tako-research/takoqa

TakoQA is a testing tool that uses an AI model to explore your web application the same way a real user would. Instead of writing scripts that click specific buttons in a fixed order, you describe what you want tested in plain language, and the tool drives a real Chrome browser to figure out how to do it. The AI reads the page visually, decides what to click or type next, and keeps going until it reaches the goal or runs out of steps. The tool is designed for finding bugs that are hard to catch with traditional automated tests: JavaScript errors that only appear during certain navigation paths, server errors triggered by unusual inputs, and confusing flows that technically work but would frustrate a real user. Every run produces a self-contained HTML report with screenshots, an embedded video recording, and a step-by-step replay of everything the AI did. You define your app's behavior in a YAML profile file. The profile says what the base URL is, how to log in, what kinds of things count as bugs (called invariants), and what goals to try (called missions). The AI figures out how to navigate to each goal on its own. This makes it straightforward to point the tool at a new application without writing new test scripts from scratch. A learning mode lets the tool build up knowledge about your app across multiple runs. It remembers which routes are gated behind login, which controls do nothing, and which missions it has already tried. Future runs use this knowledge to avoid re-testing the same dead ends. Facts that appear only once are treated as flukes and not retained, and facts that stop appearing eventually fade out. When the same issue appears repeatedly and you decide it is not actually a bug, you can mark it as muted with a reason. The tool passes that reason to the AI judge so it stops flagging the same thing in future runs. TakoQA is built in TypeScript, uses Playwright for browser control, and requires an Anthropic API key to function. The README does not specify a license.