#agent failed to reply
1 messages · Page 1 of 1 (latest)
That one is pretty direct: sandboxing is enabled, but Docker is missing.
Most likely you have agents.defaults.sandbox.mode: "non-main" or "all", and since channel/thread sessions count as non-main, this thread tried to start in Docker and failed.
Try these first:
openclaw --version
openclaw status
openclaw doctor
If you do not want sandboxing, disable it:
openclaw config set agents.defaults.sandbox.mode off
openclaw gateway restart
If you do want sandboxing, install Docker and make sure docker is on PATH, then re-run openclaw doctor.
Useful docs:
- FAQ: https://docs.openclaw.ai/help/faq
- Sandboxing: https://docs.openclaw.ai/gateway/sandboxing
- Docker: https://docs.openclaw.ai/install/docker
- Troubleshooting: https://docs.openclaw.ai/gateway/troubleshooting
If you paste the output of:
openclaw --version
openclaw status
openclaw sandbox explain
I can tell you exactly which config is causing it. 🦐
what would sandboxing do?
Sandboxing puts agent tool execution in an isolated environment instead of on your host.
In practice, that means things like:
execruns in a container- file reads/writes are limited to the sandbox workspace
- installed host binaries/files are not automatically visible
- you get a safer boundary for untrusted chats or multi-user/group setups
Good reasons to use it:
- the agent is exposed in public/group chats
- you want command/file access fenced off from your real machine
- you want stricter isolation between agents/sessions
Reasons to leave it off:
- personal/local setup
- you want the agent to use host tools directly
- you do not want the Docker dependency
A common setup is:
agents: {
defaults: {
sandbox: {
mode: "non-main"
}
}
}
That keeps your main personal session on-host, but sandboxes group/channel sessions.
Docs:
Short version: safer, but more setup and a bit more friction. If this is just your own assistant on your own machine, I’d probably turn it off.