Your memories are sensitive data. We treat security as a core feature, not an afterthought.
TLS in transit. Optional AES-256 Fernet encryption at rest with per-namespace key derivation.
Full namespace isolation. Every query is scoped. Cross-tenant data leaks are architecturally impossible.
GDPR-compliant forget endpoint. Full audit trail. SOC 2 Type II on our roadmap.
Protection at every layer
All connections use TLS. API endpoints are HTTPS-only. SSL certificates are auto-provisioned and managed.
Scale plan includes optional field-level encryption using AES-256 (Fernet) with HKDF per-namespace key derivation. Memory content and raw content are encrypted before database storage.
PostgreSQL hosted on Neon with SSL-only connections. Managed infrastructure with automatic backups, point-in-time recovery, and encrypted storage volumes.
Multiple layers of identity verification
Your data never mixes with anyone else's
Every memory operation is scoped to a namespace. Namespaces are enforced at the database query level — not just in application logic. There is no API call that can access data across namespace boundaries without an explicit cross-namespace grant.
-- Every query includes namespace scope
SELECT * FROM memories
WHERE namespace = $1
AND is_current = true
AND (expires_at IS NULL OR expires_at > now());
Cross-namespace sharing requires an admin-created grant with optional memory type filters and expiry dates. Shared memories receive a score penalty to ensure your own data always ranks higher.
Full visibility into what happens with your data
/metrics.
X-MindGlue-Signature SHA-256 signatures for payload verification.
You own your data. Always.
The DELETE /memory endpoint permanently removes memories by entity, IDs, or date range — including knowledge graph cleanup. No soft-delete, no residual data.
Export all your data at any time via GET /memory/export. Cursor-based pagination for large datasets. Your data is never locked in.
Set TTL on any memory. Expired memories are filtered at query time (invisible immediately) and permanently cleaned up within 60 seconds.
Your memories are never used to train models. LLM calls (classification, extraction) use your data only for your pipeline and are not retained by providers.
Production-grade hosting with managed services
Railway
Containerized Docker
Neon
SSL-only, pgvector
Auto-provisioned
TLS certificates
Health probes at /health/live and /health/ready for liveness and readiness checks. Automatic container restarts on failure.
Where we are and where we're headed
Found a security vulnerability? We take all reports seriously. Please email us with details and we'll respond within 48 hours. Do not open public issues for security vulnerabilities.
security@mindglue.aiHave questions about our security practices?