Best Open-Source Meeting Notetakers (2026)

Full disclosure: I make Humla, the first entry, so treat me as biased. Licences and capabilities below were checked against each project's GitHub repo in July 2026, and I've said plainly where another tool beats mine.

Two categories, and why it matters

Every tool here is genuinely open source — real licences, real repos, no "open-core" sleight of hand except where I flag it. But they don't do the same job.

A meeting notetaker solves capture. It gets the audio out of a live conversation — your mic plus the other side's voices coming out of your speakers — while the meeting is happening, then transcribes and summarises it. Capture is the hard, unglamorous part: system-audio permissions, two streams, chunking, keeping "you" separate from "them."

A transcription tool starts one step later. You give it a file; it gives you text, often excellent text. To use one for meetings you must first solve capture yourself — a recorder app, a Zoom cloud recording, an audio-routing utility like BlackHole — and then feed the file in.

Both are legitimate. Just know which problem you're solving before you install anything.

How they compare

Checked against each project's GitHub repo, July 2026. These move fast — verify before you commit.

1. Humla — the one that captures the meeting and keeps everything local

Humla is a Mac notetaker that records your microphone and the call's system audio through macOS itself, so nothing joins the call and the other side sees no participant in the list. The two streams are kept separate end to end rather than mixed down, which is what makes "you said" versus "they said" reliable before any model runs — on a remote call, everything on the mic stream is you by definition.

From there it can run with nothing leaving your machine: Whisper on the Metal GPU for on-device transcription, offline speaker diarization on the Neural Engine, and summaries against a local model through Ollama. Or bring your own OpenAI, Deepgram or Groq key if you'd rather trade privacy for speed — your account, your bill. There's also chat over your whole note history, which runs the same way: local embeddings and a local model, or your own key. I wrote up exactly how that retrieval works if you want to audit the design rather than trust it.

The self-hosting story is the part relevant to this list. Humla's client is MIT and every feature in it is free — there is no Pro tier on the app. Team sync is the only thing that costs money, and only if you want it managed: point the app at your own PocketBase server instead of Humla Cloud and every team feature works with no subscription. The paid tier buys hosting, not features. That's an unusual shape — most open-core products reserve their best features for the paid tier, and this one deliberately doesn't.

Choose it when: you want a bot-free notetaker that captures the call, labels speakers on-device today, and can be self-hosted end to end without a paywall.

Tradeoffs, honestly: macOS 13+ on Apple Silicon only — no Intel build, no Windows, no Linux, no mobile app. No CRM or Slack integrations. And the community is younger and smaller than Meetily's or WhisperX's, which matters if you want lots of third-party plugins or a big issue tracker to search when something breaks. If you need Windows, skip to Meetily; if you need Linux, Vibe.

2. Meetily — the cross-platform notetaker

Meetily is the one to reach for if you're not on a Mac. Free, MIT, bot-free, with local Whisper or NVIDIA Parakeet transcription, and it genuinely runs on Windows as well as macOS (Linux builds from source). It has one of the largest communities in open-source meeting notes and points summaries at Ollama, Claude, Groq, or any OpenAI-compatible endpoint.

Choose it when: you need Windows, want the biggest community, or want Parakeet and the ability to re-transcribe imported audio.

Tradeoffs, honestly: speaker diarization is still listed as coming soon and isn't shipped in any tier, including Pro — so "who said what" isn't available yet. And it's open-core: custom templates, higher-accuracy models and eventually diarization sit behind Pro at $10/user/month. Full Humla vs Meetily.

3. anarlog — Markdown files on disk, plus a plugin ecosystem

anarlog (the MIT, macOS-only successor to Hyprnote, renamed by the fastrepl team in 2026) is local-first and bot-free, does on-device transcription and diarization, and stores every meeting as a plain Markdown file you can sync with git, Dropbox or Obsidian. It ships a plugin system and supports a wide list of LLM providers.

Choose it when: you live in Markdown or Obsidian, or you want a plugin ecosystem. If your notes being portable text files on disk is the priority, this beats Humla's local SQLite — that's a genuine preference difference, not a shortcoming.

Tradeoffs, honestly: macOS only, no mobile, and the rename means the branding and docs are mid-transition.

