EvaluationCompare versionsVersion and roll back promptsdashboard
Version and roll back prompts
Keep every prompt version so a score change can be traced to a diff.
What you'll have
Every prompt version kept, deployed by environment, and revertible — so a score drop can be traced to a diff and undone.
You'll need
- A project
- A prompt worth versioning
Prompts change more often than code and are usually versioned worse. The registry keeps every version, so when a score moves you can see which prompt was live.
What it holds
- Every version of every prompt, with its history
- Which version was live over any time range
- Roll back to a previous version
Why it matters for evaluation
A score drop is only actionable if you can tell what changed. Without prompt versioning, "grounding fell on Tuesday" is an observation. With it, it's a diff.