What's new
⚡️ Add llama.cpp, with options to recompile it with any backend
📖 Implement a new model library for ollama / llama.cpp
🔎 Implement hybrid search, improving document reading
💻 Add command execution tool
🗂 Add tool groups
🔗 Improve MCP server adding, supporting also STDIO for non flatpak
📝 Add semantic memory handler
📤 Add ability to import/export chats
📁 Add custom folders to the RAG index
ℹ️ Improved message information menu, showing the token count and token speed
Minor Improvements:
- Newelle is now easier to package for non flatpak
- Switch to Model2Vec as default embedding
- Add a spinner while the model is loading
- Improve application responsiveness
- Implement lazy loading for messages
- Option to hide history on launch
- Add search in file explorer
- Switch to ddgs for websearch
- Add tools to show video/images
- Better message generation stopping
- Add more options to the websearch tool
- Add keyboard shortcut to stop message generation
- Add tool settings to profiles
- Add an option for parallel tool execution
- You can now delete and edit console/tool messages
Fixes:
- Fix issues with openai handlers as secondary llma
- Fix ddgs not streaming
- Fix some random crashes
- Fix issues with automatic STT