Customer record | Does not need to repeat the same details on every visit | Contact details, preferences, history, bookings, and orders | | Decision | Receives an accurate quote, product option, or appointment slot | Catalogue, inventory, tax, staff, rooms, equipment, and booking rules | | Transaction | Chooses payment, payment link, QR, or cash on delivery | Order or booking, payment state, reconciliation, invoice, and receipt | | Follow-up | Receives useful aftercare and can leave feedback | Review drafts, follow-up tasks, unresolved issues, and repeat-customer context | | Improvement | Gets a more reliable experience next time | Weekly performance, common questions, stock movement, bookings, and suggested actions | > Interface view: **The business board behind a simple customer message.** An OpenAlly canvas connecting a Google search, WhatsApp, customer, booking, payment, invoice, and review UI. ##### 1. Turn Google discovery into a trustworthy first impression For many local businesses, the first conversation begins before anybody sends a message. It begins with the business hours, phone number, photos, offers, reviews, and appointment links shown on Google. OpenAlly works alongside that listing rather than inside it. When hours change for a holiday, or you share a new review with your agent, it drafts the updated notice, a post, or a reply in the business's voice from the records it already holds. You publish it on Google yourself. Google itself is making Business Profile a more conversational surface for owners. Its official [Business Profile and Gemini guidance](https://support.google.com/business/answer/17142585?co=GENIE.Platform%3DDesktop&hl=en) covers updating hours and attributes, managing reviews and posts, and examining impressions, website clicks, directions, bookings, and search keywords. The opportunity for a business owner is practical: keep the storefront people see in search aligned with the business they actually reach. ##### 2. Continue the journey on WhatsApp or Telegram A customer should not need to learn a new system just to ask whether a product is available or a haircut can be booked. OpenAlly connects customer conversations to the [business workspace](https://openally.ai/features/apps/business). On the business's own number, [WhatsApp Business](https://openally.ai/product/whatsapp-business) connects through Meta and is included with [Premium](https://openally.ai/premium). A Telegram bot works too, and a personal WhatsApp account can be linked by QR code, though that route is unofficial and can lead to account limits. The agent can answer from approved catalogue or service information, check current availability, qualify the request, and prepare the next step without turning every enquiry into a copy-and-paste exercise. The behaviour is different for each business: - A retailer can move from a product question to a stock-aware quote, order, payment choice, and invoice. - A salon can move from a service question to an available stylist, appointment, payment, receipt, and follow-up. - A consultant can move from an enquiry to a qualified lead, booked session, payment request, and client record. - A repair business can move from an issue description to a booking, job record, quote, and status update. Meta's India launch says eligible small businesses can use Business AI on WhatsApp to answer questions, recommend products, capture leads, take bookings, and make sales. Meta also cites a 2025 Kantar study in which 91% of online adults in India reported chatting with a business every week. [Meta's India announcement](https://about.fb.com/news/2026/05/introducing-business-ai-on-whatsapp-for-small-businesses-in-india/) shows why messaging belongs inside the operating loop. ##### 3. Build customer memory without spreadsheet archaeology A conversation becomes more valuable when it improves the next interaction. OpenAlly can create and update customer records as work moves forward. A returning customer can be recognised with the right history and preferences: the service previously booked, an unfinished order, a preferred appointment time, an unpaid invoice, or an issue that needs follow-up. This is the useful core of customer relationship management for a small team. The owner does not need an enormous CRM implementation to stop treating every returning customer like a stranger. A customer record can connect: - Contact and channel details - Enquiries and preferences - Quotes, orders, and bookings - Payment and invoice history - Follow-up tasks - Review or service-recovery context Approved business context survives the end of a chat. ##### 4. Turn intent into a booking or an order The agent should not confirm a product that is unavailable or book a stylist who is already busy. For a product business, the [Sales skill](https://openally.ai/features/skills/sales) can use the catalogue, variants, prices, taxes, and current inventory to prepare a quote. When the customer confirms, the workflow can create the order and reserve the relevant stock. For a service business, it can check staff, rooms, equipment, service duration, working hours, and blackouts before proposing a slot. Once the customer confirms, the booking becomes part of the same customer history. | Business | Information checked | Customer outcome | | --- | --- | --- | | Retail shop | Product, variant, price, tax, stock, and delivery or collection choice | Accurate quote and stock-aware order | | Salon | Service, stylist, duration, availability, room or chair, and customer preference | Confirmed appointment without double-booking | | Consultant | Service type, qualification details, calendar, duration, and payment policy | Qualified lead and booked session | | Repair service | Issue type, technician, parts or service category, slot, and quote | Job booking with clear next steps | The conversation produces a usable business record, which is what separates a reply from an operation. ##### 5. Connect payments, invoices, and reconciliation A sale is not finished because a payment link was sent. The [Payments app](https://openally.ai/features/apps/payments) can create a payment link or QR, record cash on delivery where appropriate, track the transaction, and reconcile the payment when it lands. The [Invoicing skill](https://openally.ai/features/skills/invoicing) can then prepare a numbered invoice or receipt for review and sharing. The customer gets one coherent experience. The owner gets a clearer trail from the original request to the order or booking, payment state, and invoice. Money-sensitive actions remain explicit. The agent can prepare the work, while the owner retains control over actions such as capturing or refunding money. ##### 6. Turn reviews and weekly insights into the next action Reviews are a stream of customer language about what works, what disappoints, and what people repeatedly ask for. When a new Google review arrives, share it with your agent and it prepares a reply in the business's voice. You post it on Google yourself, so public replies stay under the owner's control. The weekly view then connects more than reviews. It can help an owner examine: - Hours, services, and prices the listing should reflect - Common questions from WhatsApp Business or Telegram - Booking demand and unused capacity - Orders, payment state, and unresolved invoices - Products or supplies selling fastest - Customers who need a follow-up - Reviews requiring attention - Recommended actions for the next week The owner still makes the business decision. The agent reduces the work required to see the decision clearly. #### WhatsApp Business automation that continues after the reply The easiest mistake is to measure messaging automation by response speed alone. A fast but disconnected reply can still create more work. If an agent says a slot is free without checking the diary, quotes an old price, or confirms stock that has already sold, the owner has merely automated the apology. OpenAlly's approach connects the message to the records that determine whether the answer is useful. | Customer message | Connected information | Action the agent can prepare | | --- | --- | --- | | “Is the 200 ml bottle available?” | Product variant, current stock, price, and tax | Stock-aware quote or order | | “Can I book a haircut after 5?” | Service duration, stylist availability, and opening hours | Valid appointment choices | | “Can you send the payment link again?” | Customer, booking or order, and payment status | Correct payment link and updated record | | “Where is my invoice?” | Order, payment, invoice number, and delivery channel | Invoice preview or resend | | “I came last month. Can I book the same stylist?” | Customer history and previous booking | Relevant slot options with continuity | | “Are you open on the holiday?” | The holiday blackout the owner set in the booking setup | Accurate response, plus a drafted listing update for the owner to publish | What makes the reply useful is the catalogue, stock, booking, payment, and invoice record sitting behind it. On [WhatsApp Business](https://openally.ai/product/whatsapp-business) the agent works inside Meta's rules. Inside the 24-hour window that opens when a customer writes, it replies in its own words; outside it, only an approved template goes out. A billable AI message waits for the owner's approval, or stays inside a budget the owner set, and AI replies pause for a conversation the moment the owner taps Take over. The agent runs on the owner's designated phone or Mac, which checks for new messages about every 30 seconds while OpenAlly is running on it. > Interface view: **A WhatsApp conversation that closes the sale.