Hey, the "sandbox" feature agent-canvas has, what is exactly or where can I read about how it works?

When running canvas with the agent-server from the SDK, you can either 1. run the agent-server directly on your local machine: the agent will have access to your files. It can delete anything on your desktop etc... 2. run the agent-server inside a docker container. Each conversation runs inside its own container. It can't harm your laptop, it can't access your files. That's a sandbox.

The agent-canvas itself doesn't create sandboxes right? Thats what I understood

no the code for that is in software-agent-sdk 🙂

im having a hard time to wrap my head around all the difference parts openhands has.. cli, agent-canvas, sdk, openhands itself for a problem I want to solve, will post in the same channel, if you can give me some direction or support will be much appreciated, thanks.

What exactly means openhands conversation? Is that different from a "normal conversation" via the agent-canvas? Im asking because my default agent is claude code (ACP) and the only way I saw to make it know about openhands is before starting a conversation choose the openhands tool, otherwise he has no clue about it, is this the way? or im missing something?

I mean: a conversation in OpenHands, regardless of whether its ACP or OpenHands. Tell the agent you are using OpenHands. Send the github URLs. Ask it any question it should answer with no trouble 🙂

Thanks SImon for clarifying my questions, very helpful