Ship fast.
Break nothing.

Agents that find bugs before your users.

Get Started

Bring your own LLM · Open-source coming soon

0:00

Get running

One command to start.

bash

$ npx keryai

# Installs and opens the Kery dashboard in your browser

MCP server installed automatically. Use Kery directly from your editor or agent:

Claude

Cursor

Copilot

Codex

Coming soon

Kery Cloud

Everything you love about Kery, fully managed. No infra, no config — just tests that run.

CI/CD GitHub & Vercel

Integrations Linear, Slack & more

MCP Access Any agent, anywhere

Scheduled runs Cron-based monitoring

Regression detection Catch breaks on every deploy

Team workspace Shared runs & reports

The problem

Sound familiar?

Hover any card to see what Kery changes.

The problem

You shipped. You have no idea if the happy path still works.

With Kery

Kery crawls every route, runs every flow, and hands you a report with no test code written.

The problem

You spend more time fixing broken selectors than writing features.

With Kery

Kery navigates by accessibility tree and intent, not CSS selectors. Nothing to maintain.

The problem

Every test runner dies at the login page.

With Kery

Form login, Clerk, Supabase, and OAuth work natively. Kery authenticates as part of the run.

The problem

You click through the app before every release. An hour. You still miss things.

With Kery

Point Kery at a URL. It maps the app, picks the flows, files a structured bug report.

The problem

"Something looks off on checkout." No URL, no screenshot, no repro steps.

With Kery

Every finding: screenshot, bounding box, severity, and the exact step that triggered it.

The problem

You have no idea which routes have ever been exercised by a test.

With Kery

Route dashboard shows every URL: clean, issues, stale, or untested.

The problem

You shipped. You have no idea if the happy path still works.

With Kery

Kery crawls every route, runs every flow, and hands you a report with no test code written.

The problem

You spend more time fixing broken selectors than writing features.

With Kery

Kery navigates by accessibility tree and intent, not CSS selectors. Nothing to maintain.

The problem

Every test runner dies at the login page.

With Kery

Form login, Clerk, Supabase, and OAuth work natively. Kery authenticates as part of the run.

The problem

You click through the app before every release. An hour. You still miss things.

With Kery

Point Kery at a URL. It maps the app, picks the flows, files a structured bug report.

The problem

"Something looks off on checkout." No URL, no screenshot, no repro steps.

With Kery

Every finding: screenshot, bounding box, severity, and the exact step that triggered it.

The problem

You have no idea which routes have ever been exercised by a test.

With Kery

Route dashboard shows every URL: clean, issues, stale, or untested.

The problem

You shipped. You have no idea if the happy path still works.

With Kery

Kery crawls every route, runs every flow, and hands you a report with no test code written.

The problem

You spend more time fixing broken selectors than writing features.

With Kery

Kery navigates by accessibility tree and intent, not CSS selectors. Nothing to maintain.

The problem

Every test runner dies at the login page.

With Kery

Form login, Clerk, Supabase, and OAuth work natively. Kery authenticates as part of the run.

The problem

You click through the app before every release. An hour. You still miss things.

With Kery

Point Kery at a URL. It maps the app, picks the flows, files a structured bug report.

The problem

"Something looks off on checkout." No URL, no screenshot, no repro steps.

With Kery

Every finding: screenshot, bounding box, severity, and the exact step that triggered it.

The problem

You have no idea which routes have ever been exercised by a test.

With Kery

Route dashboard shows every URL: clean, issues, stale, or untested.

The problem

Writing automated tests is a separate skill. Your team doesn't have bandwidth.

With Kery

Describe what to test in plain English. Kery plans and executes the flow end-to-end.

The problem

The same false positive keeps appearing. Real bugs get buried in noise.

With Kery

Agent memory tracks known false positives and bug patterns, so findings get sharper over time.

The problem

Your pipeline goes green. Users still find visual breaks in prod.

With Kery

Review Agent and Filmstrip Reviewer catch visual, functional, and UX regressions, not just crashes.

The problem

Confusing copy, missing feedback states, and a11y gaps don't throw errors.

With Kery

Kery's UX bug category catches unclear flows, missing affordances, and accessibility gaps.

The problem

Bugs surface in production, in support tickets, in Slack from your CEO.

With Kery

Kery runs before you ship. 8 bug categories, 3 severity levels, evidence on every finding.

The problem

AI testing tools lock you into their cloud, their models, their pricing.

With Kery

Bring your own LLM: OpenAI, Anthropic, Gemini, or OpenRouter. Every agent role configurable.

The problem

Writing automated tests is a separate skill. Your team doesn't have bandwidth.

With Kery

Describe what to test in plain English. Kery plans and executes the flow end-to-end.

The problem

The same false positive keeps appearing. Real bugs get buried in noise.

With Kery

Agent memory tracks known false positives and bug patterns, so findings get sharper over time.

The problem

Your pipeline goes green. Users still find visual breaks in prod.

With Kery

Review Agent and Filmstrip Reviewer catch visual, functional, and UX regressions, not just crashes.