** The agent checks stock and tax, quotes the customer, creates the order after confirmation, sends the payment link, reconciles the UPI payment, and leaves the numbered invoice ready to share. #### Reviews and listing updates the owner publishes A Google listing is often updated after a holiday has already begun. OpenAlly does not connect to Google Business Profile, so it never changes the listing itself: it drafts the change from the records it holds, and the owner publishes it. An OpenAlly agent can help an owner: - Keep staff, room and equipment availability current in the booking setup, including a one-off blackout for a holiday - Draft a post or offer from the [catalogue and current stock](https://openally.ai/features/apps/business) - Draft replies to reviews the owner shares, in the business's voice - Spot questions customers keep asking on [WhatsApp Business](https://openally.ai/product/whatsapp-business) that the listing should answer - Collect review replies, listing drafts, and recommended actions into a weekly summary on a schedule the owner sets That creates a loop between **what customers discover**, **what they ask**, **what they buy or book**, and **what they say afterwards**. #### What OpenAlly can automate and what remains in your hands Small-business automation works best when the control boundary is obvious. | Area | OpenAlly can handle | The owner or team decides | | --- | --- | --- | | Reviews and listing | Draft review replies, listing updates, and posts from current records | Everything published on Google, public tone, offers, and escalation | | WhatsApp Business and Telegram | Answer approved questions, gather details, check records, and prepare the next step, with billable WhatsApp messages waiting for approval | Complex conversations, complaints, exceptions, and relationship-sensitive moments | | Customer records | Create or update useful context from the workflow | Corrections, duplicate resolution, retention choices, and access | | Catalogue and inventory | Check products, variants, prices, taxes, and stock | Pricing policy, discounts, purchasing, substitutions, and stock corrections | | Bookings and orders | Propose valid slots, quote, create confirmed records, and reserve relevant capacity or stock | Capacity rules, cancellations, exceptional discounts, and fulfilment | | Payments and invoices | Create links, track status, reconcile, issue invoices, and prepare receipts | Provider policy, tax and legal review, captures, refunds, and disputes | | Reviews and insights | Draft replies to reviews you share, and assemble weekly summaries | The final public response and the business action that follows | The goal is a business in which routine coordination no longer consumes the owner's most useful hours, not an unattended one. #### How much time can an AI agent save a small business? Time savings have to start from a workload. The following ledger follows a six-day salon handling 15 repeated enquiries, 10 booking events, and 8 completed visits per day. “OpenAlly-assisted time” includes the time used to read, approve, correct, or handle exceptions. ##### Daily recurring work | Repeated task | Workload | Manual time | OpenAlly-assisted time | Time returned | | --- | --- | --- | --- | --- | | Repeated service, price, hours, and availability questions | 15 enquiries per day | 52.5 minutes | 15 minutes | 37.5 minutes | | Booking, rescheduling, and confirmation work | 10 booking events per day | 40 minutes | 10 minutes | 30 minutes | | Customer record, payment, and receipt administration | 8 completed visits per day | 24 minutes | 8 minutes | 16 minutes | | **Daily recurring total** | **Stated workload above** | **116.5 minutes** | **33 minutes** | **83.5 minutes** | ##### Weekly administration | Weekly task | Manual time | OpenAlly-assisted time | Time returned | | --- | --- | --- | --- | | Review replies, listing updates, and post drafts | 75 minutes | 25 minutes | 50 minutes | | Service catalogue and supply or inventory review | 45 minutes | 20 minutes | 25 minutes | | Weekly performance summary and follow-up list | 40 minutes | 15 minutes | 25 minutes | | **Weekly administration total** | **160 minutes** | **60 minutes** | **100 minutes** | With six operating days per week, the arithmetic is: | Period | Calculation | Time returned | | --- | --- | --- | | Average open day | 601 weekly minutes divided across 6 open days | **1 hour 40 minutes** | | Week | 83.5 daily minutes multiplied by 6, plus 100 weekly minutes | **10 hours 1 minute** | | Month | 601 weekly minutes multiplied by 52 weeks and divided by 12 months | **43 hours 24 minutes** | > Interface view: **From repeated admin to time returned.** The six-day salon workload returns 1 hour 40 minutes per open day, 10 hours 1 minute per week, and 43 hours 24 minutes per month. > [!NOTE] Recalculate with your own workload > Enquiry volume, service complexity, data quality, approval rules, staff habits, and exceptions change the total. Track one week of work, replace each workload and time value above, and run the same arithmetic for your business. The [QuickBooks Small Business Insights survey](https://quickbooks.intuit.com/r/small-business-data/small-business-insights/) gives the wider context. In its July 2026 wave, 64% of respondents said greater efficiency was the leading benefit of the digital tools they use, and three-quarters of those using AI said it was boosting their productivity. Working hours split both ways in the same wave: 24% said their workdays are shorter thanks to AI, against 12% who said AI has made their workdays longer. Those figures describe the survey population, not this ledger. The salon totals above come directly from the displayed workload and arithmetic. #### A practical seven-day setup for a small business A small pilot is more informative than trying to automate the whole company on day one. 1. Choose one measurable customer journey, such as enquiry to confirmed appointment. 2. Install OpenAlly from the [download page](https://openally.ai/download), then connect the customer messaging channel used most often, such as [WhatsApp Business](https://openally.ai/product/whatsapp-business) on the business number. 3. Add the services, products, prices, taxes, stock, staff, rooms, equipment, and business hours required for that journey. 4. Define which questions the agent can answer and which situations must reach a person. 5. Set approval rules for public review replies, discounts, cancellations, payments, and refunds. 6. Run real examples with staff before opening the workflow to normal customer traffic. 7. Compare one week of enquiry time, booking errors, missed follow-ups, payment administration, and owner interruptions. Do not begin with “How much can AI do?” Begin with “Which repeated customer journey should become easier?” #### Frequently asked questions ##### What is the best AI agent for a small business? The best fit depends on the work it can connect. Look beyond chat quality and check whether the agent can work with the business's actual customer records, catalogue, inventory, bookings, orders, payments, invoices, reviews, and reporting while preserving clear owner control. OpenAlly is designed around that connected operating loop, with customer channels and business records working together across Android, iPhone, and Mac. Most of the day is run from a phone between customers, and the same account, agents, and chat history continue on a Mac when the work moves to a desk. ##### Can an AI agent automate WhatsApp Business enquiries? Yes. An AI agent can answer common questions, collect customer details, check approved product or service information, propose valid booking options, create orders, prepare payment links, and pass complex cases to a person. Test whether the reply stays connected to the resulting customer, booking, order, payment, and invoice. With [WhatsApp Business](https://openally.ai/product/whatsapp-business), included with [Premium](https://openally.ai/premium), the agent answers on the business's own number through Meta: replies inside the 24-hour window, approved templates outside it, the owner's approval for billable messages, and Take over whenever a person should step in. Meta bills its message charges to the business directly. ##### Can OpenAlly reply to Google reviews? OpenAlly does not connect to Google Business Profile. Share a review with your agent and it drafts a reply in the business's voice; it can also draft holiday-hours notices and posts from your records. You publish them on Google. ##### Can OpenAlly automate salon bookings? Yes. A salon workflow can use services, durations, staff or resource availability, working hours, blackouts, and customer preferences to offer valid appointment options and create the confirmed booking. ##### Can the same agent manage products and service bookings? Yes. A business can use catalogue and inventory records for product sales while using availability and resources for services. A salon, for example, may book a haircut and also sell a care product through the same customer relationship. ##### Can OpenAlly create payment links and invoices? OpenAlly can create payment links and orders through a connected provider, track and reconcile payments, and issue numbered invoices or receipts. Consequential money actions such as capture and refund remain owner-controlled. ##### Does a small-business AI agent replace staff? It should replace repetitive coordination, not human judgment. Staff remain important for exceptions, complaints, relationship-sensitive conversations, service delivery, pricing decisions, and the moments where empathy matters more than speed. ##### How much time can OpenAlly save? The answer depends on volume and process. In this guide's worked salon example (six open days a week, with 15 repeated enquiries, 10 booking events and 8 completed visits per day, at the per-task minutes the ledger lists), OpenAlly returns 1 hour 40 minutes per open day, 10 hours 1 minute per week, and 43 hours 24 minutes per month. Put a week of your own workload into the same ledger to get your figure. ##### Can I run the day from my phone and continue on a Mac? Yes, and that is the ordinary pattern for a small business: answer enquiries on a phone between customers, then sit down to the paperwork on a larger screen. Your account, agents, and chat history continue from your phone to your Mac, so the conversation you were in the middle of is already there. On the Mac you connect your own AI providers and choose the model for each agent, or run a model locally. You assign skills and tools per agent and review what each one is allowed to do, and you connect channels and agent hooks so agents act on incoming messages and events. OpenAlly for Mac comes as a direct download from the [Mac download page](https://openally.ai/download/mac), not through an app store. ##### Where does OpenAlly store business information? Business records such as products and stock, customers, orders, invoices and bookings are kept in your OpenAlly account, so your phone, your Mac and the people you invite work from the same records. When a selected AI provider or connected service is needed, the request follows the route chosen by the user. WhatsApp Business messages arrive through a dedicated OpenAlly delivery service, which keeps each one encrypted until the device collects it, for seven days at most, and never runs the agent. The [OpenAlly trust and control guide](https://openally.ai/under-the-hood) explains model choice, connected-service boundaries, permissions, and working data. #### Akhilesh's salon: a connected working day Akhilesh runs a neighbourhood salon with two stylists, one treatment room, a compact retail shelf, and a steady stream of WhatsApp enquiries. Customers usually discover the salon through Google, then ask about prices, availability, directions, products, and last-minute slots. His day with OpenAlly looks like this: | Moment | Customer or business event | How OpenAlly helps | | --- | --- | --- | | Before opening | Holiday hours changed and two new Google reviews came in | Drafts the holiday notice and two review replies for Akhilesh to post | | Morning | A new customer asks on WhatsApp for a haircut after 5 p.m. | Checks the service duration and stylist availability, then offers valid slots | | Late morning | The customer chooses 5:30 p.m. | Creates the customer record and booking, then sends the confirmation | | Afternoon | A returning customer asks for the same stylist and treatment as last time | Uses the customer history to prepare relevant appointment options | | During service | A stylist sells a care product used in the treatment | Connects the product, stock movement, customer, payment, and receipt | | Checkout | The customer wants to pay digitally | Prepares the payment link or QR and reconciles the resulting payment | | End of day | Akhilesh wants to know what still needs attention | Summarises unanswered enquiries, upcoming bookings, payment exceptions, and follow-ups | | End of week | Reviews, bookings, stock, and customer questions have accumulated | Produces a digest with the useful numbers and recommended actions | > Interface view: **Akhilesh's connected salon day.** OpenAlly app UI cards connect a Google search, a WhatsApp booking, customer memory, service delivery, payment, invoice, and weekly insight. At this workload, OpenAlly returns **10 hours 1 minute in a week** and **43 hours 24 minutes in an average month**. Akhilesh still chooses the prices, handles unusual requests, speaks to unhappy customers, approves public replies, and decides how money moves. The agent keeps routine information and next steps from becoming a second job. Before OpenAlly, every customer message left Akhilesh with three more tasks: update the diary, remember the customer, and check whether payment or follow-up happened. With the routine path connected, he spends less of the day stitching tools together and more of it running the salon. For broader product comparisons, read [OpenAlly vs CellClaw](https://openally.ai/blog/openally-vs-cellclaw) and [OpenAlly vs FoneClaw](https://openally.ai/blog/openally-vs-foneclaw). #### Declared sources - [Meta Business Agent announcement](https://about.fb.com/news/2026/06/meta-business-agent/) - [Meta Business AI on WhatsApp in India](https://about.fb.com/news/2026/05/introducing-business-ai-on-whatsapp-for-small-businesses-in-india/) - [Google Business Profile and Gemini help](https://support.google.com/business/answer/17142585?co=GENIE.Platform%3DDesktop&hl=en) - [QuickBooks Small Business Insights](https://quickbooks.intuit.com/r/small-business-data/small-business-insights/) ### Gemma 4 on Android with LiteRT-LM: The Model Wasn't the Hard Part OpenAlly field notes on Gemma 4 for Android with LiteRT-LM: exact pins, S26 Ultra GPU results, 10–15s first response, token limits, and poor tool calling. Canonical: https://openally.ai/blog/gemma-4-android-litert-lm Markdown: https://openally.ai/blog/gemma-4-android-litert-lm.md Published: 2026-08-19 Updated: 2026-09-11 Reviewed: 2026-08-19 Author: Matterward Labs (https://matterwardlabs.com) Category: Engineering Keywords: Gemma 4 Android; LiteRT-LM Android; on-device AI agent; Gemma 4 tool calling; Android local LLM; Galaxy S26 Ultra AI benchmark; OpenAlly on-device models Audience: Android users curious about on-device AI; Developers building local AI agents; Teams evaluating Gemma 4 and LiteRT-LM; OpenAlly users who want the engineering story Disclosure: Matterward Labs makes OpenAlly. OpenAlly results in this article come from dated device logs and package records. Google benchmark figures are labelled separately and retain Google's published test conditions. Loading a local language model on Android sounds like the hard part. In our work bringing **Gemma 4 to OpenAlly with LiteRT-LM**, the trouble started after the model loaded. An AI agent receives more than the sentence a person typed. Before it can answer, it may also receive operating instructions, conversation history, descriptions of the tools it can use, results returned by those tools, and enough empty space to write a useful response. On a small context window, those pieces compete for the same limited token budget. **Fitting an agent into the context window was harder than fitting the model into memory.