Security and trust

A governance layer has to be harder to break than what it governs.

Sentinel holds the keys that agents prove themselves with and the record of everything they did. This is how we protect both.

Keys never on disk

Signing keys are created inside Azure Key Vault, AWS KMS, Google Cloud KMS or HashiCorp Vault (HSM-backed where available). Sentinel asks the vault to sign; private keys never leave it.

Tenant isolation in the database

PostgreSQL row-level security separates tenants below the application. Every query runs as a role that can only see its own tenant's rows.

Tamper-evident audit

Each record is hash-chained to the one before, written before the action runs, and checkpointed with signatures and RFC 3161 timestamps. Edits and deletions show up on verification.

Fail closed

If Sentinel cannot be reached, the SDKs and the MCP proxy deny. A governance outage never turns into ungoverned agents.

Least privilege for people

Operators have roles (admin, policy editor, approver, auditor, agent owner) and see only the tenants they are assigned. Single sign-on through OpenID Connect.

Signed releases

Container images are built in CI, published with SBOM and provenance attestations, and signed with Sigstore (recorded in its public transparency log).

Deployment

Your environment, your data

Sentinel is self-hosted: in your cloud account, on your network, with your database and key vault. Nothing about your agents, tools or audit trail is sent to us.

  • Docker Compose for a single host, a Helm chart for Kubernetes
  • Several instances share one database and stay in sync
  • TLS everywhere, including to the database
  • Backups of the database and audit chain on your schedule

How we build it

Tested like it matters

  • Hundreds of end-to-end checks on every change, against a production build
  • Adversarial suites that replay attacks; all must be blocked
  • Detection rules measured with exact confidence bounds and published method
  • Static checks that every authorization path is audited

Standards

Evidence for your frameworks

Sentinel maps its controls to SOC 2, ISO/IEC 42001, the NIST AI Risk Management Framework and the EU AI Act, and produces signed evidence bundles for your assessor.

These mappings support your assessment. Sentinel itself is not certified against these frameworks, and we do not claim it is.

Responsible disclosure

Found a vulnerability?

Tell us privately and give us a chance to fix it before disclosure. We acknowledge reports within two business days.

Report a security issue