Security model
What is implemented and observable in the product, what is policy, what is a target, and what depends on providers.
Last reviewed 2026-11-06
Observed controls (implemented in the product)
- Tenant isolation enforced by database row-level security on every table
- Insert-only audit log for every create, update and delete
- API keys shown once and stored only as hashes; scoped and revocable
- Outbound calls to user URLs are blocked for private networks
- PII redaction for stored conversation content, configurable per workspace
- Release gates block production without passing evals, policies and approvals
Policies
- Incident response policy: TBD: incident response policy document
- Secure development lifecycle: TBD: SDLC policy
Targets (not yet evidenced)
- Recovery objectives RPO/RTO: TBD: RPO/RTO
- Certifications (e.g. ISO 27001, SOC 2): TBD: certification status
Provider guarantees (not verified by us)
- Data residency, backups and subprocessors: TBD: hosting region and subprocessor list
- Model providers' no-training commitments: TBD: per-provider contract evidence