** This is our dated engineering account. Plain conversation was good when the prompt fit. In OpenAlly's June–July 2026 tests, however, both Gemma 4 E2B and E4B performed badly at tool calling, especially when a request needed several calls across several turns. That difference shaped the work described below. > [!IMPORTANT] The dates behind this article > Gemma 4 was released on **31 March 2026** and publicly announced on **2 April 2026**. OpenAlly pinned its LiteRT-LM runtime snapshot on **27 June 2026**, ran the Galaxy S26 Ultra device investigation on **29 June 2026**, diagnosed the tool-prompt overflow on **21 July 2026**, and verified the package records in this article on **19 August 2026**. Google's [Gemma release notes](https://ai.google.dev/gemma/docs/releases), [Gemma 4 announcement](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/), and [Gemma 4 model card](https://ai.google.dev/gemma/docs/core/model_card_4) provide the release and architecture context. The rest of this article explains what those capabilities met inside a working Android agent. #### The exact LiteRT-LM and Gemma 4 packages we pinned Reproducibility starts with more than a product name. “LiteRT-LM” and “Gemma 4 E4B” can point to different commits, exports, quantisation choices, and compiled context limits over time. OpenAlly's Android runtime uses the [LiteRT-LM project](https://github.com/google-ai-edge/LiteRT-LM). When this article was published, our runtime was pinned to the exact [LiteRT-LM commit 23f36e8](https://github.com/google-ai-edge/LiteRT-LM/commit/23f36e802e29bb71397ce0540483ed17e07ce203), committed on 27 June 2026. Git describes that snapshot as `v0.14.0-alpha.0-53-g23f36e8`. It is an exact pre-release snapshot after v0.13.1, not a claim that OpenAlly uses a stable v0.14 package. The underlying engine was pinned to the [LiteRT core commit c6de975](https://github.com/google-ai-edge/LiteRT/commit/c6de9753d8e03384d390a0f27c37e0537df0b955). On 22 August 2026 the Android runtime moved to [LiteRT-LM commit f1ec4db](https://github.com/google-ai-edge/LiteRT-LM/commit/f1ec4dbcea6667195f90c06191f6a4525efbd475), over LiteRT core b0f6c120. The table records the pins as they stood on 19 August. | Component | Exact package record verified 19 August 2026 | Why the exact pin matters | | --- | --- | --- | | LiteRT-LM | `23f36e802e29bb71397ce0540483ed17e07ce203` | The C interface was evolving, so a commit is more precise than a broad version label | | LiteRT core | `c6de9753d8e03384d390a0f27c37e0537df0b955` | The runtime underneath LiteRT-LM affects delegates, execution, and compatibility | | Gemma 4 E2B | [File revision 6e5c4f1](https://huggingface.co/litert-community/gemma-4-E2B-it-litert-lm/tree/6e5c4f1e395deb959c494953478fa5cec4b8008f), `2,588,147,712` bytes, SHA-256 `181938105e0eefd105961417e8da75903eacda102c4fce9ce90f50b97139a63c` | A smaller package gives more phones room for the model, cache, and Android itself | | Gemma 4 E4B | [File revision 28299f3](https://huggingface.co/litert-community/gemma-4-E4B-it-litert-lm/tree/28299f30ee4d43294517a4ac93abd6163412f07f), `3,659,530,240` bytes, SHA-256 `0b2a8980ce155fd97673d8e820b4d29d9c7d99b8fa6806f425d969b145bd52e0` | More model capacity also brings greater memory and thermal pressure | These are the packages in OpenAlly's Android model catalogue as verified on 19 August 2026. The exact revisions and hashes make the model bytes reproducible instead of relying on a moving filename or a broad family label. The performance and tool-calling findings below come from our June–July 2026 Galaxy S26 Ultra test cycle. On 19 August 2026, OpenAlly's in-app model catalogue gave E2B an 8,192-token window. E4B started at 4,096 with an 8,192-token package ceiling. The app used 6 GiB and 8 GiB device-memory floors for E2B and E4B respectively. Those are OpenAlly safety gates, not universal Android requirements published by Google. There are three context numbers worth keeping separate. Google's [core model card](https://ai.google.dev/gemma/docs/core/model_card_4) lists a 128K architectural context for E2B and E4B. The current E2B and E4B LiteRT-LM package pages document support up to 32K. OpenAlly's in-app catalogue then chose an 8,192-token ceiling for these mobile packages. An architecture ceiling, a downloadable package ceiling, and an app's operating limit are not the same promise. #### A local model sees far more than your message Tokens are small pieces of text that a model reads and writes. A context window is the maximum working space for those pieces during a request. Imagine a 4,096-seat room. Your message may need only nine seats, but the agent's instructions, conversation, and tool descriptions enter first. A tool description carries a machine-readable name, purpose, argument fields, allowed values, and usage guidance. The model reads that schema before it chooses whether to call the tool. On 21 July 2026, one OpenAlly prompt reported this budget against that same earlier 4,096-token export: | Prompt part | Estimated tokens | | --- | --- | | Tool schemas for 15 tools | 2,825 | | History sent on the wire, including the system instructions | 5,612 | | System instructions within that history | 5,483 | | New user message | 9 | | **Actual wire total** | **8,446** | | **Artifact context window** | **4,096** | The 5,483 system tokens are already inside the 5,612 history tokens. Adding them again would double count them. The correct total is the recorded wire total: **8,446 tokens trying to enter a 4,096-token window**. As a ratio, the request was **206% of the window** (8,446 divided by 4,096), and **4,350 tokens never got through the door**. That is more than the entire window again, discarded before the model had written a single character of an answer. The nine tokens the person actually typed were competing for seats with a queue twice the size of the room. That also explains why starting a fresh chat did not solve the underlying case. Most of the budget was spent before a long conversation had developed. > Interface view: **What reached the 4,096-token door.** The user message is tiny beside agent instructions and 15 tool schemas. The recorded 8,446-token request overflows that earlier export's 4,096-token window before a useful answer has room to form. #### Removing the tools made the conversation work We ran the most useful control test: same phone, same model path, same runtime, but with all tools removed from the prompt. The model produced a real response. That narrowed the fault to the request rather than the model: it could load and generate; what surrounded it was too large. Removing tools is not a complete agent solution. It removes the schemas, which gives the model room to talk, but it also removes its ability to take actions. A useful on-device agent has to keep both sides of that trade: - Select a smaller working set of relevant tools for each turn. - Use terse schemas that preserve essential meaning. - Reserve 1,024 tokens for the answer instead of filling every seat with input. - Cap verbose tool results before they re-enter the conversation. - Carry only the history that remains useful for the next decision. The goal is **fewer instructions competing at once**, not fewer capabilities. A [business agent](https://openally.ai/blog/ai-agent-runs-small-business) may know hundreds of possible actions while showing the local model only the small set needed for the current request. #### Google's lab numbers and OpenAlly's field result Google has published Android measurements for Gemma 4 E2B and E4B with LiteRT-LM. The table below reproduces those reference figures with their test shape: 1,024 input tokens for prefill, 256 output tokens for decode, a 2,048-token context, and warm caches. These are **Google-published reference numbers, not OpenAlly measurements**. | Model and backend (S26 Ultra) | Prefill tokens/s | Decode tokens/s | Time to first token | | --- | --- | --- | --- | | Gemma 4 E2B, CPU | 557 | 46.9 | 1.8 s | | Gemma 4 E2B, GPU | 3,808 | 52.1 | 0.3 s | | Gemma 4 E4B, CPU | 195 | 17.7 | 5.3 s | | Gemma 4 E4B, GPU | 1,293 | 22.1 | 0.8 s | Those eight figures are published on the [E2B](https://huggingface.co/litert-community/gemma-4-E2B-it-litert-lm/tree/6e5c4f1e395deb959c494953478fa5cec4b8008f) and [E4B](https://huggingface.co/litert-community/gemma-4-E4B-it-litert-lm/tree/28299f30ee4d43294517a4ac93abd6163412f07f) LiteRT-LM package cards, at the exact revisions pinned above, and every row is measured on an S26 Ultra, though the cards do not name a regional variant. Google's [LiteRT-LM performance write-up](https://developers.googleblog.com/blazing-fast-on-device-genai-with-litert-lm/) covers the same work in round numbers and sends readers to those cards for the per-device detail. The cards are also explicit about what the fastest column leaves out: time-to-first-token does not include load time, the benchmarks ran with caches enabled and initialised, and first-run latency and memory may differ. So 0.3 s is the distance between an already-loaded model and its first token, not the wait after a person chooses a model. The table still shows why accelerated prefill matters, because the phone reads a long prompt far faster before it starts writing. It does not promise the same speed for every phone, artifact, prompt, temperature, thermal state, or agent workload. Our Galaxy S26 Ultra (`SM-S948B`) test used 12 GB of RAM. **GPU-only inference ran, but the first interaction took roughly 10–15 seconds and throughput was poor.