How it works
A clear path from code to an evidence-backed release.
Turn complex regulatory requirements into a practical, evidence-based engineering process. Five signals, one decision.
- 1
Register
Add your AI system and its intended use.
- 2
Assess
Determine risk and applicable obligations against a pinned corpus.
- 3
Provide evidence
Upload and link the documents each control needs.
- 4
Evaluate
Connect your eval results and thresholds.
- 5
Monitor data contracts
Detect schema and distribution drift.
- 6
Approve
Complete human sign-off workflows.
- 7
Release
Get a clear decision: PASS, REVIEW or BLOCKED.
AI System
Claims Triage AI
Risk
High
Evidence
5/5
Evaluations
0.92
Data contract
Healthy
Approvals
3/3
Ready for release
PASS
Illustrative example with sample data.
.github/workflows/ai-release-gate.yml
name: AI release gateon: [push, pull_request]permissions: contents: readjobs: gate: runs-on: ubuntu-latest steps: - uses: SVamseekar/devernance-release-gate@v1 with: api-key: ${{ secrets.ASSURANCE_API_KEY }} system-id: ${{ vars.ASSURANCE_SYSTEM_ID }} api-url: ${{ vars.ASSURANCE_URL }}Decisions
Three outcomes, each with its reasons.
Every decision lists the controls behind it, so the next step is obvious.
PASS
Exit code 0
Cited evidence, the eval, and contracts clear the bar, and no in-force WARNING, APPROVAL_REQUIRED, or BLOCKING control holds the release. The sealed pack can be exported.
REVIEW
Exit code 2
An in-force WARNING on an otherwise clear gate, or an unsigned APPROVAL_REQUIRED control, holds the release for a person. Owner, compliance, or legal still have to act.
BLOCKED
Exit code 1
A hard stop: an in-force BLOCKING control, an open contract BREACH, a prohibited class, or missing high-risk oversight. A missing approval timestamp stays inconclusive. The scan does not invent the field.
Before AI ships, prove it is ready.
Free plan, no card. Gate your first AI system in CI today, or explore the live demo workspace.