Privacy & data handling

Chiron is built so that depth of knowledge does not require copying your data. This page states plainly what is stored, what is never stored, and how live data is governed.

Metadata out, SQL in

For connected databases, Chiron stores knowledge *about* the database — its catalog, table and column descriptions, classifications, and join structure — not its rows. The footprint per table is independent of row count, so a large estate stays small. Sensitive columns are classified *before* anything is persisted; tagged columns keep statistics only, with their values masked, never stored, and never sent to a model.

Live data is fetched, not retained

When an agent genuinely needs live rows, it queries your database directly under governance:

• a read-only role, verified at connection time
• query validation before execution
• enforced row_limit and timeout_s
• a per-user audit trail

Results flow to the agent. They never land in Chiron's storage. See Database tools and Query your connected DB from chat.

Provenance on every fact

Every fact in the ontology carries its origin — whether it was derived deterministically by a parser or interpreted by a model — and every returned claim keeps a resolvable ref. Consumers can verify rather than trust.

Tenancy

Every knowledge call is scoped to a project_id and checked at fetch time against the caller's scope; a cross-tenant ref resolves to nothing. See Governance & tenancy and Authentication.