Show HN: Yogurt – Free, private, open-source alternative to Granola
A new open-source app, Yogurt, aims to be a free, privacy-focused alternative to Granola by running meeting transcription and summarization locally across macOS, Windows, and Linux. Commenters weigh its design choices—such as using Electron and automatic model selection—against concerns about performance, diarization quality, legal consent for recording, and integration with other AI tools. The conversation situates Yogurt within a rapidly growing ecosystem of similar tools (like Muesli, OpenWhispr, Biscotti, and others), highlighting strong demand for local, customizable, and subscription-free AI note-taking solutions.
Overall Reception
- Many commenters are enthusiastic, calling Yogurt a much-needed, privacy-preserving alternative to Granola and praising that it’s free and open source.
- Multiple people say they’ve been looking to cancel or avoid SaaS note-taking tools and will try Yogurt.
- Some skepticism exists about whether this category is over-saturated and whether most apps just do “record + summarize.”
Positioning vs Other Tools
- Yogurt is framed as a Granola-like, but:
- Cross-platform (macOS, Windows, Linux), unlike several alternatives that are Mac-only.
- Local-first with on-device models, emphasizing privacy.
- It’s compared or grouped with: Muesli, OpenOats, OpenWhispr, Biscotti, Steno, Talky, Yapsnap, FluidVoice, various personal tools, Google Edge Foresight, and others.
- Some argue similar tools (e.g., OpenWhispr) already cover this space with broad model support; others note there can be many overlapping options.
Technical Implementation & Performance
- Yogurt uses Electron to leverage the React ecosystem for the UI. Some worry about memory usage; others point out that local models dominate resource cost anyway.
- It automatically chooses and downloads STT/LLM models based on hardware, with requests for a manual model selector and the ability to reuse already-downloaded models.
- Several commenters want an option to offload summarization and “ask” features to remote APIs (e.g., OpenRouter) to reduce local resource consumption.
- There is interest in how it compares technically to other backends like RealtimeSTT and Handy.
Diarization, Languages, and Features
- Yogurt supports diarization (explicitly on macOS) and multilingual transcription; details for other platforms are less clear.
- On macOS, it can use accessibility APIs to map speaker names in Zoom/Meet; diarization via models is discussed as an alternative or complement.
- Multiple commenters share diarization techniques (pyannote, Nemotron, wespeaker) and projects that remember voices across sessions.
- Requests include better diarization in “same-room” mic scenarios, hotkeys for dictation, and richer evaluation metrics vs Granola’s summaries.
Privacy, Consent, and Security
- Strong interest in Yogurt’s fully local, private design; it’s seen as safer than cloud-based tools if those are hacked.
- Some emphasize legal and ethical concerns around recording meetings, especially in two-party-consent jurisdictions, and suggest better in-app reminders when new participants join.
- A brief security review notes the local MCP server uses Unix domain sockets / named pipes with restricted permissions; no obvious token-based auth is seen.
Naming and Branding
- Mixed reactions to the “Yogurt” name: some like it and its food-theme relation to Granola; others think generic names harm discoverability or imply complement rather than replacement.