Overview
Build, evaluate, publish, and retrieve prompts with your team.
PromptLens helps your team test prompt changes against representative inputs, compare the results with production, and publish the version your application should use.
Start with your first evaluation. You will build a ticket router, check its answers with an exact-match evaluator, and save an initial baseline. Then take the prompt to production to compare a change, publish it, and retrieve it from an application.
Choose your next task
| I want to… | Start here |
|---|---|
| Write instructions and provide test inputs | Create and edit prompts |
| Check extracted JSON | Evaluate structured output |
| Judge the quality of support replies | Use LLM judges |
| Find regressions before publishing | Compare with production |
| Load a saved prompt from my application | Prompt retrieval, TypeScript, or Python |
| Manage prompts from a terminal or assistant | CLI or MCP |
| Set up my team's model access | Providers and credits |
How the pieces fit
Your organization owns prompts, provider connections, and billing. Each prompt has a personal draft for each editor, immutable saved versions, one shared dataset, and evaluators. An evaluation records what a version produced and how it scored. A label, such as production, points to the saved version your application retrieves.
Editing a draft does not change production. Saving a later version does not publish it automatically. The initial baseline establishes the first saved version and its initial labels; later releases are explicit decisions.
The dashboard guides teach configured evaluators. API, CLI, and MCP walkthroughs use their expected scoring mode for ticket routing. Application retrieval returns prompt configuration; your application makes the model call. See concepts for the full vocabulary or Product guides to browse by task.