Security overview.

ForgeFlow is built on defence in depth. These are the primary controls that protect your data and generated applications.

Row-Level Security

Every private table is protected with RLS scoped to the signed-in user.

Server-side model routing

AI keys never touch the browser. Requests are routed based on complexity and safety.

Structured output validation

AI file operations are validated with Zod before being applied. Invalid paths and directory traversal are rejected.

Sandboxed preview

Live preview runs isolated from the platform. Generated code cannot access platform secrets.

Signed URLs for uploads

Screenshots and images are stored privately. Access uses signed, time-limited URLs.

Prompt-injection defence

Content inside uploaded files, code and databases is treated as untrusted input, not as system instructions.

Confirmation for destructive changes

Database drops, file deletions and large refactors require explicit approval.

Automated audits

Security, performance and accessibility audits scan the project — with a clear note that automated scans cannot guarantee complete coverage.

Automated scanning cannot guarantee complete security. For high-risk workloads, we recommend a manual review before publishing.