Results and artifacts

What each run produces

Each run writes a single directory under run_outputs/ (or another root you configure). Inside it you get agent files plus a fixed set of run metadata:

run_outputs/
  run_{timestamp}_{chat_id}_{query_id}/
    agent_outputs/       # files the agent created while running
    run_artifacts/       # events, final result, trajectory, metadata
      events.jsonl
      result.json
      trajectory.json
      metadata.json

The default root is ./run_outputs. Set outputs.root in faraday.yaml to use a different base path.

Trajectory (ATIF) export

trajectory.json is written when the run finishes, at:

<outputs.root>/run_{timestamp}_{chat_id}_{query_id}/run_artifacts/trajectory.json

To turn trajectories off, set storage.save_trajectory: false in config. The run output root still follows outputs.root as above.

CLI: where files land

faraday "<query>" writes events.jsonl, result.json, metadata.json, and trajectory.json under that run’s run_artifacts/ folder.

One-off output location

faraday --artifacts-dir /path/to/output "Your task here"

Copy into a second collection directory (useful for Harbor-style workflows)

• Flag: --collect-artifacts-dir /path/to/collection
• Config: outputs.artifacts.collect_dir: /path/to/collection in faraday.yaml

If /logs/artifacts exists (typical in Harbor-style sandboxes), Faraday also collects there and writes a manifest.json.

Using paths from Python

After a run, FaradayAgent exposes agent.run_artifacts_dir and agent.agent_outputs_dir. See First query for a minimal example.

Batch runs

For many prompts with one artifact tree per run, see Batch runs.

Practical storage

1. Let Faraday write under the working directory (default run_outputs/).
2. After the run succeeds, sync the whole run_{timestamp}_{chat_id}_{query_id}/ folder to durable storage if you need it elsewhere.

This keeps run configuration simple and avoids losing partial uploads if a run fails midway.