Security
Security controls
This page summarizes Open Platform's security controls. Each control is implemented in the source code. The bill of materials lists how each underlying component can be replaced.
Summary
- Changes are reviewed. Review and passing checks can be required before every merge.
- Users are authenticated. Sign-in goes through your identity provider.
- Workloads are isolated. Each team's namespaces are separated by network policy.
- Data is backed up. Backups are copied off-site and restored weekly to verify them.
- AI usage is controlled. You choose the model provider and who merges AI-authored changes.
Change control
Every change is checked and reviewed
Merge policy
Repositories can require approving review and passing checks before merge; both are enforced at merge time. Whether AI-authored changes require a person to merge is configurable, and the history records which rule applied. On the forge, the AI build agent cannot merge its own changes.
Automated checks
Checks run on every pull request and again on the merged result before it is merged. A failed deployment leaves the previous version running.
Protected configuration
Changes from outside contributors that modify checks, release steps, secrets, credentials or permissions cannot be merged.
Scanning
Pushes are scanned for secrets and known vulnerabilities.
Identity and access
Every user is authenticated and authorized
Single sign-on
Applications, preview environments and the MCP endpoint authenticate through the platform's identity service. It accepts Microsoft Entra ID or any OpenID Connect provider as an upstream source; a script configures Entra ID on Azure.
Authorization
Organizations, teams and repository permissions determine who can read, propose, review, merge and administer.
AI assistants
Assistants authenticate with OAuth as a user and receive that user's permissions. See MCP access.
Audit trail
Git history and reviews record each change, its approver and the merge rule applied. The platform logs every MCP tool call.
Workloads and data
Workloads are isolated and data is backed up
Admission control
Kyverno policies validate image sources and generate network policies that isolate each team's namespaces. Each policy runs in audit or enforce mode.
Secrets
Secrets are stored in git, encrypted with SOPS. Your organization holds the encryption key; moving it to a key management service is an integration option.
Backups
Application databases are backed up daily and copied off-site, and a weekly job restores the backups to verify them. The platform's own database has point-in-time recovery.
AI model access
The AI build agent uses the model provider or gateway you configure. The application template contains no AI code.