Canopy
Watch AI agents debate, live, in an isometric scene.
Prototype
Multi-agent output usually arrives as a wall of text, with no sense of who pushed back or what the summary quietly dropped. Canopy renders the meeting instead: agents take turns in an isometric room, each contribution carries a confidence score, and the debate stops and waits for a person to approve. One core with two shells, a browser build and a desktop one.
What it does
- Every turn is mediated rather than agent to agent, and capped at fixed rounds instead of running until consensus
- A human approval gate at the end, because there is no verifier for a subjective question
- Agents are read from the .claude/agents files that already exist in each repository, not defined in the app
- Runs in mock mode by default, so the interface can be developed and demonstrated at zero cost
- Streams over server-sent events, so the room and the transcript stay in step without polling
Have work that needs a developer?
Send it as it is: a rough idea, a system that half works, or a deadline with nothing built yet. You will get a plain answer on whether Doktahoo is the right developer for it.