On-device meeting notes · Whisper · Parakeet · NVIDIA Nemotron 3.5

Meetings, captured locally.

Mityu records and transcribes meetings on your machine, with local reporting by default, live 8-speaker attribution on overlapping speech, 40-language transcription, and source-linked notes you approve line by line. No bots. No audio uploads. No account.

Up to 8 speakers · overlap-aware 40 languages · 31-language translation Encrypted local storage · Privacy guard
Download for Windows Inspect 4-Zone Shell See pricing

Windows 10/11 (64-bit), processor with AVX2 · macOS in development

Accuracy varies by language, microphone, overlap, and noise. Target-environment validation and the human pilot have not been performed; review important output against its source audio. Speaker separation is a best-effort estimate — available live with NVIDIA Nemotron 3.5 Transcribe (up to eight anonymous speakers) or after recording via the local diarization pass — and speaker labels stay anonymous; Mityu never attaches a name to a voice, because storing which voice belongs to whom would be biometric data. The live copilot is a beta: off by default, and not yet checked against a real model. Raw meeting audio remains local until you delete the meeting.

4-Zone Desktop Shell · One Private Loop

One calm workspace, from live voice to verified knowledge.

AppRail, Meetings library, report workspace, and source-linked Evidence Drawer — every step runs on your device, and every AI-drafted claim links to its exact transcript timestamp.

Speech & Intelligence Architecture

Built for complex rooms, overlapping voices, and strict privacy.

Combine fast on-device Whisper and Parakeet engines with NVIDIA Nemotron 3.5 Transcribe for multi-speaker separation, translation, and custom domain vocabulary.

NVIDIA Nemotron 3.5 Transcribe Multi-Speaker · Translation

One engine for transcription, speaker separation, and translation.

Separates up to 8 speakers even when people talk at the same time, supports 40 spoken languages, and translates into 31 languages during live or post-meeting passes.

Speaker 1
Speaker 2
Speaker 3
Speaker 4
Speaker 5
Speaker 6
Speaker 7
Speaker 8
  • ✓Tracks up to 8 anonymous speakers even when participants talk at once
  • ✓Simultaneous translation across 31 target languages during live or post-meeting passes
Promptable Audio Intelligence Domain Context · PII Guard

Steer domain vocabulary and redact PII inline.

Pass project codenames, technical acronyms, and participant roles directly into the transcription prompt, and strip sensitive emails, phone numbers, or card details before segments hit disk.

Prompt: Transcribe with speaker labels. Context: Mityu, SQLCipher, AVX2, Amira. Redact PII.
Output: [Speaker 1] Send the contract to [REDACTED_EMAIL] before Friday.
  • ✓Custom vocabulary and domain context prompt per meeting
  • ✓Opt-in inline PII redaction plus regex pre-storage masking
On-device Zero Audio Upload

Transcription that never leaves.

Recording and transcription run locally with Whisper large-v3, Parakeet, or NVIDIA Nemotron 3.5 Transcribe — no network needed from capture to finished report.

  • ✓Whisper large-v3, Parakeet & NVIDIA Nemotron 3.5 Transcribe
  • ✓SQLCipher when secure OS key storage is available
  • ✓Retranscribe or import recordings anytime
Any meeting app System + Mic Audio

No bot joins your call.

Mityu captures your microphone and system audio on the machine itself, so it works with Zoom, Google Meet, or Microsoft Teams without joining the meeting.

  • ✓Zoom, Google Meet, Microsoft Teams
  • ✓Windows 10/11 (64-bit), AVX2 processor
  • ✓macOS support in development
Human-approved Source-Linked

Drafts until you approve.

Every AI-generated line is a labeled draft linked to its source in the transcript. Nothing is final until you sign it off — and questions are answered from that transcript alone.

  • ✓“AI-generated” label on every draft
  • ✓Source links to exact timestamps in Evidence Drawer
  • ✓Export approved notes to PDF, Word or Markdown
  • ✓BYOK: Ollama, OpenAI, Anthropic, Groq, OpenRouter
The learning loop

It gets sharper every meeting.

