What Is GuardSpine? The 60-Second Explanation
GuardSpine is an open-source governance layer for AI-assisted development. It captures what changed, classifies risk, routes review, and seals proof into tamper-evident evidence bundles. Here is how it works.
Your team uses AI to write code. Your auditor asks for proof that AI-generated changes were properly reviewed. You show them a GitHub approval. They ask what was actually checked. You have nothing.
GuardSpine fixes that.
The One-Line Version
GuardSpine is an open-source governance layer that captures AI-assisted changes, classifies their risk, routes them for appropriate review, and seals everything into cryptographically verifiable evidence bundles.
That is the elevator pitch. Here is what each piece means in practice.
What It Actually Does
When a developer opens a pull request in a GuardSpine-enabled repository, four things happen automatically:
Risk classification. GuardSpine analyzes the diff and assigns a risk tier from L0 (trivial — documentation, comments) to L3 (critical — authentication, payment processing, cryptographic operations). The classification uses both static analysis and AI assessment. A change to a README gets L0. A change to your OAuth middleware gets L3.
Review routing. The risk tier determines what happens next. L0 changes get fast-tracked with minimal review. L3 changes get routed to a multi-model AI council where Claude, GPT, and Gemini each independently assess the change against your security policies and coding standards. The review depth scales with the risk.
Evidence capture. Every step of this process is recorded. The exact diff at review time. The rules that were evaluated. Each reviewer’s assessment and score. The risk classification reasoning. The authorization decision. All of it.
Evidence sealing. The captured evidence is hashed into a Merkle chain and sealed into a tamper-evident bundle. Anyone can verify that bundle offline — no API calls, no license key, no trust in GuardSpine’s infrastructure required.
The output is a JSON file that proves what happened. Not a dashboard screenshot. Not a log entry. A cryptographic proof that a specific change was reviewed against specific rules by specific reviewers with specific outcomes.
Who It Is For
GuardSpine serves three audiences with different needs:
Engineering managers who adopted AI coding tools and now face audit questions they cannot answer. “How do you govern AI-generated code?” is showing up in SOC 2 questionnaires, ISO 27001 assessments, and FDA software validation protocols. GuardSpine provides the evidence package those audits demand.
Security teams who need to enforce review policies without blocking developer velocity. The guard lane system means trivial changes flow fast while risky changes get deep review. Security gets assurance. Developers keep shipping.
Compliance officers in regulated industries — biotech, finance, healthcare, defense — where “trust me, we reviewed it” was never acceptable and AI-generated code made the problem worse. Evidence bundles give them something concrete to hand to regulators.
The Four Guard Lanes
Every change flows through one of four lanes based on its risk classification:
L0 — Express Lane. Documentation, comments, test fixtures, configuration that does not affect behavior. Automated review with minimal overhead. Evidence still captured, but the process is fast.
L1 — Standard Lane. Business logic, UI changes, non-sensitive API modifications. Single-model AI review against your coding standards. Evidence bundle generated with full review trace.
L2 — Elevated Lane. Database schema changes, API authentication, data handling, third-party integrations. Multi-model review with consensus scoring. Higher approval thresholds. More detailed evidence capture.
L3 — Critical Lane. Payment processing, cryptographic operations, authentication flows, PII handling, infrastructure security. Full multi-model council review. Mandatory human sign-off. Complete evidence chain with every decision point documented.
The lanes are configurable. You define what files and patterns map to which risk tier. Your security policy becomes executable code, not a PDF that nobody reads.
The Open-Source Stack
GuardSpine follows an open-core model, and the split is deliberate. Everything related to trust is open source. Everything related to convenience is proprietary.
Open source (Apache 2.0):
- guardspine-spec — the evidence bundle schema
- guardspine-kernel (TypeScript) and guardspine-kernel-py (Python) — the core sealing engines
- guardspine-verify — offline verification CLI
- codeguard-action — GitHub Action for governed PRs (737 tests)
- guardspine-adapter-webhook — inbound event handling
- n8n-nodes-guardspine — workflow automation nodes
Proprietary:
- Enterprise dashboard and reporting
- Team management and RBAC
- Hosted multi-model council
- Compliance report generation
- SAML/SSO integration
The trust components are open because governance that requires trusting the vendor defeats the purpose. If you cannot inspect the code that seals your evidence, you are just swapping one trust problem for another.
How It Compares
You might be thinking: “How is this different from CodeRabbit or cubic.dev or any other AI code review tool?”
Those tools review code. GuardSpine governs the review process.
CodeRabbit tells you what is wrong with your code. GuardSpine proves that CodeRabbit (or Claude, or GPT, or your internal models) actually checked your code against your specific policies, and seals that proof into a package an auditor can verify without calling anyone.
Review tools and governance tools are complementary. GuardSpine does not replace your AI reviewer. It wraps your AI reviewer in an evidence trail.
Getting Started
The fastest path to governed PRs is the GitHub Action. Add codeguard-action to your CI workflow, configure your risk tiers, and every PR automatically gets classified, reviewed, and sealed. Ten minutes from zero to governed.
For teams that need more control, the TypeScript and Python kernels let you build governance into any pipeline. The spec is public. The verification is free. The evidence format is documented.
If you want to see GuardSpine running on your actual codebase — not a demo repo, your real code — book a call at cal.com/davidyoussef/guardspine. I will walk you through the setup and show you what your first evidence bundle looks like.