Skip to content

Knowledge Base

The Knowledge Base is how an agent knows what the team already has and what has already been computed, so it builds on that instead of redoing it. Every dataset and indexed storage in the team gets an enriched description: what the data contains, its schema, sample rows, version history, and the code that produced it. Agents read it through the MCP server; people read it in the Studio UI.

Producing a description is called enrichment. It runs when someone asks for it and uses an LLM provider account that the team supplies. For the concept behind it, see Knowledge Base.

Set up

In Team settings → AI features (Admin role):

  1. Add provider: choose Anthropic or OpenAI, paste the API key, pick a default model, and check Set as active. Pricing is filled in for known models and is required before enrichment runs.
  2. Turn on Enable AI for this team.
  3. Check the Monthly budget. Enrichment stops when the month's estimated spend reaches it; the default is $200.

Only enrichment uses the provider. Browsing data and running jobs through MCP do not.

What leaves Studio

For each page, Studio sends the provider one snapshot of the dataset: its description and attributes, schema, up to five preview rows, and for each version the query script, dependencies, and what changed since the previous version. For a storage it sends the listing totals. Nothing else goes to the provider.

Generate pages

  • One dataset: open the dataset's menu in the sidebar and choose Dataset knowledge, then click Enrich. The tab shows the page, its status, and when it was last enriched; Copy copies it as Markdown.
  • One storage: open the storage's menu in the Storages section and choose Storage knowledge.
  • The whole team: in Team settings → AI features → Knowledge base, click Update knowledge base (Admin role). It enriches every dataset and storage that has no page or whose page is outdated, shows progress, and can be canceled.

Note

Currently pages are refreshed manually: Re-enrich on a dataset, or Update knowledge base for the team. When a new dataset version completes, the coverage bar in team settings counts its page as outdated. Automatic refresh on new dataset versions is coming.

A page shows one of these statuses: pending, collecting, and enriching while it is being generated; ready when stored; stale when the budget was reached or a step timed out; failed with the reason. A previous page stays readable after a failure.

Permissions

Roles and grants are explained in Security & Permissions. Admins can do everything below without grants.

Action Requires
Read a dataset's page Viewer or Editor role and a read grant on the dataset
Enrich one dataset Editor role and a write grant on the dataset
Read or enrich a storage's page Viewer or Editor role to read, Editor role to enrich
Manage providers, the Enable AI toggle, the budget, and team-wide updates Admin role

Troubleshooting

  • Provider credentials missing or invalid: set an active provider with a working key under Team settings → AI features.
  • Budget message, status stale: raise the monthly budget or wait for the next month.
  • Nothing to enrich yet: the dataset has no complete version. Wait for its job to finish.
  • Any other failure: check the error shown in the tab and click Retry. If it persists, contact your team admin.