A new Mac application, Talat, offers an alternative to cloud-based AI notetaking tools like Granola by prioritizing user privacy. Developed by Nick Payne and Mike Franklin, Talat processes all audio and transcripts locally on the user’s machine, avoiding data sharing with external servers. This approach responds to growing concerns about data security and control in AI-powered productivity tools.
The Rise of Local AI
The development of Talat stems from a desire for greater privacy in the increasingly popular AI notetaking space. While tools like Granola provide real-time transcription and summaries, they require uploading audio data to the cloud. Payne, driven by this trade-off, sought to create a solution where users could enjoy the benefits of AI without surrendering their voice recordings.
The key to Talat’s functionality is Apple’s Core Audio Taps API and the FluidAudio framework, which enables running small, fast transcription models directly on the Mac’s Neural Engine. This allows for low-latency, fully local AI processing without relying on external cloud services. Payne’s work on the open-source AudioTee library further facilitated this development, streamlining access to the necessary Apple APIs.
How Talat Works
Talat captures audio from meeting applications like Zoom, Teams, and Meet, transcribing it in real time. It attempts to identify speakers automatically, though users can manually adjust assignments. The app also generates summaries with key points, decisions, and action items using a local Large Language Model (LLM).
All notes, transcripts, and summaries remain searchable within the app, never leaving the user’s device.
Customization and Control
Beyond privacy, Talat emphasizes user control. The app allows users to select their preferred LLM (including options like Qwen3-4B-4bit, Parakeet variants, or Ollama) and configure data export settings. It also supports integrations with tools like Obsidian, webhooks, and MCP servers, giving users maximum flexibility over how their data is handled.
Pricing and Availability
Currently in pre-release, Talat is available for a one-time purchase of $49. The price will increase to $99 upon the 1.0 release. Payne and Franklin are bootstrapping the project, committed to maintaining a one-time purchase model rather than a subscription. Users with M-series Macs can download the app and test it with 10 hours of recordings before deciding to purchase.
Talat fills a critical gap in the AI productivity market by offering a fully private, local-only alternative that empowers users to control their data while enjoying the benefits of real-time transcription and summarization.
