Claude Code builds
Codex builds
Cursor builds
Video proof on every pull request
See it work, before you merge.
No tests to write. Kery reads the diff and works out what the change was supposed to do.
Use case 01 · Pull request checks
A pull request opens; Kery checks it.
Kery reads the diff, works out what the change claims to do, runs those claims in a real browser on the preview deployment, and posts each verdict back to the PR — with video proof.
Tasks: create, edit, share, delete #128
Preview deployment ready — acme-dashboard-pr-128.vercel.app
Test my sign-up form
Use case 02 · Plain-English tests
You describe a flow; Kery tests it.
Group tests into suites
Bundle sign-up, checkout, and settings into one group and run it in one go.
Put runs on a schedule
Nightly or hourly passes over the flows you care about, results waiting for you.
Auto-scan your app
Kery crawls your routes, finds the flows worth covering, and suggests the tests.
Kery Cloud
Why teams trust the verdicts.
It gets past your login.
Clerk, Supabase, Auth0, Firebase, 2FA codes, magic links, Vercel preview protection. Most agents stop at the sign-in page. Kery signs in and keeps going.
Nothing to write.
No test scripts, no acceptance criteria, no ticket. Kery reads the diff and works out what the change was supposed to do.
Video of the moment it broke.
Every check comes with a recording, and “watch this moment” seeks straight to the deciding step.
It won’t block your merge on a maybe.
Only a contradicted check fails the build. Everything else gets reported, not enforced.
It gets quieter every run.
Kery remembers your app’s paths and conventions, and the findings you’ve called false positives. Fewer repeats each time.
It runs itself.
Install the GitHub App. Kery picks up every pull request and its preview deploy. No CI config, no YAML.
Self-serve
Connect a repo, get your next PR checked
Kery Cloud runs on your pull requests today. GitHub App, one-command connect, no API keys.
Why this matters
AI made shipping cheap. Checking is still expensive.
Claude Code, Codex, Cursor, and other agents can create more product changes than a team can manually click through. The risk is not that teams stop building. The risk is that they merge faster than they can verify real user flows.
The bugs that matter hide in real flows
Login, checkout, forms, permissions, empty states, and mobile layouts are where releases usually break. Kery walks those paths in a real browser.
FAQ
What teams usually ask first.
What is Kery?
Kery checks that your pull requests do what they say. It reads the diff, works out what the change claims to do, runs those flows in a real browser, and posts each check back to the PR with a verdict and video of what actually happened.
Do I need to write test scripts?
No. Kery works out what to check from the pull request itself. You can also point it at a flow in plain English, and useful flows can be saved and rerun.
Does it work behind authentication?
Yes. Kery supports logged-in products through configured environments and credentials, including form login, Clerk, and Supabase Auth. It is not limited to simple public sites.
Which LLM providers are supported?
Kery Cloud supports OpenAI, Anthropic, Google Gemini, and OpenRouter.
Can I run Kery in CI?
Kery Cloud checks your pull requests against their preview deployments. Connect the GitHub App and Kery handles execution for you.
What can I do with MCP?
Use Kery from MCP-compatible clients like Cursor, Claude Code, and Codex CLI to run browser checks on your app, inspect route coverage, and pull results.
Where does my data live?
Kery Cloud securely stores run history, screenshots, and reports in your account. See our privacy policy for details.
From the blog
All posts Latest notes and guides.
August 4, 2026
Evidence-First CI: Proving Every Pull Request Works on Its Preview Deployment
Green checks prove your scripts ran — not that your feature works. Kery's AI tester walks every PR preview, grades each claim against real browser evidence, and attaches video proof plus the bugs found along the way.
Perspective
The verification layer for your software factory
Agents industrialised production but not proof. Four layers of a delivery pipeline, why review and tests structurally cannot cover the third, and what the missing one has to be.
Setup
Claude Code + Playwright MCP for your PRs
A working setup: what to install, the four inputs it needs, the prompt that makes it verify rather than summarise, and the auth step that stops most attempts.
Community
Join the Kery community.
Ask setup questions, share MCP workflows, compare real browser runs, and follow what lands next.
Get started
Checking this PR… acme-dashboard-pr-128.vercel.app
4 verified · 1 contradicted — ▶ video proof
Deleted the task and the list updated. After reload it was back.