** The model was running locally and producing output; it simply was not close to the responsiveness an everyday OpenAlly agent needs. For OpenAlly's daily agent workload, that observed speed was the result that mattered. > Interface view: **Reference speed beside the field result.** Google's controlled CPU and GPU figures sit beside OpenAlly's S26 Ultra result: GPU-only inference ran, the first interaction took roughly 10–15 seconds, and generation remained slow under the tested workload. #### Twelve gigabytes of RAM did not mean twelve free gigabytes The 4.58 GB model file from the June test (an earlier export compiled for a 4,096-token window, not one of the packages pinned above) was only one part of the memory bill. At runtime, Android still needs memory for the operating system, OpenAlly, model working buffers, converted weights, the key-value cache that holds context, and the graphics or CPU backend. In one June E4B load attempt, Android reported roughly **5.2 GB available memory** and the process failed under memory pressure. That is one observation on one device and artifact, not a universal minimum-RAM rule. It taught us to consider available memory and runtime headroom rather than comparing the download size with the number printed on the phone's box. The context limit created a separate failure. That earlier export had been compiled for 4,096 tokens. Forcing a 32,768-token setting could not create a larger model cache after the fact; it drove execution beyond the artifact's safe attention-mask bounds. The durable rule is to respect the context capacity packaged into the artifact. GPU execution solved only one part of the problem. In our test, the model ran on the GPU path and still felt slow: roughly 10–15 seconds for the first interaction, followed by poor generation throughput. A local-agent benchmark has to measure the experience around the model, not just confirm that a GPU delegate opened. #### Why multi-turn tool calling was the harder test Plain chat asks the model to read and respond. A multi-turn agent task may ask it to: 1. Choose the correct tool. 2. Produce valid structured arguments. 3. Read the tool result. 4. Decide whether another tool is required. 5. Preserve the user's intent across the growing history. 6. Finish with a useful answer rather than another unnecessary call. Every result adds tokens. Every round creates another chance to choose the wrong tool, lose an argument, or stop early. In our June–July 2026 tests, general conversation was good once the prompt fit, but **Gemma 4 E2B and E4B both performed badly at multi-turn tool calling**. E2B regularly damaged tool names or arguments. E4B's extra capacity did not make the full OpenAlly chain dependable once schemas, returned data, and history accumulated. These were direct results from our Android tests in OpenAlly's tool-heavy agent environment. Google's [structured tool-use guide](https://ai.google.dev/gemma/docs/capabilities/text/function-calling-gemma4) explains the expected call loop for the model. Its [Gemma 4 edge-agent work](https://developers.googleblog.com/bring-state-of-the-art-agentic-skills-to-the-edge-with-gemma-4/) also shows active progress toward useful local actions. One lesson from Google's [FunctionGemma fine-tuning guide](https://ai.google.dev/gemma/docs/functiongemma/finetuning-with-functiongemma) is especially relevant: correct function-call syntax and dependable task intent are different skills. A model may know the shape of a call while still benefiting from examples that teach which tool sequence solves a real workflow. > Interface view: **Where the multi-tool run lost state.** The trace follows one OpenAlly request through inventory and order tools before the payment-link step fails to complete. Plain conversation stayed useful; E2B and E4B both performed badly across repeated tool turns. #### What should improve from here The opportunity remains strong because the weak points are concrete rather than mysterious. **Larger safely packaged context** gives instructions, selected tools, results, and answers more room, provided the memory cost remains suitable for a phone. **Stronger accelerated backends** can reduce prompt-reading time and make local interaction feel immediate. **Better native tool support** can improve structured calls without surrounding the model with so much explanatory text. **Specialised routing and fine-tuning** can teach a compact model the smaller set of actions that matter for a particular task. Google is approaching on-device Gemma from more than one direction. LiteRT-LM gives developers direct runtime and packaging control. Android's [Gemma 4 AICore developer preview](https://developer.android.com/blog/posts/announcing-gemma-4-in-the-ai-core-developer-preview) represents a separate system-managed path. Those approaches can move at different speeds, but both increase the pressure for clearer backend reporting, dependable tool calls, and efficient context use. OpenAlly's direction follows the evidence: keep normal conversation natural, select only relevant actions per turn, protect answer space, constrain tool-result growth, and measure the backend that actually ran. How an on-device route sits beside provider keys, your own server, and OpenAlly Cloud is set out in the [trust and control guide](https://openally.ai/under-the-hood). For the reader's-eye version of the same trade, covering what a phone runs offline on Android and iPhone and where acting on real accounts becomes the harder half, see [AI on your phone: what runs offline and what real work takes](https://openally.ai/blog/ai-on-your-phone-offline). > Interface view: **A more capable local-agent path.** More usable context, verified acceleration, compact tool selection, and task-aware model training converge on an Android agent that can converse and act without wasting its token budget. #### Frequently asked questions ##### Can Gemma 4 run fully on an Android phone? Yes. Gemma 4 has official on-device packages and Android runtime paths. Whether a specific package runs well depends on the device, available memory, packaged context, backend support, and workload. Loading the model and running a multi-tool agent are different levels of difficulty. ##### Which Gemma 4 model does OpenAlly use on Android? The selected official packages verified on 19 August 2026 are Gemma 4 E2B and E4B LiteRT-LM exports. Their exact repository revisions, byte sizes, and SHA-256 values appear in the pin table above. OpenAlly can choose the package that suits the phone's practical memory headroom. ##### Why do AI tools use so many tokens before they run? The model needs a description of each available tool, including its purpose and structured arguments. Those schemas are prompt text. Fifteen detailed tools can consume thousands of tokens even when the model ultimately calls only one. ##### What worked well, and what failed, in OpenAlly's tests? Plain conversation and general-purpose answers were good when the prompt fit. Tool use was the failure point: in OpenAlly's June–July 2026 tests, Gemma 4 E2B and E4B both performed badly at choosing and chaining tools for daily agent work. The context budget, schema load, tool results, and multi-turn decisions made that workload much harder than chat. ##### How fast was OpenAlly's Galaxy S26 Ultra test? GPU-only inference ran, but the first interaction took roughly 10–15 seconds and throughput was poor. Google's E2B GPU reference reaches a first token in 0.3 s, but that figure excludes model load time and runs with caches already initialised, so it is not the wait a person feels after choosing a model. OpenAlly's field workload produced a much slower user experience. #### The hard learning On-device AI is often discussed as a model-size contest: can the weights fit, and how many tokens per second can the phone produce? An agent adds a more important question: **after the instructions, tools, results, history, and reply reserve are counted, is there enough useful context left to do the job?