知Yizhi Daily
← Back to this edition

Conversations

Developer showcases Yogurt, an open-source local AI meeting-notes tool

Edition 2026-11-12 · 08:00 Beijing time

Developer Eric shared his side project Yogurt on Hacker News. It is a free, MIT-licensed desktop application that uses on-device local AI models for meeting transcription, summarization, and question answering, so recordings, transcripts, and summaries do not leave the computer. The software supports Windows, Mac, and Linux and offers calendar integration, search, a local MCP server, and other features.

According to developer Eric, Yogurt is a free, MIT-licensed open-source desktop application positioned as an alternative to Granola-style meeting-notes tools. It can run in the background without joining users' conference calls while using AI models on the device to transcribe and summarize meetings and allowing users to ask questions about meeting content.

Privacy is the project's core selling point. Eric said meeting transcripts, call recordings, and summaries never leave the user's computer. As small open-source models have improved in speed and capability, he believes meeting-minute summarization and question answering based on minutes do not necessarily require frontier cloud models. If users want to provide notes to frontier models such as Anthropic or OpenAI, they can share them directly through Yogurt's local MCP server without having a paid cloud service act as an intermediary.

Technically, Yogurt uses Electron to support Windows, Mac, and Linux. Transcription uses whisper.cpp and can also use Apple's on-device SpeechAnalyzer when supported by the system. Summarization and question answering use local open-weight large language models; the software automatically downloads and configures them according to hardware capabilities, with the relevant part based on llama.cpp.

Eric said he had tried many open-source note-taking tools, but many products lacked features such as calendar integration and MCP servers, required connections or account creation, and were not polished enough for daily use. He therefore added details to Yogurt such as echo detection and reduction, global/folder/local hierarchical search, and automatic recording termination, while optimizing model calls and interface interactions, hoping to meet the needs of users who value privacy and cost.

Original sources

Hacker News ↗
Back to this editionOpen interactive reader