AI pull request safety gate

Stop AI-generated PRs from outrunning review.

Busleyden Guard posts a GitHub check that shows what changed, who should review it, and what evidence is missing. For supported Python authentication changes, it also breaks the changed logic on purpose and verifies that the proposed regression test catches the defect.

Try it free on a public repo Preview a public PR first

Free on public repos
Private Pro: unavailable during the evidence beta
No sales call

No install required

Paste a public PR. See what proof review may be missing.

Guard reads GitHub's public changed-file evidence and returns a bounded consequence preview. It does not execute code or pretend a preview is a complete repository check.

The review bottleneck

More code is cheap. Knowing whether it is safe is not.

AI can create a plausible change in minutes. Senior engineers still have to reconstruct its blast radius, find the right owners, and decide whether the attached proof is enough.

Reviewer time: Senior engineers manually trace unfamiliar changes.
Release delay: Missing tests and owners appear late in the merge queue.
Escaped defects: Small diffs can reach authentication, data, and dependencies.
Audit exposure: The reason a risky change merged is often not recorded.

Live counterfactual GitHub proof

It did not merely find a test. It proved the test could catch the defect.

On a public authentication PR, Guard first blocked the change because token-replay evidence was absent. After the test was added, Guard ran it on the PR, removed the new replay protection, and confirmed that the same test failed against the broken counterfactual.

Live GitHub App run / public PR #3

Reject reused authentication tokens

1 proof required

Changed surface Authentication session handling

Verified owner @AlexanderLuzDH approved

Mapped test tests/auth/test_session.py

The only missing proof

Prove that a replayed or reused token is rejected.

Evidence satisfied

The replay test passed on the PR and failed when Guard removed the replay condition. The next hosted run passed with zero blockers.

Open blocking Check Open causal Check

From install to merge decision

Set it up once. Let every PR explain itself.

Install on one repository

Connect the GitHub App and approve the setup pull request.

Open a normal pull request

Guard reads the diff, ownership, tests, and scanners, then counterfactually checks supported evidence.

Act on one clear check

Merge, route to the right reviewer, or wait for the exact missing proof.

Inspect before you trust

Do not trust the pitch. Inspect the decision trail.

The public proof includes both content-addressed certificates, both GitHub Checks, the exact code change, the test, and the targeted mutation that the test killed. Guard must earn trust the same way on your repository.

Before evidence: Certificate with one blocker
After evidence: Certificate with zero blockers
Public source trail: PR, commits, review, Checks, and workflow
What it does not claim: Guard does not prove that code is bug-free or replace human review.

Designed for a narrow trust boundary

Your source stays in your GitHub workflow.

Start with one selected repository. The analysis runs through GitHub Actions, and customer source code is not stored by Busleyden Guard by default.

Trust center Privacy Terms Data processing Security review Data handling Install guide

Repository access: Only repositories you select during installation.
Source storage: No customer source-code storage by default.
Removal: Uninstall the App and remove the workflow at any time.

Free research beta · paid checkout closed

Pressure-test Guard on a public Python repository.

The product is available free for public repositories while its usefulness is measured independently. No payment is collected, and there is no automatic conversion. Paid launch remains blocked until the public evidence gate passes.

GitHub Check Run with merge guidance
Reviewable setup pull request
No sales call or onboarding meeting required
Asynchronous support with no support SLA
Self-service installation and removal

Public repository

Free

For public repositories. No card and no reduced product.

Includes: Installed App, setup PR, and Guard Check
Scope: Add one public repository per install
Billing: $0, with no automatic renewal

Next: GitHub verifies visibility, then you review the setup pull request before merging it.

Guard Pro · not for sale

Evidence-gated

Private-repository billing will not open until independent maintainers demonstrate that Guard improves real review decisions.

Includes: Installed App, setup PR, and Guard Check
Scope: Up to 5 private repositories
Usage: 500 Guard runs per billing month; no automatic overages
Support: Asynchronous support; no calls or SLA
Billing: Closed during the public evidence beta

Paid checkout is closed. No payment can be created while the evidence gate is incomplete.

Read the thresholds and current evidence.

Five Python teams wanted

Pressure-test Guard on real pull requests for two weeks.

Install it on one public repository and mark each result Useful, Noisy, or Missed something. No card, contract, sales call, meeting, or automatic conversion.

Install free on a public repo Enroll without a call