Your code stays yours.
SourcePRD reads your repositories to keep a PRD true — so how we handle your code is the whole ballgame. Here is exactly what happens, with nothing softened.
Access is your grant, scoped and revocable
You connect GitHub, GitLab, or Bitbucket through their own OAuth flows — we never see a password. Tokens are per user: every clone runs with your permission set, never a shared bot account. We read code, read pull-request metadata, and post the statuses and review comments you enable — nothing else. Revoke at your provider any time and access dies instantly.
Credentials are encrypted at rest
Provider tokens are encrypted before they touch storage and are never displayed again after connect. Payment data never reaches us at all — checkout runs entirely inside Paddle, our merchant of record.
Analysis is isolated and ephemeral
Each repository is cloned into its own per-project workspace on sandboxed, ephemeral compute (Google Cloud Run microVMs). Workspaces exist for analysis and evaporate with the instance — your source is never written to durable storage. What we persist is what we made for you: the PRD, reports, reviews.
Never training data
Your code trains nothing. Not our models — we don't build any — and not our AI providers': analysis excerpts go to OpenAI and/or xAI under API terms that prohibit training on your data, and only when you (or webhooks you enabled) trigger analysis.
Access inside the product is role-based
Projects are private to their owner and the people they invite — viewer, editor, or admin, each enforced server-side. Invite links are one-time and stored only as hashes. Demo case studies are read-only for everyone by design.
Found something?
If you believe you've found a security issue, tell us through the contact form — security reports go to the top of the pile, and we'll keep you posted through the fix.
The formal version lives in our privacy policy and terms.
Questions about security?
Architecture, data handling, anything — we read everything.