Meetings you can come back to.

Voicewoven records calls on your Mac, transcribes them on the device, and tells the voices apart. Then it writes the summary, the decisions, and who agreed to do what.

Check availability See how it works

For macOS 14 Sonoma or later on Apple Silicon. In development, with no public download yet.

No bot joins the call.

Voicewoven listens to your Mac, not to the meeting. It records your microphone and the system audio on separate channels, so it works with whatever app the call is in, and nobody gets an invitation from a recorder.

MicrophoneYou
System audioEveryone else

Headphones work best: this version doesn't cancel echo from your speakers.

From conversation to something you can use.

You press record and stop. Everything after that happens on its own, in this order.

  1. Record

    Start from the window or the menu bar. Live meters show both channels, and an optional preview shows the words as they come in.

  2. Transcribe

    When you stop, a speech model on your Mac writes the full transcript. It handles English, Portuguese, and Spanish, and can detect which one was spoken.

  3. Tell voices apart

    Each voice becomes its own thread in the weave above the transcript. Name a voice once, play a sample to check it, and the name sticks.

  4. Summarize

    A local language model writes a summary and the decisions, and suggests tasks along with the exact quote and moment they came from.

  5. Ask

    Ask a question across many meetings, narrowed by tag and date. The answer shows which meetings it came from.

Tasks keep their receipts.

Suggested tasks arrive with the sentence someone actually said. Accept the ones that are real, ignore the rest, and set who owns it and when it's due.

Accepted tasks keep their quote even if you later delete the meeting or its audio.

Suggested from Onboarding redesign review
Send revised onboarding copy
Marina
“I'll have the new copy for the import screen by Friday.”Marina at 12:40
AcceptIgnore
Fri, Oct 10
Share drop-off numbers with the team
You
No date

Ask what was said, weeks later.

Pick the tags and dates to search, then ask in plain words. Voicewoven reads every transcript in that range and answers with the meetings it relied on.

Answers follow the language of your question. Quotes stay in the language they were spoken.

ProductThis quarter
What did we decide about the trial length?

The trial stays at 14 days for now. Theo suggested 30 days in September, but on October 7 the team chose to wait for data from the new onboarding first.

Onboarding redesign reviewOct 7, at 21:15
Pricing syncSep 18, at 06:02
Ask a follow-up

Fix what it misheard

Edit any line of a transcript. The original stays available, and earlier versions can be restored.

Tags and search

Tag meetings as you go, filter the library by tag, and combine it with text search.

Keep the audio you want

Set an audio limit or keep transcripts only. Old audio goes first; the text, names, and tasks stay.

It stays on this Mac.

Recording, transcription, voice separation, summaries, and answers all run on your computer. Models download once; after that, meetings are processed without an internet connection. Transcripts are never sent to a cloud service.

Speech
Parakeet TDT v3 or Nemotron Speech Multilingual
Voices
Community-1, through FluidAudio
Summaries and Ask
Qwen3, in 1.7B, 4B, or 8B, through a bundled Ollama
~/Library/Application Support/ com.voicewoven.desktop/ meetings/ 8f3c…/ audio.wav mic left, system right meeting.json transcript, names, summary transcripts/ earlier versions tasks.json ask.json tags.json models/

Plain files in a folder you can open in Finder. Nothing needs Homebrew, Python, or a terminal.

Where it stands.

Voicewoven is in active development. This is what works today and what's still on the way.

Works now

  • Recording, live preview, and final transcripts
  • Named voices with audio samples
  • Summaries, decisions, and tasks
  • Ask across meetings, Tags, and storage limits
  • Interface in English, Portuguese, and Spanish

On the way

  • A signed and notarized download
  • An MCP server, so tools like Claude and Codex can search your meetings
  • Starting recordings automatically when a call begins

Good to know

  • No echo cancellation yet; use headphones
  • Overlapping speech can need a quick review of names
  • Language switches mid-meeting aren't split yet

Voicewoven is coming soon.

A signed download for Apple Silicon Macs is on the way. Check back here for availability.

System
macOS 14 Sonoma or later
Mac
Apple Silicon, M1 or later
Disk
About 1.4 to 5.2 GB for the AI model you choose, plus speech models