Everything OpenAlly does
The whole surface, in one place.
15 apps, 24 built-in skills and 6 native tools — the same catalogue the app ships, generated from the app’s own manifests so it can’t drift.
One assistant, and a team
Every device gets one Executive Assistant. It holds a full grant and its own read-and-write knowledge base, and it is the thing you actually talk to. When a job is narrow enough to deserve a specialist, it delegates to a sub-agent — each with its own name, avatar, persona, model, auth profile, guardrails and a read-only view of a knowledge base. Delegation goes exactly one level deep, and that bound is the most important thing on this page.
Example team
The orchestrator delegates. Nobody else does.
These five are made up to show the shape. You build your own — however many you want, each scoped to the job you gave it.
Look at what is missing: there is no line between any two sub-agents. There is no way to draw one. Every line in this picture starts at the orchestrator, because only the orchestrator can delegate.
- Triage — sub-agent of the Executive Assistant. Reads what came in and decides what is worth your attention.
- Support — sub-agent of the Executive Assistant. Answers from a knowledge base it can read and never write.
- Research — sub-agent of the Executive Assistant. Granted the tools that read. None of the ones that write.
- Sales — sub-agent of the Executive Assistant. Its own persona, its own model, its own narrow grant.
- Ops — sub-agent of the Executive Assistant. Runs the recurring jobs, and nothing outside its lane.
The same team, on a floor
A day in the office nobody commutes to
The five specialists at their desks, the EA holding the front one. Scroll the room into view and it comes alive as a low-poly cutscene — agents wander, refill, jam the printer, sneak in some "QA" at the gaming corner, and run real jobs through the orchestrator: a refund that starts with Sales and ends with Support, a launch scheduled for 6am. Drag to spin and tilt, pinch or scroll to zoom, hover anyone for their card. The floor plan below is the real content; the 3D scene loads on demand, only when you reach it, so it never slows this page down.
Watch the reporting pattern: every conversation in the room is between a sub-agent and the EA. Sub-agents never talk to each other here either — the floor plan cannot draw that line, for the same reason the diagram above cannot.
How a request flows
- 1You ask the Executive Assistant, in whatever chat you already use.
- 2It decides: handle this itself, or hand it to a sub-agent.
- 3The sub-agent runs with its own grant, and nothing more than that.
- 4The result comes back through the EA, which answers you.
What holds it together
Depth is bound
A sub-agent cannot spawn a sub-agent. Only the orchestrator delegates, and it delegates one level. No recursion, no chain you did not draw, no surprise fourth agent three hops away from anything you approved.
The agent is the unit of access
You do not approve a contact. You approve a contact with an agent. Assign the EA and they get owner-like reach; assign a scoped sub-agent and they get exactly that sub-agent’s grant, which is usually the answer you meant.
Grants intersect
The tools an agent can actually reach are what its auth profile allows, intersected with what the agent itself was granted. Both have to say yes. There is a deny axis too, so “everything except this one” is a thing you can express without listing the rest.
Build your own team
The Agent Admin skill builds and runs sub-agents from any chat — you describe the job, it does the wiring.