AI Meeting Notes App Talat Moves Transcription On-Device for Privacy

Talat’s AI meeting notes app is a Mac-only ($49 one-time, pre-release) tool that promises “local-only processing,” keeping meeting audio, transcriptions, and summaries on the user’s device. The app is developed by Yorkshire programmer Nick Payne (with Mike Franklin) and targets privacy-conscious professionals who want to avoid cloud-based data exposure. Talat addresses a key tradeoff in hosted AI transcription: cloud tools often require sending not only your text context but your actual audio to third parties. Talat instead uses Apple Neural Engine hardware and local model execution to transcribe and summarize without sending data to company servers. Key technical points: the 20MB app defaults to an on-device summarization model (Qwen3-4B-4bit) optimized for Apple M-series chips. For transcription, users can choose between Parakeet variants or bring custom models via Ollama. Features include real-time transcription with speaker identification, local LLM summarization that generates key points and action items, full search across notes, and export options (including Obsidian). It also supports capturing audio from Zoom, Microsoft Teams, and Google Meet. Release details: Talat requires an M-series Mac (M1 or later). It includes 10 free recording hours for evaluation. Pricing is expected to rise to $99 after version 1.0, while maintaining the one-time purchase model for the core app. From a market angle, this is a privacy-focused productivity trend that may pressure cloud AI providers to offer stronger data controls, but it is not directly tied to specific crypto assets.