v2.0

TinyAI v2.0

TinyAI 2.0 adds voice dictation and introduces remote agents and live conversation as optional beta features. It also improves rich-text handling, clipboard behavior, and the menu bar voice experience.

What's new

• Voice dictation: Hold or tap a shortcut to dictate into the active app. Choose local speech recognition models powered by transcribe.cpp, or OpenAI transcription. Local audio stays on the Mac.
• Live transcription feedback: A menu bar popup shows recording status, microphone level, recognized text, and errors. The menu keeps the five most recent transcripts for quick copying.
• Remote agents (Beta): Connect voice shortcuts to HTTP APIs or MCP servers, and display, paste, copy, or speak their responses. MCP connections support browser-based sign-in.
• Live conversation (Beta): Start a full-duplex voice conversation with GPT-Live-1. It can use configured MCP tools and agents on the Mac.
• Richer text handling: Improved preservation of lists, links, emphasis, and code through translation and custom actions, with fixes for clipboard and formatting edge cases.
• Safer beta controls: Agents and live conversation stay hidden until enabled from Help → Agents and live conversation (Beta).

Requirements

• macOS 14.6 or later on Apple silicon or Intel Macs.
• An OpenAI or Google Gemini API key for text actions. OpenAI voice services use the OpenAI key; local speech models can be downloaded separately in Settings → Voice.
• Allow Microphone, Accessibility, and Input Monitoring access when prompted for the features that need them.

Install

1. Download TinyAI.zip and open it. This build is unsigned.
2. Move TinyAI.app to Applications.
3. On first launch, macOS may show an unidentified developer warning. Control-click TinyAI.app and choose Open, or use System Settings → Privacy & Security → Open Anyway.

For setup and usage details, see the README.

Full Changelog: v1.54...v2.0