Skip to content

Security and trust

Production is never attacked.

Every active test runs on a verified private copy with test keys only. These are the rules the system is built around, not settings you have to find.

  • Production is never attacked

    Every active test runs on a verified private copy. Production only ever gets passive, read-only checks.

  • Production keys never reach an agent

    Test-tier keys only: database branches, Stripe test mode, development auth instances. Production keys are never requested or stored.

  • Isolation is checked first

    Before any active test, the agent confirms the copy talks to a test database, not production. Until then, a project gets passive checks only.

  • Your code is never training data

    Model providers are used under zero-data-retention terms where available. Code is held only while a run needs it.

  • Untrusted content stays untrusted

    Pull request bodies, issues, code comments and page content are treated as untrusted input. Testing agents get no GitHub token and no repo write access; fork PRs do not run by default.

  • Secrets are scrubbed

    Secrets live in a managed store, are injected at job start, and are scrubbed from logs, traces, screenshots and videos.

  • Reports say what they are

    A launch report is evidence of what was tested and proven for one version of your product. It is not a certification.

  • Clean exit

    One click removes the config files through a pull request, restores each host's original deploy settings, revokes every token, and deletes stored data within 7 days.

Data

What we read, send and store.

The plain-English version. The privacy policy has the formal one.

We read

  • Repository code and diffs for the repos you select
  • Database schema, roles and policies (never row contents in production)
  • Deploy status and preview URLs from your hosting provider
  • Pages and responses of the private test copy during a run

We send

  • Diffs, schema summaries and page content to Anthropic under zero-data-retention terms, to drive the review and testing agents
  • Transactional email through Resend
  • Payment details to Stripe; Stableship never stores card numbers

We store

  • Your rulebook, project map, roles and compiled tests
  • Findings, verdicts and the audit log
  • Traces, videos and logs as evidence, 30 days by default
  • Connection tokens in an encrypted vault, read only by the orchestrator
Scan policyPrivacy policy

A data processing agreement is available on request and standard for teams.