Run Evaluations
Datasets, test cases, runs, and comparisons — measure template quality instead of guessing.
- Create a dataset —
POST /api/v1/datasets(evaluation.py:80). - Add test cases individually (
…/{dataset_id}/test-cases,evaluation.py:165) or in bulk (…/test-cases/bulk,evaluation.py:188). Each case pairs an input with expected outcomes. - Launch a run against your template —
POST /runs(evaluation.py:258). Runs execute the template per case and score outputs; costs accrue like normal missions. - Compare variants head-to-head with
POST /compare(evaluation.py:285) before promoting changes. - Shortcuts: seed a dataset from a template via
POST /templates/{template_id}/create-dataset(evaluation.py:661), import traces from Langfuse (POST /import/langfuse).
Cross-user privacy: community eval features share scores and metadata, not your raw payloads.
Last updated 2026-08-25 (git-derived)