Open-source AI meeting notetakers

First, a distinction most lists skip

A meeting notetaker captures the meeting as it happens (your mic plus the meeting audio), transcribes it, and turns it into structured output — summary, decisions, action items — inside a workflow built for meetings. A transcription tool turns audio into text, excellently, and stops there. Both are valuable; only one replaces Otter or Granola. Half the "open-source Otter alternatives" you'll see recommended are actually the second kind, which is how people end up disappointed. Both kinds are below, clearly separated.

The true open-source notetakers

Meetily — the biggest community

Meetily (MIT, ~24k GitHub stars) is the most established open-source notetaker: live capture with real-time transcription on Mac and Windows, built on Tauri and Rust, with local Whisper/Parakeet transcription and notes via Ollama or your own Claude/OpenAI-compatible keys. Honest caveats: it's still pre-1.0 (v0.4 pre-release at the time of writing), and the free Community Edition sits alongside paid PRO and Enterprise tiers. If you want the largest community around an open-source notetaker, this is it.

Anarlog (formerly Hyprnote) — the Mac pick

Anarlog (MIT, ~9k stars) is the polished local-first notetaker for macOS: on-device transcription, markdown notes on disk, bring-your-own LLM (including Ollama and LM Studio), no account and no paid tier. If the name is unfamiliar: it's Hyprnote, renamed twice in 2026 (Hyprnote → Char → Anarlog) and relicensed from GPL to MIT along the way. The one hard limit: macOS only, and the team has said Windows isn't planned.

Earshot — the Windows-native one (ours)

Earshot (MIT) is our entry, so judge the framing accordingly — but the facts are checkable in the repo. It's built Windows-first: dual-channel local recording (your mic and the meeting audio captured separately, so speaker attribution is physical fact rather than diarisation guesswork), no bot in the call, transcription via your own Whisper server, Groq, Deepgram or OpenAI, and notes via Anthropic, any OpenAI-compatible endpoint or a fully local model. It's a young project — you're early — and it's the only one here with a managed option: Earshot Plus runs the AI for US$9/month if you'd rather never touch an API key, which is also how development is funded (no venture money, no data business).

Speakr — the self-hosted team server

Speakr (AGPL-3.0 with a commercial dual licence, ~3.5k stars) takes a different shape: a Docker-deployed web platform your whole team logs into, with in-browser recording, uploads, diarisation, summaries and per-recording chat. It doesn't transcribe by itself — you wire it to a WhisperX container or a cloud ASR API — and it's self-described alpha. The right pick when "self-hosted" means a server for many people rather than an app on one machine.

At a glance

Star counts and statuses from the GitHub API, 13 July 2026. These move — check the repos.

Excellent open-source transcription tools (that aren't notetakers)

• Vibe (MIT, ~7k stars) — the best free cross-platform transcription GUI: files, mic, system audio, diarisation, GPU acceleration, every export format. No meeting workflow.
• WhisperX (BSD, ~23k stars) — the gold-standard library for word-level timestamps and who-said-what; a Python tool for builders, not an app.
• Speaches (MIT, ~3.5k stars) — an OpenAI-compatible self-hosted speech server; the kind of thing you point Earshot or Speakr at.
• Scriberr (MIT, ~2.8k stars) — self-hosted transcription with summaries and chat for uploaded audio; development currently paused per the author.
• noScribe (GPL-3.0, ~2k stars) — Whisper + diarisation desktop GUI aimed at researchers transcribing interviews.

Worth knowing about, with caveats: Vexa (Apache-2.0, ~2.5k stars) is an open-source meeting-transcription API that works by sending bots into Meet/Teams/Zoom — the philosophical opposite of the local-capture tools above, but genuinely open source. Whishper and Amurex appear on many older lists but haven't seen commits in roughly a year; treat them as dormant.

How to choose

• On a Mac: Anarlog is the obvious starting point; Meetily if you want the bigger community.
• On Windows: your two real options are Meetily and Earshot. Meetily brings the larger community; Earshot brings dual-channel speaker attribution, a meetings-first workflow (projects, action-item approval, Ask with citations), and the escape hatch of a $9/month managed tier when you're done fiddling with keys. Try both — they're free.
• For a team server: Speakr, wired to WhisperX.
• Just need transcripts, not notes: Vibe on the desktop, WhisperX or Speaches in the stack.

Whichever you pick, the reason to pick open source at all is the same one we lay out in local vs cloud transcription: for a tool that hears everything you say at work, "read the code" beats "trust the policy." And if you're comparing against the proprietary heavyweights instead, that's covered in Earshot vs Otter, Fireflies, Fathom & Granola.