- MacOS Tahoe updates

- Added support for open-mic conversation. Just keep the mic open at all times - Fixes that allow the AI to edit your memories - Fixes to audio processing / streaming - Allow removal of unused whisper models - Redesigned the triggers screen

- New action to allow your AI to make memories (and forget them) - Better cancellation support with Escape - (Beta) Support for a native Safari extension - Added actions for the AI to stop/sleep - Support for generating PopClip snippets - Added support for instructions (system prompts) for Ask AI actions - Added support for 'plugins' (name may change) that combine actions, system prompts, and browser scripts - Added a Text Processor action to allow for super fast transcription (skip AI) or use template logic in creative ways - Front-end performance improvements

- AI now shows what it is doing now using icons - Added action gallery - Better UX for new users - All examples now have icons - Fixed Reminder adding - Fixed audio sometimes cutting out - Added required/optional variables - Added custom waiting times for responses

- Support for the Realtime API (Low Latency, Instant Replies)

- Support for Gemini - Support for Azure - Fix for AI stopping and seeing 'no new content', when only image data was sent back - Better advanced settings - Force Whisper language - Moved logging to general settings - Hotkeys → Trigger - Defaults for AI providers auto-update (when left blank) - I recommend for previous users, to remove the default model / base urls, unless you changed them - Fixed push-to-talk - Fixed delay on recording (first run) - Fixed spacing next to icon - Fixed crash when audio input source changes - Added watch directory as a new trigger - Watch for images in a directory and run an action when a new one is added

- Advanced Voice Mode for all AI providers (even Ollama) using on-device Apple voices, or cloud voices by OpenAI - Multi-Modal interaction, send screenshots, audio, and mix them in a single conversation - Better context awareness of the AI in transcription - New actions: Look at Screen (text/image) - Support for running terminal commands through SSH

- Added support for gpt-4o-audio-preview (but the model itself is very buggy) - Added recording indicator + start/stop listening sounds - Added speak action - Added support for the model to output as text/audio aside from running actions - Fixed setting of your name from settings on first run - Fixed missing email body for some email servers (when using process) - Fixed filtering of image and audio data from logs, and other various log fixes - Fixed issues with screenshots now capturing the top part - Added instant record feature (zero delay between mic listen, but shows always recording) - Fixed non-western keyboard layouts not being able to select and paste text - Allow ESC to cancel audio playback/generation and auto cancel on new recording - Fixed multi-language support in Whisper

- Simplified General settings and moved a lot of them to advanced - Added way to disable on-device image-processing and audio-processing (advanced) - Added way to send audio to the AI model (similar to images before), only supported by one model: gtp-4o-audio-preview (advanced)

- Support for multiple hotkeys - Support for OpenRouter

- Fixed examples for Obsidian

- Added full documentation in-app - Added examples for Apple Notes integration - Fixes for URL based actions (encoding)

User actions can now return information to previous AI steps, allowing for multi-step AI actions.

Big improvements to audio commands. The AI can access your screen and selected text, allowing for much more interesting audio commands.

Added audio command mode, use your voice to trigger any action.