Skip to main content
Prompts are project objects that define what your AI application sends to a model. They can be edited, tested, evaluated, versioned, improved, and deployed. Use the Prompts section for the current prompt workflow.

​
Prompt boundaries

A prompt owns its editor state, messages, variables, model settings, linked tools, evaluators, datasets, Improve cycles, and deployment snapshots. Production evidence is connected back to prompts through traces, prompt identifiers, and deployment context.

​
Library view

The project Prompts library helps teams answer operational questions:
  • Which prompts are deployed to which environments?
  • Which prompts have not been evaluated?
  • Which prompts are missing datasets?
  • Which prompts changed recently?
  • Which prompts have deployment snapshots that should be compared before release?
Use the prompt preview sheet for a quick read, then open the prompt editor for authoring, evaluation, and deployment work.
  • Tools add callable actions.
  • Evaluators define quality criteria.
  • Datasets provide test cases.
  • Traces show production behavior.
  • Improve proposes prompt changes.
  • Deploy ships versions to environments.