Repolode
Submit
michaelwilhelmsen

michaelwilhelmsen/humla

Open-source macOS meeting-notes app that records mic and system audio without a bot, transcribes locally or via cloud APIs, and writes summaries from your notes plus the transcript.

Open-source alternative to:GranolaOtter.ai

View on GitHubWebsite
Stars
338
Forks
45
Open issues
24
Last commit
6 days ago

What is it?

Humla is a meeting-notes app for macOS. You take freeform notes while a meeting runs; Humla records your microphone and system audio as two separate streams, transcribes them, identifies speakers offline, and produces a structured summary that combines your typed notes with the transcript. No bot joins the call.

Transcription can run fully on-device with Whisper on Apple Silicon, or through OpenAI, Deepgram, or Groq, with per-language provider overrides. Summaries can use OpenAI or any OpenAI-compatible local server such as Ollama or LM Studio. Recorded audio is discarded by default unless you turn on the option to keep it.

Humla also includes a chat tab that searches and cites your notes using keyword and embedding search, and an MCP server that lets tools like Claude Code query your meeting notes in read-only mode. Optional sync to a PocketBase server is available, and the server can be self-hosted.

Key features

  • 01Records microphone and system audio as separate streams, with no meeting bot.
  • 02Transcribes locally with Whisper on Apple Silicon or via OpenAI, Deepgram, or Groq.
  • 03Identifies and labels speakers offline, and lets you rename them.
  • 04Generates summaries from typed notes and transcript, with selectable presets.
  • 05Chat answers questions over your notes and cites the source notes.
  • 06Read-only MCP server exposes notes to Claude Code, Codex, and other MCP clients.

Use cases

  • Taking meeting notes on a Mac while keeping audio and transcripts local.
  • Transcribing Norwegian or other non-English meetings with per-language engine settings.
  • Searching past meetings from an AI coding agent without switching apps.
  • Running summaries and chat entirely on a local LLM server.

Quick start

  1. 1

    Download the macOS app from the releases page

  2. 2

    Open Humla and add a transcription or summary provider key in Settings, or use local models

Things to know

  • Runs only on macOS 13 or later, and Apple Silicon is recommended.
  • Cloud transcription and summaries need your own API keys, while fully local use needs a local LLM server.
  • The server half of Humla Cloud is not published, so self-hosted sync requires you to build your own compatible server.

In collections

MCP serversOpen-source Mac appsPrivacy & local-first toolsAI video & audio tools

Topics

#meeting-notes#transcription#speaker-diarization#local-first#privacy#whisper#macos#mcp#ai#apple-silicon

FAQ

What is humla?
Open-source macOS meeting-notes app that records mic and system audio without a bot, transcribes locally or via cloud APIs, and writes summaries from your notes plus the transcript. Humla is a meeting-notes app for macOS. You take freeform notes while a meeting runs; Humla records your microphone and system audio as two separate streams, transcribes them, identifies speakers offline, and produces a structured summary that combines your typed notes with the transcript. No bot joins the call.
Is humla free to use? What license does it have?
humla is open source under the MIT license. Check the license terms before using it in commercial products.
Is humla still maintained?
Actively maintained. The last commit to humla was on Nov 1, 2026.
What can I use humla for?
Taking meeting notes on a Mac while keeping audio and transcripts local. Transcribing Norwegian or other non-English meetings with per-language engine settings. Searching past meetings from an AI coding agent without switching apps. Running summaries and chat entirely on a local LLM server.
How do I install humla?
Download the macOS app from the releases page Open Humla and add a transcription or summary provider key in Settings, or use local models
Where can I get humla?
The source code and installation instructions are on GitHub: https://github.com/michaelwilhelmsen/humla

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects