Prompt Regression Checkfree tool · runs in your browser

Export your OpenAI Evals before they go read-only

OpenAI is deprecating hosted Evals: they become read-only on 31 Oct 2026 and the platform shuts down on 30 Nov 2026 (OpenAI docs, checked 2026-10-10). OpenAI's page doesn't offer an export. This page pulls every eval, run, dataset row, model output and grader result through the API and gives you files you keep.

OpenAI API key

The key goes from your browser straight to api.openai.com. This page has no server, stores nothing and sends nothing anywhere else (view source to check). Use a key with read access to the project that holds your evals; you can revoke it afterwards.

Only evals whose name contains (optional)

Eval	Runs · rows · graders	Download

Files: bundle JSON is lossless (eval config, graders, every run with its outputs and grader results). Dataset JSONL is one line per unique input row, ready to re-use. promptfoo YAML is a best-effort starting point: string checks and model graders are mapped to promptfoo assertions; anything it can't map is kept as a comment, so review it before running.

Nothing yet.

What next?

Run the promptfoo file locally with npx promptfoo eval, or keep the bundle as your record of past results.
We're building Prompt Regression Check: these same cases run on every pull request and every model upgrade, with the diff in the PR. Get early access (free).
Want us to convert your evals into a working regression suite for you? Reply to the early-access email with what you have; agents do the conversion, a person checks it.

Built by Hieu Tran with AI agents, early. If the export fails or misses something, the log above shows what the API returned; tell us via the early-access form and we'll fix it.