What it does
- Reads eval definitions, run objects, output items, and dataset files.
- Maps
string_check,text_similarity,label_model,score_model,python, andmultigraders to promptfoo assertions, and labels each mapping exact, approximate, or manual. - Turns stored prompt objects (versions and variables) into typed functions you can keep in git.
- Also ships as a command-line tool, including a read-only exporter that only makes GET requests.
It does not call a model, compute scores, or migrate Agent Builder workflows. Manual graders are written as failing placeholders on purpose, so a result can never look green because a grader was skipped.
Get notified
No account with us is required. The waitlist is GitHub-native, so nothing is collected here.
- Watch releases. On the repo, click Watch → Custom → check Releases → Apply. GitHub will email you when a new version is published (for example, Agent Builder support).
- Pin yourself to the Agent Builder thread. Subscribe to the pinned Discussion “Notify me about Agent Builder support” (this tool does not convert Agent Builder workflows yet).
Optional email form — not enabled
This section is a placeholder. No email is collected and the form is not wired to any service. To get updates today, use Watch → Releases or the Discussion above.
Maintainers: to enable a free-tier form later, create an account at Formspree, Tally, or Buttondown (that account requires explicit approval — see APPROVALS.md), paste the form endpoint in the action below, un-hide the form, and loosen only this landing page’s form-action CSP to that provider. Do not change the converter page’s CSP (convert.html must stay connect-src 'none' and form-action 'none'). Require a consent checkbox. Collect nothing without consent.