FlowQA by TestFinch
End-to-end tests that write and prove themselves
Record a flow in a real browser, or point the exploring agent at your site and review the plan it proposes. Every generated test is replayed twice on fresh sessions and checked against a deliberately broken page before it is accepted. Then it runs on schedule, on deploy and on demand, with no model involved.
Mac and Windows. The web workbench runs tests on cloud browsers; recording a new test needs the desktop app.
01Explore
Map, plan, generate, prove
FlowQA crawls your site and shows the map. It proposes journeys with a goal and a success condition each; you keep, rename or drop them. For each journey it explores the site with a cheap decision model that picks the next element and a strong model for anything unsure, records the steps, writes the assertions, and verifies the test. In our published benchmark, the cheap model chose the recorded element in 97.5% of 163 decisions across ten sites, at a thirtieth of the strong model's cost.
- Accepted only when it can fail
- Twice green on fresh sessions, then the page is broken the way the test's own assertions describe. A test that still passes protects nothing and is rejected. This proves the test catches the break it was written for; it says nothing about the pages it never visits, which is why you review the plan.
- Asks when unsure
- Which element, what value, may it follow a host, is it done: the explorer files a question with a screenshot, parks the journey and carries on. On the desktop you can take over the browser and hand the steps back.
- Remembers the site
- Site notes hold what the map found, what journeys learned and what your team said. The planner and the explorer read them before every run.
- Repairs on change
- When a test fails on a changed page, FlowQA replays it to the failing step, proposes one small change, proves it twice, and waits for your approval.
02Run
On schedule, on deploy, on demand
Cron schedules in your timezone, a token your pipeline calls after a deploy, and a Run button in the workbench. Runs happen on hosted cloud browsers, or on your own machine for sites behind a VPN or on localhost, where they are free.
- Alerts where you work
- Email and Slack when a scheduled or triggered run fails, naming the failing tests and steps.
- Dedicated agents
- Browsers reserved for your workspace, with priority in the shared queue.
03For coding agents
An agent can run the covering tests before it says done
Over MCP an agent lists a project's tests, runs the ones that cover a snag, reads the result, and reads the site notes to understand a page the way your team does. The same account works in Snag, so the loop from bug to fix to proven fix never leaves the agent's session.
Download
FlowQA Desktop for Mac and Windows
The desktop app records tests in a real browser and runs them on your own machine for free, including sites behind a VPN or on localhost. Mac copies update themselves. FlowQA on the web needs nothing installed and runs tests on cloud browsers.