Whisper
Native toolTranscribe audio to text on-device with whisper.cpp — no audio leaves the phone. The model is an on-demand download (Android only).
- Android — supported
- iPhone — not available
- Mac — not available
How it reaches an agent
- 1It ships built inWhisper is part of the runtime — nothing to install, nothing to configure into a prompt.
- 2Grant it to an agentA tool is a capability, not an instruction: granting it costs zero context, so the agent's window stays for your work.
- 3One call unlockedThe agent calls it like a function and the result lands back in the conversation.
Zero context cost
A skill carries a teaching prompt, and that prompt takes up room in the model's context window. A tool carries none — it is a capability, not an instruction. Granting Whisper to an agent costs nothing against the window it thinks in.
What it unlocks
One call your agent can make:
transcribe_audio
Needs setting up first
Whisper has a one-time runtime setup inside OpenAlly. You do it once, from the app; after that any agent you grant it to can call it without asking you again.
Get Whisper
It ships inside OpenAlly — one install, the whole catalogue.
More tools
Python
Run real Python on-device — data wrangling, math, parsing uploaded files, quick scripts. The interpreter is an on-demand download (Android only).
- Android — supported
- iPhone — not available
- Mac — not available
Local LLM
Run a small language model fully on-device via LiteRT-LM (Gemma 4) — private, offline, zero-cost completions (summaries, classification, drafting). Models are .litertlm files downloaded on-device.
- Android — supported
- iPhone — supported
- Mac — not available
PDF Reader
Read PDF documents and answer questions about them — contracts, reports, statements, forms. Extracts the text on-device and works with every model (no special PDF support needed). Handles multiple PDFs, page ranges, and encrypted files.
- Android — supported
- iPhone — supported
- Mac — not available
PDF Maker
Create polished, print-ready documents from HTML — reports, letters, summaries, one-pagers, statements. The agent authors the HTML; it is laid out on A4 and returned as a document you can preview, print, or save as a PDF on-device. Works with every model. Author HTML is sanitised (no scripts / external URLs).
- Android — supported
- iPhone — supported
- Mac — not available
Context Retrieve
Let the agent pull back the full original of anything that was compressed earlier in the chat to save context — when a tool result shows a `<<ccr:KEY N>>` marker, the agent fetches the dropped rows back by key. Pure local lookup — works with every model, no setup.
- Android — supported
- iPhone — supported
- Mac — not available