Correct the same thing a few times and Mityu turns it into a plain-language rule that shapes your next report — automatically. Every rule stays visible and editable, and the loop runs on your device by default.

Transparent Licensing

Choose the plan.

Get a taste

Trial

$0 — 14 days

Everything, free for 14 days.

Download free

Plan includes:

  • ✓Every feature
  • ✓Works fully offline
  • ✓No account required
Most popular

Pro

$79 one-time

One license, yours to keep.

Buy Mityu Pro

Plan includes:

  • ✓2 devices
  • ✓1 year of updates
  • ✓Email support
For teams

Business

$59 /user /year

Roll Mityu out across your team.

Contact us

Plan includes:

  • ✓Priority support
  • ✓Volume licensing
  • ✓Team workspace on the roadmap

Mityu is available now for Windows 10/11 (64-bit) with an AVX2-capable processor (Intel Core 2013+, AMD 2015+; some budget Celeron and Pentium Silver chips are not) — download the free trial or buy a Pro license above. Every plan includes the current release, with the live copilot beta off by default.

Every purchase comes with a 14-day money-back guarantee.

Frequently asked questions

When can I get Mityu?

Mityu is available now for Windows 10 and 11 (64-bit). Download the free 14-day trial, or buy a Pro license — both from the buttons above. The download always serves the current release, and every version’s notes are on the release notes page. macOS support is in development.

Is my audio uploaded anywhere?

No. Recording, transcription, and storage all happen on your device. If you connect your own cloud model key, only the transcript text you choose to summarize or ask about is sent to that provider — your audio never leaves the machine. The beta live copilot ships stricter still: its “Allow cloud insights” setting is off, so it answers only with a model running on your device and refuses rather than sending anything away. Turning that setting on under Settings → Beta sends the last few minutes of transcript to the provider you configured, on every action you press.

Does a bot join my meetings?

No. Mityu listens to your microphone and system audio locally. Nothing joins the call, and participants see no extra attendee.

Which platforms are supported?

Mityu is available on Windows 10 and 11 (64-bit). macOS support is in development, and Linux is being explored. Write to us and we’ll tell you the moment new platforms land.

Does it work offline?

Yes. Capture, transcription, and local summaries work with no network at all. An internet connection is only used if you bring your own cloud model key.

Which models and API keys can I use?

Transcription runs locally on Whisper large-v3, Parakeet, or NVIDIA Nemotron 3.5 Transcribe (30B-A4B MoE, served via a local vLLM or NIM endpoint, or your own OpenAI-compatible audio endpoint). For summaries, use the built-in local model, run one through Ollama, or bring your own key for OpenAI, Anthropic, Groq, or OpenRouter. Ask this meeting uses that same model choice. The beta live copilot ships restricted to a model running on your device, and will use your cloud provider only if you turn on “Allow cloud insights” under Settings → Beta.

What is the live copilot, and is it on?

It is off by default, and it is a beta. Turn it on under Settings → Beta and a small always-on-top panel follows a recording you start. When you ask, it drafts a recap, a suggestion, a definition or follow-up questions from the last few minutes of the transcript; every answer is labelled AI-generated and cites the transcript segment it came from, or is refused. It ships with “Allow cloud insights” off, so it answers only with a model running on your device and refuses rather than sending anything away; turning that on sends the last few minutes of the conversation to the provider you configured, on every action you press. It never records on its own, never hides Mityu from you or anyone in the room, and on Windows it can stay out of your screen share. It has not yet been checked against a real model, which is why it is a beta and why we publish no accuracy or latency figure for it.

Can I ask a meeting a question afterwards?

Yes. Open a finished meeting and ask; the answer is drawn only from that meeting’s transcript, each line opens the segment it came from, and anything the model cannot trace to the transcript is dropped rather than shown. Every answer is marked as AI-generated and needs your review.

What’s the refund policy?

Every purchase comes with a 14-day money-back guarantee. If Mityu isn’t for you, write to us within 14 days and we’ll refund it in full.