CabraFlow
Demonstrate your business, the test writes itself.
UI testing tools always cost you somewhere. Code-first frameworks are for developers only. Classic record & playback captures dead selectors that break on the first screen change. And XPath-riddled tests spend more time being repaired than catching bugs. The common thread: the test captures the technical how (a selector) instead of the business what (an intention).
The essentials
- What: a keyword-driven test orchestrator on top of Playwright, designed so a business profile creates a test as naturally as a developer.
- The language: business micro-phrases created on the fly with the words of the person demonstrating ("Approve the urgent order"), never a tool vocabulary, never code.
- The targets: resolved through the accessibility tree (role and accessible name, the way a user or a screen reader sees the page), no CSS, no XPath. A local agent proposes the repair when the UI changes.
- The capture: a live demo narrated out loud is enough: you navigate, you explain, the flow writes itself before your eyes. Transcription and reconciliation run locally.
- Replayable everywhere: the same test runs on the native runner, exports to pure Playwright or Robot Framework, and documents itself as an ISTQB test case.
- Status: an AI Cabra R&D program at the design stage; nothing is announced that is not executable and measured.
How it works
You open your application in the driven browser, turn on the microphone, and demonstrate your business while narrating: "I log in as an administrator, I fill in the form, I approve the order". Two probes work silently: one captures every gesture with its accessibility role, the other transcribes your voice locally. A reconciliation engine aligns the two streams: the voice provides the intention and the keyword name, the gesture provides the technical target. When the demo ends, the test exists: named in your words, executable, robust.
Three ways to create a test, one artifact
- The business user gives a narrated live demo: keyword blocks appear in real time while they navigate.
- The techno-functional profile assembles and refines keywords in a graphical studio: block composition, typed parameters, instant trial of a single keyword.
- The developer edits the text file (YAML) directly in their editor, with autocompletion, and can view the generated Playwright code at will.
All three doors lead to the same file, versionable in git like code. And AI agents have their own door: capture, replay and repair are exposed through MCP, the open standard our tools already use.
Neither a fork nor a reinvention
CabraFlow forks neither Robot Framework nor Playwright: it borrows the ideas, not the code. From Robot Framework, the keyword-driven principle and keyword composition. From Playwright, the execution engine and the user-facing locators (role and accessible name). The difference: the visible vocabulary is neither the tool's nor TypeScript, it is the words of your business. And nobody is locked in: a Robot Framework team replays the flows at home through the .robot export, a Playwright team through the .spec.ts export.
Where the program stands
CabraFlow is an R&D program at the design stage, run the AI Cabra way: architecture and proof criteria first, executable next, communication last. We will publish executable demonstrations and measurements, not promises. To discuss it for your QA team, write to us.
Frequently asked questions about CabraFlow are gathered on the site's FAQ page.
Let's talk about your test suite
Tell us where you stand. You will get an opinion, not a brochure.