** On 21 July, removing the tools turned a failed request into a real conversation on the same device. That single test changed the engineering focus: treat context as a strict product budget, rather than abandoning local models or pretending every tool could fit. Gemma 4 and LiteRT-LM make serious local Android inference possible. Building a dependable local agent on top takes the less glamorous work too: exact package pins, verified backends, safe memory margins, compact schemas, selective tools, bounded results, and honest measurements. #### Declared sources - [Gemma release notes](https://ai.google.dev/gemma/docs/releases) - [Gemma 4 announcement](https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/) - [Gemma 4 model card](https://ai.google.dev/gemma/docs/core/model_card_4) - [LiteRT-LM Android performance](https://developers.googleblog.com/blazing-fast-on-device-genai-with-litert-lm/) - [Gemma 4 agentic skills on the edge](https://developers.googleblog.com/bring-state-of-the-art-agentic-skills-to-the-edge-with-gemma-4/) - [Function calling with Gemma 4](https://ai.google.dev/gemma/docs/capabilities/text/function-calling-gemma4) - [FunctionGemma fine-tuning guide](https://ai.google.dev/gemma/docs/functiongemma/finetuning-with-functiongemma) - [LiteRT-LM repository](https://github.com/google-ai-edge/LiteRT-LM) - [OpenAlly LiteRT-LM runtime pin until 22 August 2026](https://github.com/google-ai-edge/LiteRT-LM/commit/23f36e802e29bb71397ce0540483ed17e07ce203) - [OpenAlly LiteRT-LM runtime pin from 22 August 2026](https://github.com/google-ai-edge/LiteRT-LM/commit/f1ec4dbcea6667195f90c06191f6a4525efbd475) - [OpenAlly LiteRT core pin until 22 August 2026](https://github.com/google-ai-edge/LiteRT/commit/c6de9753d8e03384d390a0f27c37e0537df0b955) - [Gemma 4 E2B LiteRT-LM package](https://huggingface.co/litert-community/gemma-4-E2B-it-litert-lm/tree/6e5c4f1e395deb959c494953478fa5cec4b8008f) - [Gemma 4 E4B LiteRT-LM package](https://huggingface.co/litert-community/gemma-4-E4B-it-litert-lm/tree/28299f30ee4d43294517a4ac93abd6163412f07f) - [Gemma 4 in Android AICore](https://developer.android.com/blog/posts/announcing-gemma-4-in-the-ai-core-developer-preview) ### OpenAlly vs FoneClaw: Mobile AI Agents Compared Compare OpenAlly and FoneClaw across platforms, Android screen control, business records, model choice, privacy, and their documented editions. Canonical: https://openally.ai/blog/openally-vs-foneclaw Markdown: https://openally.ai/blog/openally-vs-foneclaw.md Published: 2026-08-17 Updated: 2026-10-06 Reviewed: 2026-08-25 Author: Matterward Labs (https://matterwardlabs.com) Category: Comparison Keywords: openally vs foneclaw; foneclaw alternative; mobile AI agent comparison; AI agent for personal and business use; Android phone agent Audience: Individuals choosing a mobile AI assistant; Small businesses comparing AI agents; People who want Android screen automation Disclosure: Matterward Labs makes OpenAlly. We checked FoneClaw against its official public pages and repository, and we link the evidence so you can verify the comparison yourself. OpenAlly and [FoneClaw](https://www.foneclaw.ai/) are both mobile AI agents, but they start from different jobs. FoneClaw centres on asking an Android phone to carry out supported actions. OpenAlly centres on helping an individual or business run work across mobile and Mac, with Android screen control available through the separate Aster companion. So the useful question is which product matches the work you want to do, not which has the longer feature list. > [!IMPORTANT] A fair comparison > Matterward Labs makes OpenAlly. FoneClaw facts in this article come from FoneClaw's official website, public documentation, privacy policy, and repository. Product details were reviewed on 17 August 2026, and the download page and privacy policy were rechecked on 25 August 2026. #### Short answer Choose **OpenAlly** when you want one place for personal work and business operations: knowledge, sites and apps, customers, products, stock, orders, bookings, invoices, payment links, and focused creative tools. Its [current feature catalogue](https://openally.ai/features) spans Android and iPhone, and your account, agents, and chat history continue from the phone to the [Mac app](https://openally.ai/download/mac). You choose how AI requests are processed, and you can change that choice without rebuilding your work. Choose **FoneClaw** when your main goal is to tell an Android phone to perform supported phone actions. Its [public feature pages](https://www.foneclaw.ai/features.html) emphasise visible execution, phone controls, permissions, and more than 120 built-in Android tools. If Android screen control matters inside OpenAlly, [**Aster**](https://github.com/satyajiit/aster-mcp) adds it as an optional, separate, open-source Android companion. That feature is Android-specific; OpenAlly itself is not defined by Android. > Interface view: **One body of work, two useful surfaces.** An OpenAlly interface view connecting the current mobile and Mac continuity flows. #### What each product is actually built to do ##### OpenAlly starts with the work OpenAlly is arranged around outcomes for individuals and businesses. You can ask questions over your own knowledge, build a site or web app, manage products and stock, turn a customer request into an order, prepare an invoice, create a payment link, or use a focused app for something useful or fun. Those records belong to one working system rather than a collection of unrelated phone actions. A product can move through stock, sale, payment, and invoice without making you rebuild the context at every step. A customer conversation can be one way into that flow, including supported WhatsApp and Telegram connections, but the channel is not the product. ##### FoneClaw starts with the Android phone FoneClaw describes itself as an Android phone agent. You speak or type a request; FoneClaw plans the steps, selects supported tools, and applies permission and approval controls according to the task and your settings. Its public feature catalogue covers device checks and settings, messages and calls, connectivity, screenshots, email, calendar, maps, web tasks, workflows, and skills. That focus can also help a small business. FoneClaw's [own business guide](https://www.foneclaw.ai/small-businesses-use-ai-agents.html) shows it operating the apps a business already uses for replies, posts, orders, and inventory. Its [current public feature catalogue](https://www.foneclaw.ai/features.html) does not document a built-in product, inventory, order, invoice, or payment system of record. #### Before comparing: FoneClaw ships a Full APK and a Lite Play edition FoneClaw's [download page](https://www.foneclaw.ai/download.html) now presents two editions side by side: a **Full** build behind a Download Full APK button noted as Complete features, and a **Lite** build behind a Get Play Lite button noted as Lite features, Play updates. Its [privacy policy](https://www.foneclaw.ai/privacy.html) is written for that Lite edition and explains how the Google Play edition processes information. - The **Full APK** documents deeper Android actions, including screen interaction through accessibility and additional device integrations. - The **Lite edition on Google Play** is defined by its own policy as one that does not use an Accessibility Service, read SMS or call logs, scan all installed apps, monitor your photo library in the background, install APK plugins, or update itself outside Google Play. It uses the Android photo picker for media you select yourself, and calls and text messages are handed off to an Android system screen for you to complete. Rows about FoneClaw screen control below refer to the Full APK documented by its download, feature, and repository sources. Where the Lite edition is mentioned, we use only the boundary stated in its own policy, in full: the photo-library clause is part of that boundary and counts in FoneClaw's favour. #### OpenAlly vs FoneClaw at a glance | Question | OpenAlly | FoneClaw | | --- | --- | --- | | What is the product centred on? | Personal work and business operations across a shared set of apps, records, agents, and AI choices. | Supported actions on an Android