MonkeyTest CoreAn AI agent that uses your site like a real user — and reports the bugs.

Open-source. Playwright + your favourite LLM. Zero glue code. Built for devs who'd rather ship than write yet another Selenium script.

Get started in 60 seconds

How it works

GitHub

An open library of intelligence with a stylised AI monkey agent inspecting a webpage and pinning a bug

🧰

Drop-in, not framework

One npm package, one constructor. No Selenium grids, no test DSL, no YAML — just `await monkey.executeAndTriage(url)`.

🧠

Bring your own LLM

OpenAI, Gemini, Anthropic, Mistral, Groq, or any OpenAI-compatible endpoint (Ollama, vLLM, OpenRouter). Pick a different model per phase.

📸

Real evidence, not log dumps

Per-step screenshots, optional WebM video, console + network failures, and a deduplicated bug report with stable fingerprints.

📡

Live activity stream

Subscribe to an EventBus and pipe explorer / planner / runner / triager events into your own dashboard or CLI.

🔬

Predictable, hackable, typed

Pure TypeScript, zod-validated LLM output, structured `RunResult` you can serialise, diff, and store. No magic.

🤝

AGPL & community-first

Free for everyone to use, fork, and self-host. The Affero clause keeps modifications shareable when run as a service.

Why MonkeyTest Core?

Most QA tools ask you to write tests. MonkeyTest Core writes its own — by looking at your app first.

It opens a real browser, crawls a few pages, asks an LLM "what would a real user try here?", runs those flows step by step, and hands you back a structured report with screenshots, bug fingerprints, and a live event stream you can wire into your own dashboard.

import { Monkey, autoLLMConfig } from '@mtai/monkeytest-core';

const monkey = new Monkey({ llm: autoLLMConfig({ quality: 'balanced' }) });

const { run } = await monkey.executeAndTriage('https://example.com', {
  explore: { maxPages: 8, screenshotsDir: '.monkeytest/shots' },
  plan:    { maxFlows: 4 },
  run:     { screenshotsDir: '.monkeytest/shots' },
});

console.log(`${run.bugs.length} bug(s) found across ${run.stats.flowsTotal} flows`);

That's it. No fixtures, no page objects, no recorders.

Where it shines

Pre-release sanity checks — point it at a staging URL and let it find the obvious "the form doesn't submit" / "checkout 500s" issues before a human does.
PR previews — run on Vercel/Netlify preview URLs in CI, fail the check if a flow regresses.
Daily smoke runs against prod — cheap LLM tier, screenshots stored as artefacts.
Embedding in your own product — the engine is a library, not a CLI. Wrap it in your own SaaS, internal tool, or VS Code extension.

What it isn't

Not a replacement for deterministic critical-path E2E tests — it's a complement.
Not a load tester or a11y auditor (those exist; pair them).
Not magic — quality of bug reports tracks the quality of the LLM you give it.

Don't want to run it yourself?

MonkeyTest Cloud is the same engine, fully managed, with hosted browsers, history, dashboards, and team accounts. Use the OSS engine for hacking and self-hosting; use Cloud when you want it to "just run".

Get started →