How Revyse handles your code and data
What leaves your machine, what Revyse keeps and for how long, and how to delete it.
What leaves your machine, and what Revyse keeps
| Data | Sent to | Kept by Revyse |
|---|---|---|
| Your request text, answers and revisions | Revyse API → the AI model provider | For the life of the change (it is the change's history) |
| The selected element's tag, text, classes and position | Revyse API → the AI model provider | With the change. Never form values, passwords or payment fields |
| Source files chosen for the change | Revyse API → the AI model provider | Not kept. Used for the request, then dropped |
| The plan, the list of changed files, and why | — | With the change |
| The diff, and the checks' output | — | 90 days after the change was last touched, then removed |
| The checks' verdicts (passed, failed, skipped, with reasons) | — | With the change |
| Screenshots (browser check, request screenshot, preview check) | Revyse's artifact bucket | 30 days, then deleted from the bucket |
| Cloud workspaces (clones of your repository) | The workspace provider | Not kept. Destroyed when the stage ends, or after 30 minutes |
| Audit trail (who did what, when; ids and counts, never content) | — | For the life of the organization. Append-only |
| Usage and credits (tokens, minutes, charges) | — | For the life of the organization (your bill) |
| Integration tokens (GitHub installation, Vercel) | — | GitHub: never stored, minted per call. Vercel: stored encrypted |
Never sent anywhere: .env files, private keys, credentials, node_modules, .git, or files outside your configured source directories; files in your exclude list are never read. Revyse never logs source code, prompts, or secrets, and refuses to write them into the audit trail.
In local mode, edits are applied only on your machine and stay there until you commit them; the local history under .revyse/ is yours to delete. Local execution does not mean data stays local: planning uses a cloud AI model, through Revyse's API, under the model provider's no-training API terms.
Deleting
- A project: An owner can delete it from the project page, typing its name to confirm. That removes the project, its changes, diffs, checks and screenshots, its GitHub and Vercel connections, its CLI tokens and previews, at once. The audit trail and the credit ledger keep their records (by id, no content). Pull requests already opened on GitHub stay on GitHub.
- Access: Revoke a CLI token on the project page; uninstall the GitHub App or the Vercel integration from GitHub or Vercel, and Revyse disconnects the projects that used them.
- An organization: Ask Revyse. Deleting an organization removes everything above including its audit trail and ledger, and is done by hand during the pilot.
Where things run
Model inference runs with the provider named in the dashboard's model tier list. Cloud changes run in an isolated workspace with network access limited to GitHub and the package registries plus the Revyse API. Previews are built by your own Vercel account.