Document Lake is built for the most sensitive documents a startup has. Security is architectural, not bolted on — and because the platform is open source, every claim below is verifiable in the code.
Tenant isolation
Every workspace’s catalog lives in its own isolated storage object (a dedicated SQLite-backed Durable Object) — there is no shared table where a missing filter could leak another customer’s rows. Storage locations are derived from the authenticated request context, never from client input, and workspace references are unforgeable unique IDs.
Encryption
- TLS 1.3 in transit; provider encryption at rest underneath everything.
- Application-layer envelope encryption on top: a master key wraps a per-workspace key, which wraps a per-document key; documents are encrypted with AES-256-GCM in authenticated chunks (tampering, truncation, or reordering fails decryption).
- Workspace deletion destroys the workspace keys — crypto-shredding — making any residual ciphertext permanently unreadable.
Access control & auditability
- Role-based access control (owner / admin / member / viewer) enforced centrally at the API gateway.
- MCP/agent access uses OAuth with tenant-scoped, read-only tokens that owners can revoke.
- An append-only, per-workspace audit log records uploads, views, downloads, searches, member changes, and external agent calls — exportable by owners.
Data residency
EU-region workspaces keep documents, derived text, and vectors in EU-jurisdiction storage, pinned at the infrastructure level. AI inference is processed by the providers on the sub-processors page.
Development practices
All code is open source and developed with a mandatory test gate (unit, integration, and browser end-to-end suites run on every change). Secrets never live in the repository; deployment credentials are least-privilege and rotated.
Reporting a vulnerability
Please report suspected vulnerabilities to security@document-lake.com. We commit to acknowledging reports within 72 hours and will not pursue good-faith research conducted without data exfiltration or service disruption.