Introduction
qaitai is open-source AI QA. Describe a journey in plain English; parallel workers test your app in real browsers and report bugs with repro steps and screenshots.
You write a journey, for example sign up, add a card, check out. qaitai starts one or more workers.
Each worker drives a real browser from a different angle (happy path, invalid input, navigation edge
cases, mobile and keyboard, boundary values) and reports findings: a severity, a title, a
description, repro steps and the screenshots it took. On a pull request, results show up as a GitHub
check.
Workers really use your app
Workers click, type and submit forms like a user. Point them at staging or a preview deploy, not production. Safe mode, when enabled, blocks actions it judges irreversible, such as payments and deletes.
Ways to run it
| What you need | Start here | |
|---|---|---|
| qaitai Cloud | A browser. Cloud runs test domains you have verified. | Cloud |
| CLI | Node.js 22+ and a model key. No account, runs on your machine. | CLI |
| Self-hosted, Docker Compose | A host with Docker and ~4 GB RAM | Docker Compose |
| Self-hosted, Kubernetes | A cluster, Postgres, Hatchet and S3 | Helm |
| Runner in your network | Docker next to the app under test | Remote runner |
Every option can use your own model key: see BYOK models. To run on pull requests, see GitHub PR checks.
Licensing
The control plane (apps/web) is AGPL-3.0. The runner, agent, CLI, database and job packages are MIT.
Enterprise features in ee/ use a commercial license, and the core works without them. See
LICENSING.md.