Freedom251 › AI Meeting Assistants
Millet

What Millet is
Millet is free, open-source desktop software for recording meetings, transcribing speech, and producing summaries, action items, and PDFs on Linux and macOS. It is listed for meetings across Zoom, Google Meet, Microsoft Teams, Slack, Discord, Signal, Telegram, WhatsApp, Keet, Jitsi Meet, Webex, Skype, FaceTime, GoTo Meeting, RingCentral, Amazon Chime, and BlueJeans. On Linux, microphone and system audio are recorded on separate channels. Whisper large-v3-turbo supports 99 languages; English, German, French, Spanish, Turkish, and Farsi are listed as tested. Summaries can include an overview, key topics, decisions, action items, and open questions or follow-ups. Exports include TXT, SRT, JSON, Markdown, and PDF. After models are downloaded, transcription, diarization, and PDF output work offline. Summaries use either a hardware-attested Tinfoil TEE or local Ollama. Tinfoil is listed at approximately $0.03 per meeting, while Ollama is free. Transcripts and summaries can be pushed to any Git repository on a schedule.
Who it suits
Millet suits Linux users who need meeting recordings, transcripts, and structured summaries across many call apps. macOS users can process captured sessions, but audio capture requires Linux.
What it does well
- Free and open source
- Supports many meeting services
- Exports several text and document formats
- Offline transcription after downloads
- Offers local Ollama summaries
What to know first
- Windows is unsupported
- Alpha software
- Overlapping speech is handled poorly
- Diarization needs a Hugging Face token
Verdict
Millet combines desktop capture with transcription, structured meeting notes, and multiple export formats. Its Linux-first capture model and alpha status matter when choosing it, especially for users needing macOS capture or Windows support.
Facts from Millet's own pages
- Audio capture
- On Linux, Millet records microphone audio and system audio on separate channels. · github.com, 12 Oct 2026
- Desktop interface
- The GTK3 GUI provides record and stop controls, a timer, status indicators, and shortcuts to the resulting PDF and session folder. · github.com, 12 Oct 2026
- Exports
- Millet can produce .txt, .srt, .json, .summary.md, and .pdf files. · github.com, 12 Oct 2026
- Integrations
- Millet can push transcripts and summaries to any Git repository on a configurable schedule. · github.com, 12 Oct 2026
- Intended users
- The README describes Millet as a tool for recording and processing meetings, including with browser-based meeting apps and desktop clients. · github.com, 12 Oct 2026
- Languages
- The README says Whisper large-v3-turbo supports 99 languages and lists English, German, French, Spanish, Turkish, and Farsi as tested languages. · github.com, 12 Oct 2026
- License
- The repository identifies Millet as licensed under GPL-3.0. · github.com, 12 Oct 2026
- License and maturity
- The repository identifies Millet as alpha software and provides it under the GPL-3.0 license. · github.com, 12 Oct 2026
- Limitations
- Windows is not supported, and the README says overlapping speech is not handled well. · github.com, 12 Oct 2026
- macOS support
- On Apple Silicon macOS, Millet supports post-capture transcription, labeling, and sync, while audio capture requires Linux. · github.com, 12 Oct 2026
- Meeting app compatibility
- The README says Millet works with call apps that play audio through system speakers, including Zoom, Google Meet, Microsoft Teams, and Webex. · github.com, 12 Oct 2026
- Meeting capture
- desktop · github.com, 20 Sept 2026
- Meeting platforms
- Zoom,Google Meet,Microsoft Teams,Slack,Discord,Signal,Telegram,WhatsApp,Keet,Jitsi Meet,Webex,Skype,FaceTime,GoTo Meeting,RingCentral,Amazon Chime,BlueJeans · github.com, 20 Sept 2026
- Offline operation
- After models are downloaded, transcription, diarization, and PDF output work offline; summaries require either the networked TEE or local Ollama. · github.com, 12 Oct 2026
- Output formats
- Millet can produce TXT, SRT, JSON, Markdown summaries, and PDFs. · github.com, 12 Oct 2026
- Platforms
- Full recording and processing are supported on Linux; macOS Apple Silicon supports post-capture transcription, labeling, and sync, while Windows is not supported. · github.com, 12 Oct 2026
- Private processing
- Summary generation supports a hardware-attested Tinfoil TEE or a fully local Ollama model; the README says both are private backends. · github.com, 12 Oct 2026
- Purpose
- Millet transcribes meetings, summarizes them, and produces PDF output. · github.com, 12 Oct 2026
- Requirements
- Speaker diarization requires a free Hugging Face account and token, and the README recommends an NVIDIA GPU with at least 8 GB of VRAM while also allowing slower CPU processing. · github.com, 12 Oct 2026
- Summaries
- Meeting summaries include an overview, key topics, action items, decisions, and open questions or follow-ups. · github.com, 12 Oct 2026
- Summaries and action items
- Yes · github.com, 20 Sept 2026
- Summary cost
- The README lists Tinfoil at about $0.03 per meeting and Ollama as free. · github.com, 12 Oct 2026
- Summary privacy
- Summary backends are a hardware-attested Tinfoil TEE or fully local Ollama, and the README says both are private. · github.com, 12 Oct 2026
- TEE cost
- The README estimates Tinfoil summarization at approximately $0.03 per meeting. · github.com, 12 Oct 2026
Where Millet stands on Freedom251
- AI Meeting Assistants#55 of 82