Resource
Granola is not open source, and it transcribes in the cloud. If you want an open-source tool instead, the best pick depends on what you liked about Granola: the notepad, the bot-free capture, or the meeting memory your AI can read. Here are nine open-source projects, what each is actually for, and where each falls short.
Quick answer
For the editable AI notepad Granola is known for, Anarlog (formerly Hyprnote) is the closest open-source match. For local transcription and summaries, Meetily is the most widely used. For meetings saved as Markdown that your existing AI assistant can search, consider Minutes, which is what this site is for.
All three capture device audio without a bot. Check two things before you choose: whether transcription actually runs on your machine by default, and where the notes are stored.
Granola runs on macOS, Windows, iOS, and Android and records without adding a bot to the call. Transcription goes through third-party cloud providers, and notes are stored on Granola’s servers. It has a free Basic plan, paid Business and Enterprise plans, and a hosted MCP server that lets AI assistants read your notes. The source code is not published.
| Project | License | Platforms | On-device transcription | Storage | MCP | Capture |
|---|---|---|---|---|---|---|
| Minutes | MIT | macOS (Apple Silicon), Windows; Linux CLI | Yes, by default | Markdown + YAML files | Yes, local | Mic and system audio, no bot |
| Anarlog (formerly Hyprnote) | MIT; enterprise directory commercial | macOS, Windows, Linux; mobile in beta | On Mac only; cloud is the easy path | SQLite and files; Markdown export | Yes | Device audio, no bot |
| Meetily | MIT; PRO is separate and paid | macOS, Windows; Linux from source | Yes | Local SQLite | Not found | Device audio, no bot |
| OpenWhispr | MIT | macOS, Windows, Linux | Your choice of local or cloud | Local SQLite; optional cloud sync | Yes | Device audio; detects Zoom, Teams, FaceTime calls |
| StenographAI (StenoAI) | MIT | macOS; Windows and Linux alpha | Yes, by default | Markdown | Agent skill; no server found | Mic and system audio, no bot |
| OpenOats | MIT | macOS 14.2+ (Apple Silicon) | Yes | Plain-text files | Not found | Device audio, no bot |
| LokalBot | GPL-3.0 | macOS 15+ (Apple Silicon); Windows, Linux alpha | Yes, by default | Not stated | Optional, read-only | Device audio, no bot |
| Omi | MIT | macOS, iOS, Android, wearable; Windows from source | No; Omi backend | Omi backend | Hosted | Device, phone, or wearable |
| Vexa | Apache-2.0 | Self-hosted server, or hosted | Only with your own GPU service | Postgres and a Markdown workspace | Yes | Meeting bot |
Markdown conversation memory for your own AI
Meetings, voice memos, and dictation transcribed on your machine (Parakeet v3 on Apple Silicon, Whisper elsewhere) and saved as Markdown that Claude, Codex, or another MCP client can search. Written by the Minutes maintainer, so weigh it accordingly.
The closest open-source take on an AI notepad
An editable local notepad with SQLite storage, Markdown export, a CLI, and MCP. On-device transcription is Mac-only; the docs point new users to Anarlog Cloud for the easiest setup.
Local transcription and summaries
The most-starred project here. Transcribes locally with Whisper or Parakeet and summarizes with Ollama or a cloud model. Meetily PRO is a separate, paid codebase.
Dictation and meeting notes in one app
Started as a dictation app and now also transcribes meetings, with on-device speaker labels and calendar integration. Local or cloud models for every feature, plus optional cloud sync and team sharing through OpenWhispr Cloud.
A local AI notepad for confidential calls
Records, summarizes, and answers questions about meetings with local AI by default; cloud providers are opt-in. Parakeet v3 on Apple Silicon for live text, Whisper for other languages, and notes saved as Markdown.
A live in-call assistant on a Mac
Transcribes on an Apple Silicon Mac and saves plain-text transcripts to your Documents folder. Closer to a live assistant during the call than a notepad afterward.
A small on-device Mac notetaker
Transcribes on-device by default and offers an optional read-only MCP server and CLI. A young, single-developer project with Windows and Linux in alpha.
Memory across desktop, phone, and a wearable
MIT-licensed apps and wearable hardware. Audio streams to the Omi backend for transcription, so the open license does not make it local by default. MCP is hosted.
Self-hosted meeting-bot infrastructure
An API that sends a bot into Meet, Teams, Zoom, and Jitsi and returns transcripts, with MCP. Transcription needs a Vexa token or your own GPU service. Something you operate, not a desktop notepad.
Pick Anarlog if you want to write and edit notes during the meeting, the way Granola works.
Pick OpenWhispr if you want dictation and meeting notes in one app with local or cloud models, or StenographAI for a local-first notepad aimed at confidential conversations.
Pick Meetily, OpenOats, or LokalBot if the priority is transcription that stays on your machine, and you are happy with that project’s platform support.
Pick Minutes if you want meetings, voice memos, and dictation in one Markdown archive that Claude, Codex, or another assistant can search through MCP.
Pick Omi for a phone or wearable, or Vexa if you want to run meeting-bot infrastructure yourself.
If what you want is Granola’s polished hosted experience, shared team notes, and workspace features, none of these is a drop-in replacement. Open-source tools win on control, inspectable files, and local processing. They ask more of you in setup, and features such as sharing are thinner or absent.
Each entry was checked against the project’s own repository, license file, and documentation on 2026-10-29. This is a documentation review, not a hands-on benchmark of accuracy or reliability. Projects were included only if they are open-source licensed, actively maintained, and built for meetings. Tools with no license, a license that now restricts commercial use, or no updates in over a year were left out.
Next step
The compliance answer that sends people looking.
The transcription layer underneath most of these.
Which of them an assistant can actually read.
Getting the existing bot out first.