Semantic memory
How DriftGuard remembers failures and recalls them at PR review time.
Description
Every finding from every PR analysis is converted into a 384-dimensional embedding using the Voyage-3-lite model. The embedding captures: the resource type, the change intent, the blast radius, and the severity. It is stored in a pgvector index alongside the original finding text, the repository, the PR number, and the outcome (blocked / allowed).
How it works
When a new PR arrives, its plan diff is embedded using the same model. A cosine similarity search returns the top-k most similar past incidents with similarity ≥ 0.5. Results are attached to the PR comment as citations.
Description
Memory is isolated per organisation. No cross-tenant recall. On the Team plan, retention is 365 days. On Enterprise, unlimited. Memory is encrypted at rest (AES-256).
API reference
You can query the memory index directly via POST /api/v1/memory/recall. See the API reference for the full schema.