It is a CLI, not just a dashboard
Everything the dashboard does, the CLI does too: remoteagent chat, remoteagent exec, remoteagent status, remoteagent doctor. If you live in a terminal, you never have to leave it.
Real commands, not a wrapper
remoteagent chat "why is the build failing?"— ask and watch it think.remoteagent exec— run a one-shot tool call.remoteagent doctor— check the install and environment.remoteagent audit— read the audit trail locally.
The sandbox is the point
Commands are argv-based and allowlisted, deletes go to the trash, and the workspace is bounded. You can hand the agent a real task and not worry it will wipe a directory because a web page told it to.
Keys stay yours
Bring your own keys on-device, or use the vault. Either way, nothing is hard-coded in your repo and nothing ends up in git history.
Scriptable and open
The MIT license and a plain HTTP API mean you can build it into CI, cron, or a plugin. The audit trail exports as JSONL if you want to feed it back into training.