One project. One token.
Create a project, mint a token for it, and log the CLI in. Every run you sync lands here.
Describe what “good” means — or let Cairn draft it.
Model evals live in one plune.yaml. Cairn, our QA agent, explores your app and proposes UI
tests. Nothing joins the suite until a human accepts it.
Your model answered. Is it right?
Plune checks every answer like a unit test: does it say the fact, is it grounded in your docs, does it answer the question, would a strict reviewer pass it — and what did it cost.
Run the whole suite. Watch it go green.
plune run evaluates every case against your model — locally, or on every pull request.
plune sync keeps the history on the platform.
Swapped the model? Plune noticed.
A new model started promising 60-day refunds. The answer still sounds fine — faithfulness to your policy fell from 0.97 to 0.41. Fix the prompt, re-run, green again.
Every pull request takes the same exam.
The GitHub Action runs your suite on each PR, diffs it against main and keeps one comment up to date. Make the check required, and a regression blocks the merge — not your customers.
Every synced run adds a bar.
A project’s Trends tab charts pass rate and cost run by run, so the run after the model switch stands out at a glance. Analytics sums up every project on one page.
Bring the whole team in.
One company can hold every project. Owners and managers invite, members do the work, read-only reviewers look — each sees what their role allows.
