# Developer governance · Neural Inverse Govern > A merge gate for pull requests that AI coding tools write, control of CI/CD runs, and governance of the dependencies, tests, documentation, APIs and infrastructure code AI produces. Source: https://neuralinverse.com/govern/developer-governance/ --- ![](/photos/bench.jpg) [Neural Inverse Govern / Developer governance](https://neuralinverse.com/govern/) # AI-written changes pass your review rules before they merge. A merge gate for pull requests that AI coding tools write, control of CI/CD runs, and governance of the dependencies, tests, documentation, APIs and infrastructure code AI produces. Source control GitHub and GitLab, SaaS or self-managed Gate A required check on the head commit Attribution Claude Code, Copilot, Cursor, Codex, Devin and more Approvals Never by the requester Record Every gate decision audited Overview Govern receives signed webhooks from GitHub and GitLab, works out which AI tool wrote each pull request, evaluates your review rules and publishes the result on the head commit. Branch protection that requires the check then blocks the merge until the gate passes. Around it, CI/CD runs, new dependencies, AI-written tests and documentation, API specifications, refactorings and infrastructure code each have their own policy, and an approval is never given by the person who asked. [What it does ↓](https://neuralinverse.com/#capabilities)[How it works ↓](https://neuralinverse.com/#workflow)[Enforced and recorded ↓](https://neuralinverse.com/#enforcement)[Where it runs ↓](https://neuralinverse.com/#where)[Get started ↓](https://neuralinverse.com/#get-started) What it does ## What developer governance covers. Only what works today. - **AI pull request rules:** Rules by repository pattern, for AI-authored changes or for every change: blocked tools, a minimum review score, a person other than the author after the AI, a cap on AI-written lines and DLP on the description. The most specific rule applies. - **Review scores from registered sources:** A required score must come from a review source an administrator registered, for the current head commit. A member's own session cannot post one. - **Signed webhooks:** GitHub, GitHub Enterprise Server, GitLab and GitLab self-managed deliver signed webhooks. A bad signature or an unknown connection is refused before the body is read, and a delivery cannot be replayed. - **AI attribution:** Bot authors, commit trailers, author emails, labels, description markers and branch prefixes identify Claude Code, Copilot, Cursor, Codex, Devin, Aider and others. Attribution stays once seen, even if a marker is removed. - **Result on the commit:** The gate publishes a Check Run through a GitHub App, or a commit status on GitHub or GitLab, so branch protection or merge checks can require it. - **Source control governance:** Commits, pull requests, changelogs, releases and versions are validated with secret and blocklist scanning, and push and pull request events are recorded from signed deliveries. - **CI/CD control:** Register CI platforms with policies, hold pipeline runs for approval or rejection, scan runs with DLP and post commit statuses. Source control tokens are kept in the vault. - **DevOps actions:** AI-proposed deployments, rollbacks, environment changes and CI configuration are gated, with Dockerfile scans and remediation actions. - **Dependencies:** Organisation dependency policies, and requests for new dependencies that someone other than the requester approves or rejects. - **Tests:** Test suites with coverage and mutation trends, assertion analysis, organisation thresholds and quality gates. AI-written tests are evaluated, with DLP on their test data. - **Documentation:** AI-written documentation is scanned for leaked sensitive data and for references to code that does not exist, under an organisation policy. - **API specifications:** A registry of API specifications with analysis for personal data and authentication, version comparison, and review that the author cannot approve. - **Refactorings:** Refactoring sessions with impact analysis, an architecture gate, staged rollout plans and rollback, approved by someone other than the requester. - **Code search, environments and debugging:** Approved code search tools with DLP on queries and results, named environments with rules for actions in each, and AI debugger sessions with sanitised output. - **Infrastructure as code:** IaC policies, analysis sessions with findings, an organisation cost budget with a cost gate, and a check for privilege escalation. - **Database changes:** Queries analysed against database policies, migrations checked for safety, query cost tracked, and approved query templates. How it works ## How an AI-written pull request reaches the merge button. 