Memory & knowledge sources
Opt-in reusable context and five separate collections with visible provenance.
Saved memory is optional
In Account or a selected project, explicitly enable saved memory before adding information. Inspect entries, edit a specific revision, or delete them. Disabling memory stops future use while preserving entries for inspection and deletion. Conversation history remains separate.
Scope and bounds
Memory is scoped to the authenticated user and workspace, with optional project scope. Each entry is limited to 2,000 characters and 4 KB of UTF-8 text, with at most 50 entries. Context inclusion is bounded; a saved entry is not a guarantee that all content will be sent.
Five source collections
Vulnerabilities, documentation, rules and techniques, threat intelligence, and private context are separate collections. The registry shows canonical identity, source/parser/schema versions, license, commercial permission, attribution, hashes and freshness timestamps. Unknown freshness is not represented as current.
Updating and revoking sources
Private source forms save only the content explicitly entered. A canonical identifier does not trigger an arbitrary URL fetch. Administrative updates require the current source generation and explicit replace or delta semantics. Revocation blocks future retrieval; erasing historical records is a distinct operation under the retention policy.