phone, with visible plans and permission controls. | | What is the platform story? | Workflows span Android and iPhone. Your account, agents, and chat history continue on [Mac](https://openally.ai/download/mac), which is a direct download rather than a Mac App Store listing. Current install options are kept on the [download page](https://openally.ai/download). | Its [download page](https://www.foneclaw.ai/download.html) documents Android 9 or later. No official iPhone or Mac app is documented there. | | Are business records built in? | Products, variants, stock, customers, orders, bookings, invoices, payment links, and payment records are part of the product. | It can operate existing business apps; its public catalogue does not document a native business system of record. | | How does Android screen control work? | Through Aster, an optional separate open-source Android companion. App Automations are Android-specific. | It is a central capability of the Full APK. The Lite edition on Google Play has a narrower permission set. | | Can work continue on a computer? | Yes. Your account, agents, and chat history continue on [Mac](https://openally.ai/download/mac), where you connect your own AI providers, assign skills and tools per agent, and connect channels and agent hooks. Remote Deck can turn a phone into a controller for approved Mac actions. | No official desktop product is documented on the reviewed public pages. | | What model choice is offered? | 46 current provider entries, covering your own keys, supported subscriptions, your own server, supported on-device choices, and OpenAlly Cloud. | A free default model, a custom OpenAI-compatible online endpoint, or a compatible local model imported in the app. | | Does it connect to customer conversations? | Messaging channels such as Telegram, or a personal WhatsApp account linked by QR code, can carry a selling or service workflow. They are integrations, not the centre of the product. | Official examples describe WhatsApp notification summaries alongside general Android screen control. The reviewed sources describe UI and notification automation, not a documented WhatsApp bot or channel integration. | | Can a business answer on its own WhatsApp number? | Yes, through [WhatsApp Business](https://openally.ai/product/whatsapp-business), a separate app that connects the business's number through Meta and is included with [Premium](https://openally.ai/premium). | Not documented. The reviewed sources describe UI and notification automation, not a WhatsApp bot or channel integration. | | What is publicly open? | Aster is open source. OpenAlly is not presented as an open-source app. | The public repository includes documentation, skills, workflow templates, and plugin source. Its licence statement gives an MIT licence to the public documentation, skills, and workflow templates. | | Who is it a natural fit for? | Individuals and businesses that want personal work, business records, creation, and AI choice to stay connected across devices. | People who primarily want an AI agent to carry out supported Android phone actions. | #### Three workflows that reveal the difference ##### 1. Running a small business from the same place you talk to customers Suppose a customer asks for a product in a supported chat. In OpenAlly, that request can become a quote, an order that reserves stock, a payment link or cash-on-delivery choice, and an invoice or receipt. The owner enables the Sales skill and decides which shop and actions the agent may use. What matters is that products, stock, orders, payments, and invoices remain connected after the conversation moves on, whether the request arrived through WhatsApp, Telegram, or anywhere else. The practical guide to an [AI agent for small business](https://openally.ai/blog/ai-agent-runs-small-business) follows that journey from Google discovery and WhatsApp enquiry through booking, payment, invoice, review, and repeat business. > Interface view: **From product to paid order.** OpenAlly's inventory, sales, payments, and invoicing flows, using sample shop values. FoneClaw can help operate the Android apps a business already uses. That is useful when the job is navigating those apps. OpenAlly is designed for the second case: keeping business records and their relationships together. ##### 2. Starting on mobile and continuing on Mac OpenAlly treats mobile and Mac as different views of the same work. Your account, agents, and chat history continue from the phone to the [Mac app](https://openally.ai/download/mac), so a request captured while moving can be finished with more room at a desk. What the Mac surface gives an agent, as the [Mac download page](https://openally.ai/download/mac) and the [trust and control guide](https://openally.ai/under-the-hood) describe it: - Your own AI providers, with the model chosen for each agent, or a model running locally. - Skills and tools assigned per agent, with a review of what each one is allowed to do. - Channels and agent hooks, so an agent can act on incoming messages and events. Android screen control does not travel. [App Automations](https://openally.ai/features/skills/app-automations) reach the screen through the separate Aster companion, so they stay scoped to the Android device that has Aster installed. What continues to the Mac is the account, the agents, and the chat history, without the screen permissions. The Mac app is also a direct download from OpenAlly rather than a Mac App Store listing. [Remote Deck](https://openally.ai/features/apps/remote-deck) adds a focused version of that relationship: the phone becomes a custom control deck for actions the Mac has approved. FoneClaw's current public product is built around the Android handset itself. Its [download page](https://www.foneclaw.ai/download.html) is headed Download FoneClaw for Android, and the reviewed pages document no iPhone or desktop product. If that handset is where the task begins and ends, its tighter centre may be exactly what you want. ##### 3. Asking Android to use an app for you This is where the products overlap most clearly. FoneClaw's Full APK makes Android phone operation part of its core; its official [scenario guide](https://github.com/FoneClaw-AI/foneclaw-android/blob/main/docs/scenarios.md) and [security guide](https://github.com/FoneClaw-AI/foneclaw-android/blob/main/docs/security.md) explain that boundary. OpenAlly handles that job through Aster and Android-specific App Automations. Aster is deliberately separate. It asks for its own permissions, can be stopped independently, and keeps Android screen access out of the OpenAlly product boundary on iPhone and Mac. That is harder to put in a slogan and easier to weigh when you are deciding what access to grant. > Interface view: **Android screen control stays a separate choice.** An approved Android screen task moving from OpenAlly to the optional Aster companion. #### Model choice and privacy: follow the request path Neither product can be described accurately with one blanket word such as local or cloud. With OpenAlly, working messages and knowledge are stored on the device, while business records and the notes in the Notes app are kept in your OpenAlly account. A request using your own provider key goes directly to that provider. A supported on-device route avoids a remote model; our dated field notes on [running Gemma 4 on Android](https://openally.ai/blog/gemma-4-android-litert-lm) show what that route costs in context budget and first-response time. An OpenAlly Cloud request is processed by the OpenAlly service. A connected service receives the information needed for the action you requested. The public [trust and control guide](https://openally.ai/under-the-hood) explains the choices, and the [privacy policy](https://openally.ai/privacy) carries the full disclosure. The practical advantage is choice without lock-in. OpenAlly's public catalogue contains 46 current provider entries across provider keys, supported subscriptions, your own server, on-device choices, and OpenAlly Cloud. Changing the model route does not require rebuilding your agents, apps, or business records. FoneClaw offers a free default model, a custom online model through an OpenAI-compatible address and key, and compatible local models imported through its in-app Hugging Face path. Its [current privacy policy](https://www.foneclaw.ai/privacy.html) is written for FoneClaw Lite and explains how the Google Play edition processes information. That policy says some databases, settings, recordings, and downloaded local models remain on the device, while online AI, attachments, transcription, speech, feedback, email, web, and related features can use network services involving FoneClaw's service, the selected model provider, or other named processors. FoneClaw's [security guide](https://github.com/FoneClaw-AI/foneclaw-android/blob/main/docs/security.md) separately says the AI model may be cloud-hosted and receives conversation text, action requests, and the results needed for the task; the reviewed sources do not establish that every Lite edition processor or retention rule also applies to the Full APK. > Interface view: **Choose the route that fits the job.