1. Open An AI coding tool, or a person using one, opens a pull request. 2. Receive GitHub or GitLab sends a signed webhook; Govern verifies it before reading it. 3. Attribute Govern identifies the tool that wrote the change from the strongest signal. 4. Evaluate The most specific rule is applied: tools, score, DLP and AI line count. 5. Review A person other than the author approves the current head commit. A push after the approval needs a new one. 6. Publish The result is posted on the commit; branch protection blocks the merge until it passes. Who uses it ## Built for every role that answers for AI. - **Engineering leads:** Write review rules for AI-authored changes and decide which repositories require the gate. - **Platform and DevOps engineers:** Connect source control and CI platforms, and approve held runs and DevOps actions. - **Security teams:** Register review sources, set DLP on descriptions and runs, and approve new dependencies. - **Engineers:** See on the pull request why the gate is pending or failed, and what it needs to pass. Enforced and recorded ## Fails closed at the gate, recorded with every delivery. The gate decides on the server and blocks at the source control platform where your branch protection requires it. Enforced - **Fails closed:** A missing score from a registered source, or a missing approval by someone other than the author, leaves the gate pending. Blocked tools, DLP findings, a low score or too many AI lines fail it. - **Signed deliveries only:** Unsigned, mis-signed and replayed deliveries are refused, and source control events are recorded only from signed deliveries. - **No self-approval:** CI/CD runs, DevOps actions, dependencies, API specifications and refactorings cannot be approved by the person who asked. - **Secrets in the vault:** Webhook secrets and source control credentials are held in the vault; a webhook secret is shown once. Recorded - **Gate:** Every received pull request and push, every change of gate state and every result posted, or the failure to post it. - **Rules and sources:** Every rule change, review source registration and revocation, and every submitted score. - **Connections:** Every source control connection change and webhook secret rotation. The gate blocks a merge only where branch protection or merge checks require its check. AI attribution is detection, not proof: a rule that applies to every change cannot be evaded by removing markers. Where it runs ## Next to your source control, wherever it runs. [Deployment options](https://neuralinverse.com/deployment/) - **Cloud source control:** GitHub.com and GitLab.com deliver webhooks to your installation's receiver. - **Self-managed source control:** GitHub Enterprise Server and GitLab self-managed on your network. Calls to private addresses are allowed by your operator. - **Self-hosted and in your cloud:** Rules, results and credentials stay on your installation. - **Hosted by us:** We run it for you, with the same controls and audit record. Get started ## Set it up with the documentation. - [Neural Inverse Govern](https://neuralinverse.com/docs/govern/) - [Agent policy](https://neuralinverse.com/docs/govern/agent-policy/) - [Approvals](https://neuralinverse.com/docs/govern/approvals/) What's next ## Being built now. These are in development and not part of the product today. - **Bitbucket:** The AI pull request gate for Bitbucket repositories. - **Review bot registry:** Registered review bots with their own policies, action approval and merge protection rules. - **Documentation drift:** A measure of how far documentation has drifted from the code it describes. - **Infrastructure deployments:** IaC module deploys, Kubernetes cluster and ingress configuration, and platform deployments applied to real targets with rollback. - **Database registry:** Health checks and backups of the databases you register. Related [Agent policy and approvals → Allow, ask and deny rules for coding agents, with named approvers.](https://neuralinverse.com/govern/agent-policy/)[Third-party AI tools → The control level Govern has over each third-party AI coding tool.](https://neuralinverse.com/govern/third-party-tools/)[Audit and evidence → A hash-chained audit log you can verify offline, and evidence toward your frameworks.](https://neuralinverse.com/govern/audit/) ![](/photos/team.jpg) ## Talk to our team. Neural Inverse is available by request. We work with your engineering, security and procurement teams from the first conversation. [Request access](https://neuralinverse.com/contact/?topic=access)[Contact us](https://neuralinverse.com/contact/)