Yogurt: Local Open-Source Alternative to Granola

At a Glance
| Item | Value |
|---|---|
| Published | 2026-11-12 |
| License | MIT |
| Source type | Primary source (the publisher itself) |
Values determined by this site’s code when the information was collected. Dates are JST.
Overview
Yogurt is a desktop application that handles meeting transcription, summarization, and Q&A entirely using on-device AI models. It is free under the MIT license and written primarily in TypeScript. Creator Eric (brownrout) released it on Show HN. Similar to Granola, it does not join calls and instead runs in the background.
This material is an introductory post for the initial release rather than a version update release notes document, so version numbers and change logs are not included. The point of greatest impact for users is that transcription data, recordings, and summaries reportedly never leave their PC. It is worth trying for anyone who cannot send meeting content to the cloud.
Key Changes
Locally Contained Processing Architecture
According to the author, whisper.cpp is used for transcription. Apple’s on-device SpeechAnalyzer is also used if supported by the system. Local open-weight LLMs are used for summarization and Q&A. This part runs on llama.cpp.
Models are automatically downloaded and set up to match hardware performance. The author explains they are chosen within a range that “keeps the cooling fans from making a huge fuss." Specific model names, parameter counts, and required memory are not stated in the material and remain unverified.
Supported OS and Implementation
Built with Electron to support Windows, Mac, and Linux. The author states they spent time optimizing UI behavior, animations, and model calls. Comments on HN appreciated the fact that it can be used outside of Apple Silicon.
Local MCP Server
Yogurt features a local MCP server. The author explains that if users want to pass meeting minutes to external models like those from Anthropic or OpenAI, they can share them directly without going through a paid cloud service. Calendar integration is also cited as a feature lacking in other OSS tools. No account creation or connection is required.
Recording Warnings
According to HN comments, a warning prompting users to notify call participants that they are being recorded is displayed by default. The same comments included requests for re-notification of late joiners. Whether this is implemented cannot be confirmed from the material.
Points to Know Before Adoption
The following are opinions raised in HN comments, all of which are officially unverified.
- There are observations that local model processing for summarization and Q&A consumes a lot of resources. Some voices requested an option to skip post-processing or offload to OpenRouter.
- Concerns have been raised regarding memory consumption by Electron.
- Questions were raised about speaker diarization and multilingual support, but could not be confirmed in the material.
- Comparisons of summary quality (evals) against Granola are not posted on GitHub, though requests have been made.
- Similar projects such as Muesli, OpenOats, OpenWhispr, and anarlog are mentioned in the comments. The author’s answers regarding differences are not in the material.
The body of the material cuts off midway. What is written here is based on what could be read.
Supported Models and Hardware
Based on what can be understood from the material, the details are as follows:
- Supported OS: Windows / Mac / Linux. Implemented with Electron. HN comments appreciated that it is not limited to Mac or Apple Silicon.
- Transcription: Uses whisper.cpp. Apple’s on-device SpeechAnalyzer is used when supported by the system.
- Summarization & Q&A: Uses local open-weight LLMs running on llama.cpp. According to the author, models that run comfortably on the hardware are automatically downloaded and set up.
The following items are not listed in the material and remain unverified:
- Names of automatically selected models
- Parameter counts
- Quantization formats
- Required memory
- Supported GPUs
- Presence of speaker diarization
- Scope of multilingual support
Whether it will run on your own environment must be verified by actually trying it.
How to Update
As this is an introductory post for the initial release, the material contains no installation steps or commands, so steps will not be guessed and written here. Please check the following pages for acquisition and installation:
- GitHub Repository: FrigadeHQ/yogurt
- Official Website (listed in the repository’s homepage field): yogurt.so
- Demo video by the author: YouTube
Version numbers and release notes are not in the material either. Those who already have it installed should check the repository’s releases page for the latest version.
What to Read Next
- Engines mentioned in this article → whisper.cpp / llama.cpp