** OpenAlly's user-facing processing choices, focused on outcomes and control. #### OpenAlly includes focused apps beyond chat OpenAlly's broader shape also shows up in focused apps that open as their own screen. Each one below is a current entry in the public [app catalogue](https://openally.ai/features/apps), and the platform support given for it is the support that catalogue listed on 25 August 2026. [App Builder](https://openally.ai/features/apps/app-builder) helps create, edit, and preview websites and web apps across Android, iPhone, and Mac. [Vibe Pages](https://openally.ai/features/apps/vibe-pages) turns a mood into a shareable page with a background, fonts, and a playlist, across Android, iPhone, and Mac. [Slap](https://openally.ai/features/apps/slap) turns a tap or shake into a playful reaction across the same three platforms. [Remote Deck](https://openally.ai/features/apps/remote-deck) turns Android or iPhone into a foreground controller for actions approved on a Mac. > Interface view: **Focused tools, one connected workspace.** App Builder, Vibe Pages, Slap, and Remote Deck across OpenAlly's mobile and Mac surfaces. [Web to App](https://openally.ai/product/web-to-app), which joined the app catalogue after that 25 August review, sits beside them: it turns a website you run into a signed Android app, built on an Android phone or a Mac. You install it, share it, or upload it to your own Google Play listing. None of this argues that everyone needs more features. It shows where the product's centre sits: useful work, business, and creative ideas can become focused experiences while OpenAlly stays useful across its mobile and Mac surfaces. #### Which one fits you? Choose OpenAlly if most of these sound like your day: - You want one assistant for personal work and running a business. - Products, inventory, orders, bookings, invoices, and payments need to stay connected. - You start from mobile but want the same work available on Mac. - You want to choose among many AI providers and processing routes without rebuilding everything. - You want Android screen control as an optional companion capability, not the definition of the whole platform. Choose FoneClaw if most of these sound closer: - Your main target is an Android phone. - You want spoken or typed requests to become visible, supported phone actions. - You prefer a large catalogue of Android device tools and workflows. - Your business information already lives in other Android apps and operating those apps is the main job. OpenAlly is the more natural choice for an individual or business that wants connected work across mobile and Mac. FoneClaw is the more direct choice for someone whose centre of gravity is Android phone operation. If the work outlives the conversation that started it, as a customer, an order, a payment, or a published site does, choose OpenAlly. If the job starts and finishes on the handset in your hand, choose FoneClaw. Running both is a legitimate answer, and its honest cost is two permission boundaries to keep track of instead of one. If direct Android control is your main comparison point, also read [OpenAlly vs CellClaw](https://openally.ai/blog/openally-vs-cellclaw) for a second evidence-led look at where phone operation ends and connected work begins. #### How this comparison was made We reviewed FoneClaw's official homepage, features, download page, privacy policy for its Lite edition on Google Play, [public repository](https://github.com/FoneClaw-AI/foneclaw-android), scenario documentation, security documentation, and small-business guide on 17 August 2026, and rechecked the download page and privacy policy on 25 August 2026. We kept capability and privacy statements scoped to the edition each source describes, and used stable wording such as more than 120 tools because FoneClaw's own pages give a rounded figure rather than an exact count. For OpenAlly, we checked the current product catalogue and source-verified release evidence for business records, sales, payments, invoicing, App Builder, Vibe Pages, Slap, Remote Deck, provider choice, privacy controls, and Aster. The comparison includes only features verified for current public use. Mac statements come from the current [Mac download page](https://openally.ai/download/mac) and are limited to capabilities that hold from one release to the next, which is why no version number, file size, or release date appears in this article. The WhatsApp Business, Premium and Web to App lines were added on 10 September 2026 from their product pages. Products change, so the review date and the official source links below are part of the comparison. #### Frequently asked questions ##### Is OpenAlly a FoneClaw alternative? Yes, if you are comparing mobile AI agents. It is not a like-for-like replacement. OpenAlly centres on personal and business work across mobile and Mac; FoneClaw centres on supported Android phone actions. ##### Which is better for inventory, orders, and invoices? OpenAlly has built-in products, stock, customers, orders, bookings, invoices, payment links, and payment records. FoneClaw can operate existing Android business apps, but its current public catalogue does not document an equivalent built-in system of record. ##### Does FoneClaw support iPhone or Mac? No iPhone or Mac product is documented on the public pages reviewed for this article, and its [download page](https://www.foneclaw.ai/download.html) documents Android 9 or later. That is a statement about the current public material, not a prediction about FoneClaw's future. OpenAlly answers the question differently: your account, agents, and chat history continue from the phone to the [Mac app](https://openally.ai/download/mac), which is a direct download rather than a Mac App Store listing. ##### Can OpenAlly automate Android apps? Yes. Android screen control and App Automations use Aster, an optional separate open-source Android companion. Those capabilities are Android-specific; the rest of OpenAlly is not. ##### Is FoneClaw open source? FoneClaw publishes a public repository. Its own licence statement covers the public documentation, Skills, and Workflow templates; the repository also contains plugin source. The reviewed materials do not establish that the entire shipped Android app is open source. ##### Can OpenAlly help sell on WhatsApp? Yes. With the Sales skill enabled and the right shop permissions, a supported customer conversation can lead to a quote, stock reservation, order, payment link or cash-on-delivery choice, and invoice. The WhatsApp channel, which links a personal account by QR code, is one integration surface for that workflow, not the whole product. For a business's own number, [WhatsApp Business](https://openally.ai/product/whatsapp-business) is a separate app: it connects through Meta and is included with [Premium](https://openally.ai/premium); Meta bills its message charges to the business directly, and billable AI messages wait for the owner's approval by default. ##### Is all AI processing local in either product? No blanket claim is accurate. Both products can keep some work on the device and can also use network services, depending on the model, feature, and, for FoneClaw, the edition. Check the chosen route, edition, and current privacy documentation for the specific job. #### Declared sources - [FoneClaw official website](https://www.foneclaw.ai/) - [FoneClaw features](https://www.foneclaw.ai/features.html) - [FoneClaw download and edition details](https://www.foneclaw.ai/download.html) - [FoneClaw Lite privacy policy](https://www.foneclaw.ai/privacy.html) - [FoneClaw official repository](https://github.com/FoneClaw-AI/foneclaw-android) - [FoneClaw scenario guide](https://github.com/FoneClaw-AI/foneclaw-android/blob/main/docs/scenarios.md) - [FoneClaw security guide](https://github.com/FoneClaw-AI/foneclaw-android/blob/main/docs/security.md) - [FoneClaw small-business guide](https://www.foneclaw.ai/small-businesses-use-ai-agents.html) - [Aster open-source companion](https://github.com/satyajiit/aster-mcp)