AI writes your code.
Decern makes sure it follows your rules.
A CI gate that evaluates every pull request against your architecture decisions using your own LLM. Violations of blocking ADRs fail the build. Everything else gets flagged, not blocked. New patterns get surfaced as signals for the tech lead to formalize.
Built for platform teams whose developers ship with Copilot, Cursor, and Claude Code.
$ npx decern gate
ADRs: found 4 approved in docs/adr
Changed files: 3
Evaluating ADR-007: Use PostgreSQL for persistence...
pass (92%)
Evaluating ADR-012: No direct DB access from handlers...
BLOCKED (87%) — handler queries DB directly, bypassing service layer
SIGNAL: New Redis caching layer detected
Gate: blocked. 1 new decision also detected.
No ADRs yet?
You don't have ADRs?
Decern extracts them from your codebase.
An ADR (Architecture Decision Record) is a short markdown file that captures one technical decision your team made: what you chose, why, and what the trade-offs are. It lives in your repo alongside the code it governs. Most teams have dozens of these decisions but never wrote them down which database to use, where business logic lives, which patterns are approved. Run `decern init` and Decern analyzes your directory structure, dependencies, git history, and entry points to propose 15-25 ADR drafts. You review each one, approve or skip, and commit the ones that matter. In 30 minutes you go from zero ADRs to an enforced architecture baseline.
Sound familiar?
AI codes faster than
your team can review
An AI agent introduces a new caching layer nobody discussed. Review misses it in an 800-line diff. It ships, and now it's a de facto architecture decision that was never decided.
You deprecated a pattern last quarter and wrote an ADR. Every new Copilot suggestion brings it back. Nobody checks the ADR before merging.
Three teams, three interpretations of the same architecture decision. AI amplifies the inconsistency at 10x the speed. The drift report shows it, but by then it's already in production.
This isn't a review problem. It's an enforcement problem. Decern fixes it in CI.
Why now
Review doesn't scale when AI writes
30-70% of your PRs.
Until 2024, architecture governance was a review problem the tech lead read every PR and caught the drift. Since AI coding became mainstream, it's a throughput problem. Copilot, Cursor, and Claude Code ship code fast, but they don't know what your team decided. The old way of catching drift in review stopped working when the volume tripled. Decern moves the check from the reviewer's head into CI, where it runs on every PR, for every developer, including AI agents.
Works with your CI and your LLM
How it works
Three phases. One CI gate.
Bootstrap
Run `decern init` in your repo. Decern analyzes your codebase and proposes 15-25 ADR drafts. You review, approve the ones that matter, and commit them to /docs/adr/. Done in 30 minutes.
CI gate
Add `decern gate` to your CI. Every PR gets evaluated against your ADRs. Violations on blocking ADRs fail the build. Warnings get logged. Confidence scores prevent false positives.
Evolution
Every PR is scanned for new patterns not covered by any ADR. The tech lead sees them as signals in the dashboard, generates a draft ADR with one click, and manages the full lifecycle (approve, promote, supersede) all via PRs on the repo.
ADRs in the repo are the source of truth.
The dashboard is the control plane. The gate is the enforcement.
Compliance
Audit-ready by design
Every gate run produces a hash-chained, Ed25519-signed evidence record: which ADRs were evaluated, each verdict with its confidence score, the diff hash, and who authored the change. Exportable, timestamped, cryptographically verifiable.
When your compliance team asks how you ensure human oversight on AI-generated code, you have an answer, and the evidence chain to prove it.
Designed for SOC 2 and ISO 42001 audit trails. Compatible with EU AI Act Art. 17 and NIST AI RMF.
Who it's for
One tool, three outcomes
Platform / DevEx
One CI step. Zero workflow changes. Architecture governance your leadership can measure.
Enforcement without friction, evidence when it's asked for.
Engineering Manager
AI is 10x-ing your team's output. Decern makes sure it follows the decisions your team already made.
Speed without architectural drift.
Tech Lead
Stop repeating architecture rules in every review. Write them as ADRs once, enforce them in CI. New patterns get surfaced as signals.
From architecture police to architecture reviewer.
Trust and control
Your LLM. Your repo.
Your rules.
Decern doesn't manage your LLM or your ADRs. Your repo is the source of truth. Your LLM runs the evaluation. Decern is the orchestration layer that connects them in CI.
Bring your own LLM
The gate calls the LLM you choose. OpenAI, Anthropic, or any OpenAI-compatible endpoint. Your API keys stay in your CI environment, never stored by Decern. Confidence scores and model warnings prevent false positives.
Repo is the source of truth
ADRs live in /docs/adr/ as markdown files. The gate reads them locally. The dashboard caches them for display. Every lifecycle change (approve, promote, supersede) creates a PR. Nothing enters the repo without a merge.
Evidence chain for audit
Every gate run produces a signed, hash-chained evidence record. Export bundles include records, signatures, and public keys. Verifiable offline. Retention configurable per workspace.
Enterprise deployment
Deploy in your own VPC
For regulated teams in finance, healthcare, defense, and public sector. Decern runs self-hosted: your code, your LLM, your infrastructure. Nothing leaves your network.
Why Decern
I built Decern after watching tech leads spend their Fridays writing the same review comment for the tenth time, while Copilot introduced patterns nobody had discussed. The code was shipping, but the architecture was drifting. Review was becoming a bottleneck, not a safety net. This tool is what I wish I'd had: write the decision once, enforce it in CI, and get back to building. Alessandro
Built in 2026. Looking for design partners.
Start free to try it, or get in touch if you want to shape the product.
AI doesn’t know
what your team decided.
Decern does.
Free forever. Full CI gate.
No credit card required.