All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Telemetry: Agent now sends structured
agent_eventdata to PostHog on everyinvoke/ainvokecall, capturing query, steps, model, provider, success/failure, and error details. Runs silently with no console output. - Public API:
Agent,Browser,AgentEvent,EventType,BaseEventSubscriberand friends are now importable directly fromwindows_use—from windows_use import Agent, Browser. - CI/CD: GitHub Actions workflows for linting (ruff via pre-commit) and tests on Windows across Python 3.10–3.13, plus automated PyPI publishing on version tags.
- Tree traversal:
NoneType.strip()crashes when Windows UI Automation returnsNonefor element names — fixed with(value or '').strip()in 6 locations. - Event system:
ValueError: list.remove(x): x not in listwhen removing an event subscriber that was already removed or never added. - Telemetry user ID:
PermissionErroron systems whereTEMPpoints toC:\WINDOWS\TEMP— now usestempfile.gettempdir().
- Performance Optimization:
- Significant improvement in Tree Traversal speed (0.2-0.8s).
- Desktop State Capture optimized to (0.4-1.0s).
- Reduced computation expense by removing reliance on the root level children.
- Minimized COM calls to UIA3 using enhanced caching mechanisms.
- LLM Wrappers:
- Updated LLM wrappers for better reliability.
- Implemented Minimal Schema for Ollama to reduce token usage and improve stability.
- VDM: Fixes for Virtual Desktop Manager interaction.
- Windows 11 Support: Updated VDM to support Windows 11 Build 26100.4349.
- Ollama: Added schema sanitization for Ollama.
- Tool Schema: Updated tool schema generation to resolve understanding issues for some models.
- Vision: Minor fixes to the vision capabilities in LLM wrappers.
- Multi-screen Support: Enhanced capabilities to interact with multiple monitors.
- Desktop Tool: New tool for Windows 10/11 users to enable desktop switching, creation, and deletion.
- Annotation Support: Added
use_annotationparameter in the agent to allow requesting plain or annotated screenshots.
- Performance Optimization: Introduced caching to significantly speed up tree traversal.
- Resource Usage: Reduced usage of the UIA module for app retrieval to enhance desktop state speed.
- Tool Merging: Merged
drag_toolandmove_toolinto a single, unifiedmove_tool.
- Memory Leaks: Fixed memory leaks in the system to improve stability.