Talk to your Mac while the model works.
aistudio is the provisional codename for voice-first local computer use: audio in, audio out, screen vision, and macOS Accessibility in one loop.
Pre-launch. Phase 1 is stabilization: audio reliability, computer-use grounding, and tool-execution concurrency. Public naming is unresolved under the caletta labs parent frame.
What it does- Voice-first loop — speak to Gemini Live and hear the model answer while it reasons about the screen.
- Screen grounding — screenshot and vision context flow into the model instead of relying on text-only prompts.
- Native control — macOS Accessibility actions route through axmcp, so local apps are reachable.
- Terminal-native — runs from the terminal with the user's own Gemini API key.
One conversation carries voice, screen context, and local app control. Gemini Live supplies the model session; axmcp supplies native macOS actions. The integration is the product, and audio reliability, screen grounding, and concurrent tool execution are the work still being stabilized.