Security at AI CodeOps
AI agents write and run code on your behalf, so the platform is built to keep each workspace's code, secrets and sessions apart, and to leave a record of everything that happened.
Isolation
- The project's tests for an AI change, and every terminal, run in their own sandboxed containers with restricted network egress.
- Workspaces are separated in the database by row-level security: a query made for one workspace can't read another's rows.
- Repository access uses the connection your workspace set up, and only for the repositories you choose.
Accounts and access
- Two-factor authentication with authenticator apps (TOTP) and passkeys, with recovery codes.
- Single sign-on with SAML or OIDC on Enterprise, with verified domains.
- Roles from owner to viewer, custom roles, and separation-of-duties roles such as security and billing admins. See audit and roles.
- Rate limits on sign-in, codes and passwords against brute force.
Secrets
Workspace secrets, model provider keys and integration credentials are encrypted at rest (AES-256-GCM) with a key kept outside the database. Secret values are never shown again after they're saved.
Evidence and audit
- An audit log of sign-ins, membership, settings, approvals and code changes.
- Every change can pass a release gate: tests, an AI findings review and approvals, with two-person review once a workspace has two or more approvers.
- A signed evidence pack per change that anyone can verify at /evidence/verify.
- SBOM records (CycloneDX and SPDX) per project. See SBOM and security.
Self-hosted deployment
Enterprise customers who need their code to stay on their own infrastructure can run AI CodeOps there. It's an Enterprise option: contact sales.
Reporting a vulnerability
See our security.txt for the contact. We'll acknowledge your report and keep you updated while we fix it. Service health is on the status page; data handling and subprocessors are on the trust page.
Security questions
Does AI CodeOps train AI models on my code?
No. Your code and prompts are sent to the model provider that runs your session only to run it. See the trust page for the providers we use.
Where does AI-written code run?
The project's tests for an AI change, and every terminal session, run in their own sandboxed containers on our servers, with outbound network access restricted, so one workspace can't reach another's.
How are my secrets stored?
Workspace secrets and provider keys are encrypted at rest with AES-256-GCM, using a key that is never stored in the database.
Do you have SOC 2 or ISO 27001?
Not yet. We don't claim certifications we don't hold; the trust page will say when that changes.
How do I report a vulnerability?
Email the security contact in our security.txt with 'Security' in the subject. Please don't test against other customers' data.