Security controls, stated precisely
What AgentBrain enforces today, what depends on your deployment and providers, and what is still planned. See the Trust page for status labels.
— Access control
Organization and resource permissions
One permission model covers folders, knowledge items, media, connectors, knowledge graph entities and relations, and workflows. Grants go to users, roles or groups and can be scoped to a department subtree.
- Read, write, admin, run and approve actions per resource
- Row filters on data access
- Agent task and workflow run permissions are checked at the parent workflow only
— Audit
Append-only audit log
The application database role cannot update, delete or truncate audit rows, and database triggers block mutation and schema changes. Retention is configurable from 30 to 3,650 days, with legal hold. There is no cryptographic hash chain, so a database superuser outside the application is not prevented from altering records.
- Append-only for the application role, enforced in PostgreSQL
- Retention 30–3,650 days, with legal hold
- Audit log export to SIEM tools is planned
— Authentication
Sign-in and identity
Sign-in is handled by the NextLevelBuilder identity service (builder-auth), with users created on first sign-in, and API keys for service access. SAML, OIDC and LDAP / Active Directory federation are planned and not available today.
- Managed sign-in through builder-auth
- API keys for service-to-service access
- SAML, OIDC and LDAP federation: planned
— Encryption
Encryption of stored secrets
Connector credentials and AI provider keys are encrypted by the application with AES-256-GCM. Disk encryption and TLS settings depend on the hosting environment. Customer-managed keys (BYOK) and KMS integration are not available.
- AES-256-GCM for stored credentials and provider keys
- Disk and transport encryption depend on the deployment
- BYOK / KMS: not supported
— Deployment
Where AgentBrain runs
AgentBrain runs today as a managed service on our Kubernetes infrastructure. Private deployments on customer infrastructure are scoped per project. A packaged self-hosted release and air-gapped operation are not available today.
- Managed cloud: production-supported
- Private deployment on your infrastructure: scoped per project
- Air-gapped: not supported (sign-in and default AI providers need network access)
| Topic | Managed cloud | Scoped private deployment | Air-gapped |
|---|---|---|---|
| Where data is stored | Our managed region | Your infrastructure (project-scoped) | Not supported |
| Network access needed | Yes | Yes (sign-in, AI providers) | Not supported |
| Encryption key control | Managed by us | Scoped per project | Not supported |
| Certification | None yet | None yet | Not supported |
| Status | Production-supported | Scoped per project | Not supported |
| Typical plan | Business / Enterprise | Custom | — |
— Compliance
Certification status
AgentBrain holds no third-party certification or attestation today. We do not claim SOC 2, GDPR, PDPA, HIPAA, FedRAMP or ISO 27001 compliance. Whether your provider settings train on your data depends on the AI provider you configure.
- ISO 27001, HIPAA, FedRAMP: no certification and no claim
- AI provider data use follows your provider's terms
- security.sections.compliance.bullets.2
Need the details for your security review?
We will walk through data flow, AI providers and deployment assumptions with your team.