An open source AI note taker lets you run the recording and transcription yourself, but the summary still needs a language model. If that model is a cloud provider, your transcript text goes to it. For a private setup, run a local model, and check whether the tool keeps some features in a separate edition.
The common assumption is that an open source note taker keeps your meetings private. It might, but the licence has nothing to do with it. What decides privacy is where the audio is processed and, above all, where the summary gets written.
That's the thread through this page. Eight tools follow, in no particular order, with a table showing how each one hears the meeting, its licence and where the summary comes from.
Where your transcript goes when the summary is written
Nearly every tool here needs a language model to turn a transcript into a summary. Some default to a local model on your own machine. Others let you plug in a cloud provider. Meetily's README spells out the trade: audio and transcription stay on your device, but if you choose a cloud AI provider, the transcript text is sent to it. A fully private setup means running a local model, such as Ollama, yourself.
Licences differ too. Three of the desktop apps are open core: the free app is open, while a Pro edition, commercially licensed enterprise components or a cloud add-on sit alongside it. Two of the server tools use copyleft licences, AGPL-3.0 and GPL-3.0, which only start to matter if you modify one and offer it to others. Everything here was active on its repository as of Nov 2026.
Audio source, licence and summary engine
| Tool | How it gets the audio | Licence and summary source |
|---|---|---|
| anarlog (formerly Hyprnote) | Device audio, no bot | MIT community app; local model or provider |
| Humla | Device audio on a Mac, no bot | MIT, open core; provider you choose |
| Meetily | Device audio, no bot | MIT, open core; local model by default |
| Millet | Recordings fed to a command line | GPL-3.0; local Ollama or enclave service |
| Scriberr | Uploaded audio and video files | MIT; Ollama or OpenAI-compatible provider |
| Speakr | Recordings uploaded to your server | AGPL-3.0; summary engine you configure |
| Vexa | A bot that joins the call | Apache-2.0; transcripts via your own API |
| VoiceFlow | Mic plus system audio, no bot | MIT; summary from an LLM you choose |
Eight note takers, in no particular order
anarlog (formerly Hyprnote)
Best for: Notepad-style meeting notes from your device audio, with no bot joining the call and data kept in a local database. It began as Hyprnote.
Humla
Best for: Mac users wanting bot-free recording, speaker separation and summaries that blend their own notes with what was said, without an account.
Meetily
Best for: Recording, transcribing live and summarising meetings on your desktop with a local model by default, and a cloud provider only if you choose one.
Millet
Best for: Command-line users turning a meeting recording into a speaker-labelled transcript, an AI summary and a PDF, working offline with local models.
Scriberr
Best for: Transcription first: local transcripts of audio and video, then summaries or chat over them through Ollama or an OpenAI-compatible provider.
Speakr
Best for: Teams wanting recordings turned into transcripts, summaries and searchable notes on their own server, choosing their own transcription engine.
Vexa
Best for: Developers who need a bot in Google Meet, Teams and Zoom calls streaming speaker-labelled transcripts through an API they can host themselves.
VoiceFlow
Best for: Windows and Linux users who mainly dictate but want a meetings mode that records, transcribes locally and summarises through an LLM they pick.
How each one hears the meeting
An app on your own machine, no bot
Humla and VoiceFlow join the desktop recorder from the section above and the notepad app that began life under another name. They listen to your device's audio, so nobody sees a bot join the call. One is Mac only, and the dictation app's meetings mode is a side job rather than its main purpose.
A server you host
Vexa sends a bot into the call and gives developers transcripts through an API. Speakr and Scriberr take recordings you upload; the second is transcription-first, with summaries added through a model you connect.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
A command-line pipeline
Millet is for people comfortable in a terminal: feed it a recording, get back a transcript, a summary and a PDF.
My advice on choosing for privacy
Decide first whether the summary may leave your machine. If it mustn't, pick a tool that supports a local model and set that up before your first real meeting, not after. If a cloud model is fine, any of the bot-free apps will feel closest to the hosted note takers you may already know, and our guide to alternatives to a popular hosted note taker covers those.
If you only need transcripts, a transcription-first tool is simpler, and our round-up of AI transcription software lists hosted options too. Either way, check which features sit in a separate edition before you plan your team's workflow around them.
Consent comes before any recording
Bot-free recording is quieter, and that's exactly why it needs care. If nobody sees a bot join, people may not know they're being recorded. Recording and consent rules vary by country and sector, so check the ones that apply to you, and tell everyone on the call that notes are being taken. It's the polite thing to do anyway, and it keeps the notes usable if anyone asks later.
Do you make an open source AI meeting note taker? Tool makers can ask to be considered for a featured place on this list.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.
Frequently asked questions
Is an open source AI note taker private?
Only if the summary is written by a model you run yourself. If you connect a cloud AI provider, the transcript text is sent to it, even when the app is open source.
Do open source note takers need a bot in the meeting?
Not all of them. Some record your own device audio with no bot, while server-based tools can send a bot into the call or work from uploaded recordings.
What does open core mean for a note taker?
The main app is open source, but some features, such as speaker identification, integrations or sync, sit in a separate edition or add-on under different terms.
Can I run an AI note taker completely offline?
Some tools here say transcription and summaries can run with local models. You'll need to install and run the model yourself.
Which open source note taker works on a Mac?
Several desktop apps support macOS, and one is Mac only. Check each project's README, since one dictation-first tool doesn't officially support macOS.
Should I tell people I'm recording a meeting?
Generally, yes. Rules vary by country and sector, so check what applies to you and tell participants before you record.