The problem

Confusing copy, missing feedback states, and a11y gaps don't throw errors.

With Kery

Kery's UX bug category catches unclear flows, missing affordances, and accessibility gaps.

The problem

Bugs surface in production, in support tickets, in Slack from your CEO.

With Kery

Kery runs before you ship. 8 bug categories, 3 severity levels, evidence on every finding.

The problem

AI testing tools lock you into their cloud, their models, their pricing.

With Kery

Bring your own LLM: OpenAI, Anthropic, Gemini, or OpenRouter. Every agent role configurable.

The problem

Writing automated tests is a separate skill. Your team doesn't have bandwidth.

With Kery

Describe what to test in plain English. Kery plans and executes the flow end-to-end.

The problem

The same false positive keeps appearing. Real bugs get buried in noise.

With Kery

Agent memory tracks known false positives and bug patterns, so findings get sharper over time.

The problem

Your pipeline goes green. Users still find visual breaks in prod.

With Kery

Review Agent and Filmstrip Reviewer catch visual, functional, and UX regressions, not just crashes.

The problem

Confusing copy, missing feedback states, and a11y gaps don't throw errors.

With Kery

Kery's UX bug category catches unclear flows, missing affordances, and accessibility gaps.

The problem

Bugs surface in production, in support tickets, in Slack from your CEO.

With Kery

Kery runs before you ship. 8 bug categories, 3 severity levels, evidence on every finding.

The problem

AI testing tools lock you into their cloud, their models, their pricing.

With Kery

Bring your own LLM: OpenAI, Anthropic, Gemini, or OpenRouter. Every agent role configurable.

Architecture

Four agents. One run.

Each agent has one job. Bugs from all sources are merged and deduplicated.

navigator

Drives a real Chromium session. 15 action types. Flags broken UI, failed submissions, and state drift in real-time.

reviewer

Post-run pass over the full flow. Catches functional gaps and state mismatches the navigator couldn't see mid-action.

filmstrip

Compares screenshots across the entire journey. Flags layout shifts, broken nav, and visual regressions.

memory_curator

Extracts navigation paths and bug patterns after each run. Proposes memory entries with confidence scores.

Bug Detection

Not pass/fail.
Actual bugs.

Screenshot, bounding box, step number, severity. Every finding is actionable.

8 bug categories · 3 severity levels

Functional

Buttons with no effect, forms that silently fail, broken state transitions, missing network calls.

Visual

Element overlap, layout shifts, broken responsive design, z-index stacking, missing images.

Dead click zones, missing loading states, empty states not handled, broken pagination.

Accessibility

No focus ring, missing aria labels, keyboard traps, interactive elements unreachable.

HTTP Errors

4xx / 5xx responses, CORS failures, auth token expiry, API timeouts surfaced silently.

Performance

Slow time-to-interactive, render-blocking requests, memory growth, excessive re-renders.

Data

Wrong values rendered, stale cache shown, missing fields, type errors surfacing in the UI.

Console

Uncaught exceptions, unhandled promise rejections, React warnings, silent errors logged.

Works behind auth

Your login page is not a wall.

Most testing tools stop at login. Kery authenticates natively, including multi-factor, so you can test real app flows behind real auth.

Supports form-based login, Clerk, and Supabase Auth.

Clerk

Supabase

Form Login

More providers coming soon

FAQ

Short answers.

How does Kery find bugs without any test scripts?

The Navigator agent drives a real Playwright browser using the accessibility tree and screenshots, not CSS selectors. It maps your app, plans flows, executes them, and reports bugs. Nothing to write, nothing to maintain.

What happens when the DOM changes?

Stagehand self-healing kicks in. When an element moves or the layout shifts, Kery finds it by intent rather than a stale selector. Tests don't break when your UI changes.

Which LLM providers are supported?

OpenRouter (recommended), OpenAI, Anthropic, and Google Gemini. Navigator, Review, Auxiliary, and Stagehand roles are each configurable independently so you can optimize cost and quality per task.

Does it work behind authentication?

Yes. Kery supports form-based login, Clerk, and Supabase Auth out of the box. Set credentials once in the dashboard and every run authenticates automatically.

Can I run Kery in CI?

Yes. Use the TypeScript client SDK or trigger runs via the REST API from any pipeline. Results come back as structured JSON you can parse, diff, or fail a build on.

What can I do with MCP?

Scan your app, run tests, pull bug reports, and browse route coverage directly from Claude, Cursor, or any MCP-compatible editor. Over 20 tools available including kery_scan, kery_get_bugs, kery_list_routes, and more.

Does Kery get smarter over time?

Yes. The memory curator agent learns successful navigation paths and records known false positives after each run. Confidence scores decay over time so memory stays fresh, not compounding.

Where does my data live?

Kery is fully self-hosted. Screenshots, run history, and bug reports live in your own PostgreSQL instance. Nothing leaves your environment except the prompts sent to your LLM provider.

Is Kery open-source?

Open-source is coming soon. Watch the GitHub repo for updates.

Join the community

Ask questions, share workflows, follow what's being built, and shape where Kery goes next.

Join Discord

Get started