Voice
Talk to the chat instead of typing. Answers can be read aloud (with a speed control and a prompt that keeps spoken answers short), dictated text can submit itself, and with speech-popup 0.5.0+ installed a microphone button starts a hands-free conversation: you speak, the answer is read back, and the dictation window returns on its own. Demo
One engine shared with LLM Hub and Local LLM Hub
The chat UI, the turn lifecycle, Gemini streaming for both APIs, Vault tools, MCP, RAG, workflows, skills and settings now come from one shared package instead of three copies. Where the same feature existed in three generations, the newest one won, so behaviour no longer differs between the plugins and a fix reaches all of them at once.
Fixes
Follow-up questions can see what tools did earlier; tool limits, web sources and usage reporting behave the same on Interactions and GenerateContent; every RAG search keeps its sources instead of only the last one; skill workflows stay inside the folders they are allowed to touch; MCP app windows move and stay in sync; and a multi-window vault opens menus in the window you are actually in.