Talat, a new macOS application, has emerged as a privacy-focused alternative to high-profile AI notetakers like Granola. Developed by Nick Payne and Mike Franklin, the software processes all audio data locally on the user’s hardware, ensuring that sensitive meeting information never touches the cloud. Unlike subscription-based competitors currently dominating the market, Talat is available as a one-time purchase.
The Quest for Local AI Privacy
While industry leaders like Granola—recently valued at $1.5 billion—have set a high bar for AI-assisted productivity, developer Nick Payne identified a significant gap in the market. The concern was not the utility of the tools, but the trade-off required: users must surrender their raw audio data to third-party servers.
Payne’s journey began with research into Apple’s Core Audio Taps API, which eventually led to the creation of the open-source library AudioTee. The project gained momentum upon the discovery of FluidAudio, a Swift framework that enables low-latency AI processing directly on Apple’s Neural Engine.
How Talat Works
Talat is a lightweight, 20MB application designed for users who prioritize data sovereignty. It captures audio from standard meeting platforms like Zoom, Teams, and Google Meet, providing real-time transcription and speaker identification. Once a meeting concludes, a local Large Language Model (LLM) generates a summary, highlighting key decisions and action items.
Key features include:
- Full Local Processing: No account creation required; zero analytics data shared.
- Configurable Intelligence: The app defaults to the Qwen3-4B-4bit model but allows users to swap in other LLMs or utilize MCP servers.
- Integration Ready: Supports auto-export to tools like Obsidian, with future plans for Google Calendar and Notion integration.
Pricing and Availability
The app is currently optimized for Macs equipped with Apple’s M-series silicon. Users can test the software with a 10-hour free recording allowance before committing to a purchase.
Currently, Talat is available for a one-time fee of $49 during its pre-release phase. Payne and Franklin intend to maintain a one-time purchase model, with the price set to increase to $99 upon the official 1.0 release. By bootstrapping the project, the founders aim to avoid the pressure of recurring subscription models, focusing instead on user control and transparency.