Getting started¶
In five minutes you'll watch heal repair a deliberately broken locator at runtime and produce a report. You need Python 3.10+, a Robot Framework Browser setup, and access to any OpenAI-compatible model endpoint.
1. Install¶
2. Configure a model¶
Create a .env in your project root (the nearest .env is loaded automatically
and overrides the environment):
Any OpenAI-compatible endpoint works — see Configure a model provider for vLLM, Ollama, MiniMax, Azure and others. Check it before running:
3. Add the listener¶
*** Settings ***
Library Browser timeout=3s
Library Heal
Suite Setup New Browser chromium headless=True
Suite Teardown Close Browser ALL
*** Test Cases ***
Login
New Page https://the-internet.herokuapp.com/login
# this id is wrong on purpose — heal will find the real one
Click id=does-not-exist
4. Run it¶
You'll see heal classify the failure, propose a verified locator, rerun the keyword, and pass the test:
heal: 'Click' failed (locator-drift) and was healed:
Replaced broken locator 'id=does-not-exist' with verified 'css=...'.
5. Read the report¶
Open results/heal/heal_report.html — the dashboard shows the healed failure,
the evidence, the cost, and a fix proposal with a side-by-side diff of the
suggested permanent change (your original file is untouched). See
What you get for a tour of the artifacts.
Next steps¶
- Configure a model provider for your backend
- Fix test files to turn heals into permanent changes
- Failure taxonomy to understand what heal can repair