Operate agentbox

How-to. Everyday operations. Run them from the repo directory.

Task	Command
Start / stop	docker compose up -d / docker compose down
Status	docker compose ps · uv run --script cli.py models
Apply models.yaml changes	docker compose restart agentbox agentbox-launcher llm-bridge
Apply .env / secrets.env changes	docker compose up -d
Upgrade	git pull && docker compose --profile build build && docker compose up -d
Rebuild only the agent image	docker compose --profile build build agent
Delegation log	tail -f $AGENTBOX_DATA/logs/delegations.jsonl
MCP server / launcher / bridge logs	docker logs -f agentbox · docker logs -f agentbox-launcher · docker logs -f llm-bridge
vLLM backend log	journalctl -t agentbox-<model-id> -f
Stop a resident vLLM model	it stops when another family is needed; to stop now: docker stop agentbox-vllm-<model-id>
Rotate the MCP token	edit AGENTBOX_MCP_TOKEN in .env, run docker compose up -d, re-run claude mcp add …
Clean up a branch after merging	cleanup_delegation(repo_path, branch) or uv run --script cli.py cleanup --repo … --branch …

Stale scratch clones (for example after a crash) live in $AGENTBOX_DATA/worktrees/ and $AGENTBOX_DATA/git/. cleanup_delegation removes a task's clone, and first hands ownership back from the agent uid if needed.

Uninstall: docker compose down, remove the images (docker image rm agentbox/server:local agentbox/agent:local), then delete $AGENTBOX_DATA and the Ollama model dir.