This release adds new AI model choices and a "thinking-level" control, more reliable web search, a reorganized Settings window with startup options, and multi-monitor support. It also lowers CPU usage, speeds up responses, and fixes a number of stability and speech issues.
New Features
- Kimi K2.6 is now the default AI model.
- Added Opus 4.8 (Anthropic) as a selectable AI model, best for very complex or long-running tasks.
- Added a "thinking-level" setting that lets you choose how much the AI reasons before answering. Off is fastest; higher levels think more deeply but are slower. Low is recommended for everyday use.
- LIMA now supports multi-monitor setups.
Changes
- Web search is now powered by Tavily, with an automatic fallback for more reliable answers.
- Models Llama 4 Maverick and GPT-5.3 have been retired.
- Mouse actions such as clicking have been refactored, bringing a 30% speedup.
- Faster responses and lower usage thanks to smarter reuse of context between requests.
- Refreshed the app's appearance to use the standard Windows look for better accessibility.
Bug Fix
- Fixed replies and screen narration being cut off before they finished.
- Fixed an issue causing web search answers responding with symbols.
- Fixed assistant messages occasionally appearing twice in the conversation.