agent-evals
Runs a suite of cases against one or more models through an adapter the consumer writes, scores each reply with a scorer the consumer writes, and prints leaderboards, per-case breakdowns and Braintrust experiments. The gem knows nothing about what is being evaluated: it owns discovery, the model lineup, providers, parallel execution, reporting and the command line.
Activity
- Latest release
- 2w ago
- Total releases
- 2
- Cadence
- ~daily
- Last 12 months
- 2
Reach
- Downloads
- 0
Details
- License
- MIT
- First release
- Oct 15, 2026
Releases