Skip to main content

Admin guide · Platform

Prompt Vault

Version-controlled prompts, grouped into projects and compiled server-side at request time. Manage them under Prompt Vault.

Projects

Prompts are organized into projects, each showing prompt / draft / published counts. Create one with New project, or bring prompts in with Import.

The Prompt Vault projects list with prompt, draft and published counts per project.
Prompt Vault — projects with their prompt and publication counts.

The editor & blocks

Source on the left, live preview on the right. Use {{name}} to insert reusable blocks — the preview shows them expanded.

The prompt editor: a file tree of prompts on the left, the prompt source in the middle with an UNACCEPTABLE RISK badge and a Mitigate Risk action, and the expanded preview on the right.
The editor — a file tree, the source, and the preview with blocks expanded.

With the EU AI Act classifier on, each prompt carries its risk level next to the content, with Mitigate Risk proposing a safer rewrite. A dot next to a file in the tree means it has unpublished changes.

Versioning: commits, history & publishing

Versioning works like git: pending edits accumulate as changes, and each save is a commit with a message.

The Changes and History panel next to the prompt editor: one modified file, a Publish changes button, and the commit log below.
Changes & History — pending edits, then Publish changes.
  • Commit messages can be generated by Mato when it is enabled.
  • Publishing vs deploying — prompts are org-wide; the prompt:publish and prompt:deploy permissions separate authoring from releasing.

note · Published prompts are consumed everywhere

Guardrail judges, evaluation runs and API clients resolving prompts on /v1/responses, /v1/chat/completions and /anthropic/v1/messages all compile against the published versions — see the developer page Prompt Vault for the request-side API.