Security at DoryAI

Current security boundaries

• Clerk authenticates users before dashboard access.
• Convex operations scope saved links, categories, tags, and chat history to the authenticated user.
• Clerk secrets, the backend shared secret, and webhook signing secrets remain on the server.
• Convex-to-backend requests require a shared server secret plus the intended user and chat session.
• General webpage and remote-image fetching rejects local and private network targets, rechecks redirects and DNS answers, and limits time, response size, and content type.
• URL normalization makes equivalent save retries idempotent.

External services

DoryAI currently relies on Clerk for authentication, Convex for app data and server functions, Google Gemini for AI processing, source websites and media tools for link metadata, and Polar for the existing billing path. Their infrastructure and data practices are governed by their own terms.

Known limitations

• DoryAI has not documented an independent penetration test, security certification, or formal 24-hour monitoring program.
• The product does not provide end-to-end encryption. Server-side processing must read submitted links and chat content to perform the requested work.
• LinkedIn and X may block extraction. DoryAI records a limited, labeled result instead of claiming unavailable post content.

Your controls

You can delete individual saved links and clear chat history in the app. The current product does not claim a self-service account export or complete account-deletion workflow.

Report a vulnerability

Do not publish suspected vulnerabilities or private user data in a public issue. Send the affected revision, impact, and a minimal reproduction to nmamanipantoja@gmail.com. Never include live credentials in the report.