AI settings
How system administrators enable AI Gateway models, pick default models per use case, edit per-function instructions, and set the agent policy and runtime defaults under /admin/ai.
Plans
How system administrators review the Starter, Pro, Business and Enterprise plan catalog, edit credit package prices, and publish prices to Stripe under /admin/plans.
AI settings
How system administrators enable AI Gateway models, pick default models per use case, edit per-function instructions, and set the agent policy and runtime defaults under /admin/ai.
Plans
How system administrators review the Starter, Pro, Business and Enterprise plan catalog, edit credit package prices, and publish prices to Stripe under /admin/plans.
How system administrators manage platform-wide knowledge collections under /admin/knowledge, and how ingestion and hybrid retrieval work for AI chat.
System knowledge holds platform-wide content for retrieval-augmented generation (RAG) in AI chat and tools. Manage it at /admin/knowledge, under Admin > Operations > Knowledge.
| Route | Purpose |
|---|---|
/admin/knowledge | List system collections. |
/admin/knowledge/create | Create a collection. |
/admin/knowledge/{collectionId} | Edit a collection and manage its files. |
System collections are shared with every organization. Each organization also has its own collections at /{orgSlug}/knowledge. The seeded "CentraKit product guide" collection is an example of a system collection.
A collection groups related files. It has a name, a description, a retrieval priority and a visibility:
| Visibility | Meaning |
|---|---|
| Automatic | Everyone with knowledge access can use the collection in chats. |
| Restricted | Only granted teams and users can use the collection. |
| Disabled | The collection is excluded from retrieval. |
Team and user grants are managed in the Access section of organization collections. System collection pages have no Access section, so use Automatic or Disabled for system content.
Retrieval priority breaks ties between equally relevant results; a lower number ranks first.
Files are uploads or web pages. Each file shows a status: Pending, Processing, Ready, Ready (text only), Failed, Stale or Deleted. A failed file shows the reason next to its status.
On the collection detail page:
GitHub and API connections are not yet available.
From the files table you can Delete a file or Reingest it, for example after changing the embedding model.
Each new file starts the system knowledge-ingestion workflow:
Only configure an embedding default that produces exactly 1536 dimensions. A model with the wrong size fails the file with an actionable error. A temporary provider outage marks the file Ready (text only), so full-text search still works; reingest the file later to add vectors.
Chat and AI tools search knowledge with hybrid retrieval. Vector similarity and full-text rank are combined with reciprocal rank fusion, then ordered by retrieval priority. If the query embedding fails, search falls back to full text so chat keeps working.
Retrieval always runs in the caller's organization and permission context. Row Level Security limits results to collections the user may read, so restricted collections need a grant and disabled collections never appear. The number of chunks retrieved per turn is set by the knowledge limits in AI settings.
| Key | Effect |
|---|---|
system.admin.access | Opens the admin area. |
system.knowledge.manage | Creates, edits and deletes system collections and their files. |
Organization collections use the organization.knowledge.* keys instead.
Last updated on
AI settings
How system administrators enable AI Gateway models, pick default models per use case, edit per-function instructions, and set the agent policy and runtime defaults under /admin/ai.
Plans
How system administrators review the Starter, Pro, Business and Enterprise plan catalog, edit credit package prices, and publish prices to Stripe under /admin/plans.