4. Vibe — the cross-platform wildcard

Vibe is an MIT desktop transcription app for macOS, Windows and Linux — the widest platform support here. It's built around transcribing files, but it can capture system audio and the microphone, which puts it closer to a notetaker than the rest of this section. It does speaker diarization, supports Whisper, Parakeet and Nemotron models, and can summarise via the Claude API or Ollama.

Choose it when: you're on Linux, or you want one local transcription app across three platforms.

Tradeoffs, honestly: it isn't designed as a meeting notetaker, so the meeting-shaped workflow — recording a call, keeping your typed notes alongside the transcript, a summary tuned for meetings — isn't really the point of it. You get transcription plus a summary, not a notes app.

5. WhisperX — the best pure transcription engine here

WhisperX (BSD-2-Clause, 23k+ stars) is the most technically impressive project on this list and I'd rather say so than bury it. It adds three things to plain Whisper: word-level timestamps via forced alignment with wav2vec2, speaker diarization, and batched inference that hits roughly 70× realtime with large-v2.

If you're building something, this is the engine you build on. If you need precise word timings — subtitles, search-into-audio, editing tools — nothing else here competes.

Choose it when: you need word-level accuracy or you're assembling your own pipeline.

Tradeoffs, honestly: it's a library and CLI, not an app. It won't record your meeting, there's no UI, no notes, no summaries — you bring the file and the glue code.

6. Scriberr — the self-hosted server with chat

Scriberr (MIT) is the most complete self-hosted option: deploy via Docker or Homebrew, and you get transcription with Parakeet, Canary or Whisper, speaker identification, AI summaries, and chat over your transcripts — plus a built-in recorder, notes, an API, and folder monitoring for automation. Word-level timing included.

If your goal is "a private transcription service on my own hardware that the whole household or team can post files to," this is the closest thing to it here, and its feature list overlaps Humla's more than anything else on this list.

Choose it when: you want a self-hosted service rather than a desktop app, or you want to automate transcription through a watched folder or API.

Tradeoffs, honestly: it doesn't capture live meetings the way a notetaker does — the workflow starts from a file. And be aware of the bus factor: the maintainer recently noted they've paused active development after losing their job, while staying open to contributions. That's not a knock on the software, but it is worth knowing before you build a workflow on it.

7. Whishper — good idea, currently stalled

Whishper (AGPL-3.0) is a self-hosted, Docker-deployed web UI for transcribing audio and video, with translation via LibreTranslate and a genuinely nice subtitle editor.

I'd normally recommend it for subtitle work, but honesty first: it is not actively maintained. The repo carries a notice that the author is working on a complete rewrite and the current branch won't get new releases or updates. It also has no diarization and no summaries.

Choose it when: you specifically want its subtitle-editing workflow and you're comfortable running unmaintained software, or you're waiting on the rewrite.

Which self-hosts, and how completely

Since self-hosting is usually the reason people want open source in the first place, here's the honest breakdown — because "open source" and "self-hostable as a service" aren't the same claim:

• Runs entirely on your machine, no server needed: Humla, Meetily, anarlog, Vibe. Nothing to deploy; the app is the whole thing.
• Deploy as a service on your own hardware: Scriberr, Whishper. This is the classic self-hosted shape — a box on your network you send files to.
• Self-host the team layer too: Humla, uniquely on this list, because it's the only one with a sync/teams layer at all. Point it at your own PocketBase server and every team feature is free, with no paid gate.
• Build it yourself: WhisperX. It's the engine; the service around it is your job.

How to choose in one minute

1. Need notes from a live call, on a Mac? Humla.
2. Need that on Windows? Meetily — accepting no speaker labels yet.
3. On Linux? Vibe.
4. Live in Obsidian? anarlog.
5. Have recordings already, want a box on your network to process them? Scriberr.
6. Building your own pipeline, or need word-level timestamps? WhisperX.
7. Want subtitles specifically? Whishper, with the maintenance caveat.

The honest summary: if the job is meeting notes, the shortlist is genuinely short, because capture is the hard part and most open-source projects in this space sensibly don't attempt it. If the job is transcription, you're spoiled, and WhisperX and Scriberr are excellent.

Try Humla

Download Humla for macOS — free, MIT, no bot, no sign-up, and it runs on your Mac. If you want to see on-device transcription before installing anything, the local page has a browser demo that runs the model in your tab and uploads nothing. The source is on GitHub, including the self-hosting instructions.

If you're comparing specific tools, there's Humla vs Meetily, and a Mac-focused roundup that goes deeper on the three notetakers.