Forge

Privacy

Forge stores account and connection details so it can work with GitHub and carry out changes you approve.

Access is limited to operator-configured GitHub accounts. Each operator supplies their own accounts, credentials and retention policy.

What Forge stores

Account and GitHub access

Your GitHub numeric user id, current login and Forge GitHub App installation id. Forge also keeps one GitHub user credential encrypted at rest. Repository reads, writes, approvals, personal repository creation and explicit public GitHub search use it. GitHub restricts access to the intersection of your permissions and the Forge GitHub App's installed repository access.

OAuth connection

Registered client names and redirect addresses, hashes of short-lived authorization codes, and the user id they belong to. Forge access tokens are signed. Refresh tokens are opaque, client-bound, rotate on every use and are stored only as hashes; inactive refresh tokens expire after 30 days. Used and expired token records are removed with the Forge account and may remain until then so replay can be detected.

Changes and approvals

GitHub remains the only copy of repository files. For a requested merge or discard, Forge stores the repository name, change branch, expected commit, changed-file evidence, expiry, decision and outcome. An approval link can act for seven days; its record currently remains until the Forge account is deleted.

Public-page captures

Forge renders public pages on demand and returns screenshots in the tool response. It stores only your daily capture-attempt count; it does not keep a screenshot gallery or persistent copy of captured pages. Capture is restricted to operator-enabled accounts chosen by the operator.

Usage and analytics

Forge stores one daily capture-attempt count per user. Shared inference admission keeps anonymous daily attempt, completed, unknown-outcome and returned token totals, plus random reservations that expire after 60 seconds and are removed on the next admission. These records contain no user or repository identifiers, source or queries. Shape-only operational measurements are written to the existing Cloudflare logs as application events. These measurements contain no repository names, source, queries, captured URLs, user identifiers or tokens.

Cloudflare also processes automatic request logs separately from these measurements, including request paths, timestamps, IP addresses and browser information. The source configuration requests removal of URL query strings from logs and traces. Operators choose their log retention and check that redaction is enabled. Existing logs are kept under their retention policy.

What Forge does not keep

Who processes data

GitHub provides identity and repository operations. Cloudflare hosts the Worker, database and Browser Rendering. Semantic operations use the configured JEV inference route through Cloudflare. Relevant source sent to that route is external processing, not local-only analysis. This deployment permits bounded relevant private-repository source to be processed by the configured JEV route when you ask a semantic repository question. Exact source reads do not require inference. Forge keeps no persistent source index and does not sell personal data.

Your controls

For support or deletion, contact the operator who gave you access, using the private contact route they supplied. Do not send tokens, private repository details or captured-page contents.

Security and changes

Report Forge source vulnerabilities through GitHub’s private reporting form. Do not include live credentials or approval links.

Forge uses repository-scoped GitHub App access, encrypted storage for the one user credential, signed approval links, and explicit approval before Forge merges or discards a proposed change. This notice will be updated when the stored data, processors or retention behaviour changes.