Stop your agents from repeating the same mistakes, enforce your team’s quality standards, and catch runtime backend bugs and security issues in every agent session.
npm install -g @kerno/clikerno auditAI adoption: Feb 2026 · Figures read Period A → Period B. Green moved the better way, amber the worse way.
was 71 in 1 Sept 2025 – 31 Jan 2026
Healthy Improving · +201.6 by May 2027 at the current rate (−0.71 a month; 95% range −3.9–7.1).
200.0 confirmed as duplicates or reimplementations, from 550 candidate pairs.
Code rewritten within 90 days. Extensions are iteration; repairs are waste.
| Date | Sha | Author | Subject | Lines | AI | Complexity Δ |
|---|---|---|---|---|---|---|
| 5 Nov 2026 | 10e9ff1 | dependabot[bot] | deps: bump amqplib from 2.2.0 to 2.2.1 | +10 −10 | Unknown | — |
| 5 Nov 2026 | 3006f97 | Konstantin | [BE-3317] Port the remaining MCP suite fixes | +388 −195 | Unknown | +7 |
| 5 Nov 2026 | 1fceddb | Diogo | BE-3695: report a Sentry release | +44 −0 | AI | 0 |
| 5 Nov 2026 | 10c4202 | Maximiliano | BE-3369: report a cancelled re-check as cancelled | +213 −10 | AI | +3 |
| 4 Nov 2026 | b245108 | Konstantin | fix(BE-2871): keep endpoint-test preconditions read-only | +296 −27 | AI | +3 |
| 4 Nov 2026 | 34dd2f7 | Konstantin | BE-3110: report a cancelled run as cancelled | +87 −10 | AI | +2 |
| Repository | Detected adoption | Quality score | PR size | Rework | Complexity / 1k | Human review | Median time to merge |
|---|---|---|---|---|---|---|---|
| kerno | Feb 2026 | 71 → 90 Healthy | 63 → 299 | 49% → 57% | 10.4 → 6.0 | 55% → 24% | 35m → 51m |
| kerno-backend | Feb 2026 | 69 → 89 Healthy | 18 → 186 | 34% → 27% | 1.3 → 0.5 | 38% → 10% | 16m → 23m |
| kerno-frontend | Feb 2026 | 74 → 72 Watch | 70 → 149 | 58% → 63% | 5.5 → 11.4 | 69% → 8% | 2.1h → 23m |
agent/refunds · Own worktree, ports and database · Claude Code · started 4m ago
| Result | Change | Stopped at | Duration | When |
|---|---|---|---|---|
| Passed | Keep partial refunds partial b682d434 | — | 4m 12s | now |
| Failed | Add partial refunds to POST /refunds 2476f16f | browser-e2e | 3m 40s | 12m ago |
| Passed | Refund webhooks for the ledger 0d2b5e10 | — | 4m 05s | 1d ago |
| Failed | Cap refunds at the amount paid 5e0f8b19 | unit-tests | 1m 12s | 2d ago |
| Passed | Retry failed refund webhooks 9c41e7a2 | — | 3m 58s | 2d ago |
Every run on your code, and what it found.
| Kind | Flow / Entry point | Result | Run type |
|---|---|---|---|
| Flow | A customer refunds part of a paid order | 1 diff 7 of 8 tests unchanged | Validation |
| HTTP | POST /refunds | Passed 6 of 6 tests unchanged | Validation |
| HTTP | GET /orders/{id} | Passed 4 of 4 tests unchanged | Validation |
| MCP | refund_order | Passed 5 of 5 tests unchanged | Validation |
| HTTP | GET /refunds/{id} | Passed 4 of 4 tests unchanged | Validation |
| HTTP | POST /orders/{id}/cancel | Passed 7 of 7 tests unchanged | Validation |
| Flow | An admin approves a large refund | Passed 5 of 5 tests unchanged | Validation |
Agents and engineers at these companies use Kerno
Audit your codebase for the complexity and mistakes AI introduces, and feed the lessons back to your team and agents.
It’s hard to tell whether AI made your code better or worse, or where.
See where quality slipped after AI adoption, in complexity, rework and test coverage.
Agents repeat the same mistakes, and every round burns tokens and engineering time.
See which code keeps getting reworked and why. Every lesson reaches your agents through MCP, so the mistakes stop coming back.
Teams change how they work with agents, and it’s hard to tell whether it helped.
Run it after each change and watch your quality score, reliability and test coverage move.
was 71 in Period A
Enforce your engineering policies, security controls and compliance requirements on every change, from one agent to a whole fleet.
Each agent and engineer runs a different set of checks, so quality varies from change to change.
Every agent works in its own isolated worktree and passes your deterministic gates, the same way every time, before it opens a pull request.
Reviewers mostly rely on the agent’s own summary of what it did.
Every run keeps verifiable artifacts that show the agent did the work. When something goes wrong, full observability traces every agent, MCP and tool call, so you see what happened and what to fix.
AI spend is hard to pin to a specific engineer, session or ticket, so you estimate where it went and guess where to focus.
See the cost of every ticket and task, so you know where the budget goes, what it delivered and where to focus next.
Test your critical flows and your entire backend surface against your real stack after every change, and keep every test up to date as your code evolves.
Unit tests pass on mocks, runtime bugs slip through CI into production, and fixing them there means expensive rework.
Every backend flow a change touches runs across your APIs, services and real dependencies, and you see the exact step that broke.
Agents change your backend faster than reviewers can check who can access what and how it fails.
Your entire backend surface, from APIs and queue consumers to MCP tools, is tested for authorization, OWASP Top 10 issues, malformed input and error handling.
Test suites fall behind as agents change the code, and coverage slowly erodes.
Tests self-heal as your product changes, and flows are regenerated when behavior changes on purpose, so coverage keeps pace with every pull request.
Customer B asks for the receipt of customer A’s refund.
Suspected bug. Kerno flagged this when it implemented the test, and matched the test to what the code did at the time.
Investigate with your agentCritical flows, end to end
Every route, contract and edge case
Tools, schemas and permissions
Run Kerno in your own cloud, keep your code inside your network, and sign in through the identity provider you already use.
Self-host Kerno or deploy it in your own cloud account or VPC, so the agent, the gateway and your run metadata stay inside your infrastructure.
The agent runs on your machines and indexes your repository locally. Only the code a task needs reaches the model, through Kerno’s gateway.
Your private code is never stored and never used for training. Every model provider we use is under a zero-data-retention agreement.
Data is encrypted in transit with TLS 1.2 and at rest with AES 256-bit encryption.
Sign in with SAML through your own identity provider, with Owner, Admin and Member roles.
Every run records who started it, what it checked and the exact code it validated.
AI is helping developers ship more code, but it is also accelerating complexity that compounds fast. Left unchecked, that complexity will build up silently, in code that is becoming slower and riskier to change.
npm install -g @kerno/clikerno audit