Skip to content

Migrating from 0.3

0.4 is a ground-up rewrite. The SelfHealing library still works as a deprecation shim that routes to the new engine, but configuration moved from LLM_* to HEAL_* environment variables.

Library import

*** Settings ***
# still works (deprecated shim, emits a warning):
Library    SelfHealing

# preferred:
Library    Heal

Library Heal is the short, canonical form; Library heal.rf.HealListener is the equivalent fully-qualified import.

Configuration

0.3 0.4
LLM_API_KEY / LLM_API_BASE HEAL_API_KEY / HEAL_BASE_URL
LLM_TEXT_MODEL / LLM_LOCATOR_MODEL HEAL_MODEL / HEAL_LOCATOR_MODEL
LLM_VISION_MODEL HEAL_VISION_MODEL
heal_assertions=True (kwarg) HEAL_HEAL_ASSERTIONS=true
locator_db_file=... (kwarg) HEAL_HISTORY_DB=...
fix=, use_locator_db, collect_locator_info, use_llm_for_locator_proposals dropped — the run store and capability-resolved proposals replace them

See the full configuration reference for every HEAL_* setting.

What's new

The 0.4 engine adds failure triage beyond locators (timing, viewport, overlay, form, assertion), tiered locator selection, cross-run heal memory, a fix engine with visual diffs, an MCP server and CLI, and capability-tiered model support. Browse the Explanation section for the concepts.