Prompt tools API
⚠️
Preview — this feature is being built; the design below is the spec.
Three endpoints help write and maintain prompts. Each one is an ordinary model call with a carefully built instruction behind it, billed at the target model’s rate, and each returns its working so you can review it before adopting anything.
| Endpoint | Use it when |
|---|---|
POST /v1/prompts/generate | You have a task and no prompt yet. |
POST /v1/prompts/improve | You have a prompt and know what is wrong with it. |
POST /v1/prompts/templatize | You have a filled-in prompt and want a reusable template. |
Preview flag: syntic-beta: prompt-tools-2026-12-01.
Fit with evaluation
Generated and improved prompts are drafts. The intended workflow is: generate or improve, save the result as a prompt template version, then compare it with the version in production through an evaluation run. Only promote it if it wins on the dataset.
What the tools will not do
- They never see your production traffic. They work only on what you send in the call.
- They do not invent policy. When the task description leaves a rule open, the response lists it in
notesinstead of guessing. - They do not tune sampling parameters or choose a model for you.
Was this page helpful?