Hi — looking for guidance on the current state of Agent Canvas after ~15 minutes of usage. Concepts I'm unclear on: - What is a "backend" in this context? - What is a "workspace," and why am I unable to create one? - More broadly, the purpose of each UI element isn't obvious from the interface itself. Issues I'm hitting: - "SDK settings schema unavailable." shown across roughly half the settings screen. - Recurring error every ~10 seconds:

undefined is not an object (evaluating 'e.items.map')

. - Most buttons throw errors when clicked. - No way to start a conversation. - The in-app help links out to docs for what looks like the current SaaS, not Agent Canvas. Is this the expected state, or is there onboarding/setup I've missed?

hey @victorious-market-46627 Thanks for trying it out What is the backend? it is an Agent Server, a REST API that runs agents. It can run locally on your laptop (Docker or native), on a VM you own, or in our hosted cloud. The canvas frontend can talk to multiple backends and switch between them, so think of the canvas as a control plane for all your agents. Ref: OpenHands Agent Server What is a Workspace? it is just a directory on disk. When you start a conversation, a workspace is assigned and the agent works inside that directory. Example: if you keep your projects in

~/git/

and one of them is

~/git/hello_world

, and you want to work on this project with your agent, then

hello_world

is the workspace for that conversation. To make those workspaces selectable, you point the canvas at the parent directory via

PROJECT_PATH

when starting it (e.g.

PROJECT_PATH=~/git

). The canvas then lists the subdirectories as available workspaces. This is needed because when the backend runs in Docker, it has to mount a host directory to give the agent access to your code. One gotcha: by default Docker only mounts

PROJECT_PATH

plus a few config dirs (

~/.openhands

, ...). When you say: why am I unable to create one? you mean: you click New Workspace, and the Add workspace and give you an error? UI affordances aren't well labeled yet. That's a fair point. What do you think we could do better, or what was your experience like? Specifically, what was not intuitive? (cc. @elegant-branch-54423 @hundreds-twilight-32833) About the errors you are hitting, the project is in alpha, so it is worth to opening GitHub issues for the concrete ones. Can I ask you what command are you using and which OS are you on?

Thanks for the feedback and questions @victorious-market-46627! Hopefully the extra context from Vasco helps and if you can share a bit more about your environment we can hopefully help get those error states/bugs sorted out

This is exactly the type of feedback we are looking for! Thanks Mark. More work to do to make sure docs.openhands.dev reflects Agent Canvas. I made a few tweaks to the README to help with explaining the concept of a backend: https://github.com/OpenHands/agent-canvas/pull/368