How To Use Overmind

Overmind is the fastest way to improve an existing Python agent with measurable, regression-aware changes.

1) Install

pip install overmind

2) Initialize your project

From your agent project root:

overmind init

This creates .overmind/ and collects required API keys and model defaults.

3) Register your agent entrypoint

overmind agent register my-agent agents.my_agent:run

Your entrypoint should accept a dictionary input and return a dictionary output.

4) Set up policy + evaluation

overmind setup my-agent

You can also bring your own policy document:

overmind setup my-agent --policy docs/my_policy.md

5) Run optimization

overmind optimize my-agent

Overmind will:

1. run your agent against a dataset
2. trace and score outputs
3. generate candidate fixes
4. keep only changes that improve results

6) Review artifacts

Results are written to:

• .overmind/agents/<name>/setup_spec/
• .overmind/agents/<name>/experiments/

You can inspect report.md, review diffs, and continue iterating.

Next

• Overmind Optimizer Reference
• Getting Started