Prompt Management
The Prompts feature is a versioned registry for the prompts your application uses. Edit prompts in the UI, label a version as production, and your app picks it up — no code change or redeploy needed.
Open it from Prompts in the left nav. Prompts are fetched at runtime by your app via the SDK, so updating a prompt here updates production behavior.
Concepts
| Concept | Meaning |
|---|---|
| Prompt | A named, reusable template (text or chat), with {{variable}} placeholders. |
| Version | Every save creates a new, immutable version, numbered automatically. |
| Label | A pointer to a version — e.g. production, latest, or a custom label. Your app requests a prompt by label. |
| Tags | Freeform labels for organizing and filtering prompts. |
| Folders | Use / in a prompt name (e.g. support/refunds) to organize prompts into a tree. |
Versioning and labels
Prompt management is built around labels so you can change what's live without touching code:
- Each edit is a new version — nothing is overwritten, so history is preserved.
- Move the
productionlabel to any version to deploy it; move it back to roll back instantly. - Apply custom labels (e.g.
staging) to run different versions in different environments. - Protected labels (enterprise) restrict who can move critical labels like
production.
Using a prompt in your app
Create the prompt
In Prompts, create a prompt, write the template with {{variables}}, and save. This is version 1.
Label a version
Assign the production label to the version you want live.
Fetch it at runtime
Use the SDK to fetch the prompt by name and label, fill its variables, and send it to your model. See the Python and JS/TS SDKs.
Iterate
Edit the prompt (creating a new version), test it in the Playground, then move the production label when you're ready.
Linking and metrics
Prompts link to the traces that used them, so you can see how often a prompt version is used and how it performs (latency, cost, scores) — useful for comparing versions before promoting one.
Next steps
- Playground — test and refine prompts interactively.
- Experiments — benchmark prompt versions against a dataset.
- Prompt Operations — the broader PromptOps concepts.