Harbor

Faraday works as a Harbor installed agent. Your normal faraday.yaml controls everything — no Harbor-specific config is needed beyond the launch command.

Quick start

harbor run -d "<dataset@version>" \
  --agent-import-path faraday.integrations.harbor.agent:FaradayHarborAgent

faraday.integrations.harbor.agent:FaradayHarborAgent

That’s it. Harbor handl es installing Faraday from your repo and routing task instructions to the agent.

Configuration

Place a faraday.yaml at your repository root (start from faraday.example.yaml). This is the same config file used for local and Docker runs. The most relevant settings:

app:
  mode: host              # runs inside the Harbor task container
  workspace:
    source_root: /workspace

sandbox:
  backend: docker         # code sandbox (default for Harbor)
  workspace:
    container_path: /workspace

These are the defaults — you only need a faraday.yaml if you want to change something.

Output artifacts

After each run, artifacts are written to /logs/artifacts:

File	What’s in it
result.json	The agent’s final answer and run summary
events.jsonl	Timestamped stream of every agent event
metadata.json	Run parameters (model, steps, config)
trajectory.json	Full trajectory in ATIF format for replay

Override the output location with --artifacts-dir:

harbor run -d "<dataset@version>" \
  --agent-import-path faraday.integrations.harbor.agent:FaradayHarborAgent \
  --artifacts-dir /custom/output