For docs, I'm considering consolidating all the information about "agent servers", "automations server", etc to just "agent canvas" This way there is just one command the user needs to know, and they don't need to know the details of how all the different services get stitched together I can add an advanced section if need be, but I think we'd recommend users just launch a whole canvas instance as opposed to individual agent servers for remote deployments

I think it makes sense to simplify. Are you suggesting: when a user wants to deploy a remote backend, the instructions will basically be to deploy a new agent canvas instead of an agent-server? I wonder if we want a special mode or flag for this, so the user knows how to easily get the API key etc.

Yep thats what I mean And the recommended way to do this is via

LOCAL_BACKEND_API_KEY=<choose-a-strong-secret> agent-canvas --public

I think that makes sense to me - users shouldn't have to learn how to use all the different services. No objections from me unless there's a technical reason not to do this. Also, we may to make sure pages under the

Connect and Manage Backend

reference this install path

Also heads up I reworked the splash page to distinguish "why agent canvas" first instead of "what is agent canvas" (which listed a few basic features)

🤔 I tend to agree on automation server, but isn't agent-server deployment too necessary and too common?

I think we had plans to assign agent canvas the

openhands

command (and deprecate the existing openhands command to

openhands-cli

) In the future we'd support something like

openhands --agent-server

which would launch just the agent server, but in the upcoming launch I don't think we've planned for these convenience commands

Re: this splash page, sounds good... > run conversations with OpenHands, Claude Code, Codex, or Gemini CLI through... I'm reading here that Agent Canvas runs OpenHands like it runs Claude Code, in other words, that

OpenHands = agent-server

. Are we leaving agent-canvas out of the OH/OH repo at least for now? Or I'm reading too much into it? 😅

maybe reading too much into it? i was referring to ACP feature, i don't think that affects whether we move agent canvas into oh/oh repo