Memory governance for enterprise AI.
Nightwatch governs what can be stored, retrieved, and forgotten — with audit-grade proof.
allow / redact / blockPersistent memory is a new attack surface.
RAG and agents store data indefinitely. Without governance, you're exposed.
PII + secrets persist in vector stores
User data, credentials, and sensitive documents get embedded and never expire.
Retrieval can leak across subjects/tenants
Queries return context they shouldn't access. No scoping, no filters.
Deletion requests are hard to prove
GDPR unlearning without receipts is just a promise, not a guarantee.
AI memory lifecycle
App / Agent
write / query
Nightwatch
Memory stores
Vector DB
SQL / NoSQL
Nightwatch intercepts writes, runs detection (PII/secrets/doc-type), applies policy, and stores governed text + metadata.
Queries go through Nightwatch so retrieval stays scoped and compliant.
Unlearning jobs delete events + vectors and produce an audit record.
What you get today
Ship-ready components, not vaporware.
Dockerized API + worker + Postgres + console
Governed write/read endpoints (allow / redact / block)
Subject-level unlearning jobs with deletion counts
Audit logs for every memory event
Python SDK + JS reference client
YAML policies + detection pipeline (PII/secrets/doc-type)
Document ingestion with fingerprinting + chunking
Deployment
→Deploys inside your environment — no data leaves your VPC
→No model replacement required — works with your existing stack
→Sits between your app and memory stores — transparent integration layer
Design partner program
Integrate in days. Help define the Memory Governance category.
What you get:
- →Deployment modes: Docker Compose today; cloud later
- →Data stays in your environment (no data leaves your VPC)
- →Direct access to engineering team
- →Shape roadmap and define category standards