Speak it into shape.

No. Mumble records your microphone and your Mac’s audio directly. Nothing joins the call. Tell the other people you’re recording, the same as with any recorder.

Yes. Because Mumble listens to your Mac’s audio rather than joining the call, it works with Zoom, Teams, Meet, Slack huddles or any browser call.

Yes. Put your Mac on the table and Mumble listens through the microphone, so workshops, 1:1s and whiteboard sessions get mapped the same way.

Audio never does. Transcription runs locally with Whisper. To build the map, the transcript and canvas go through your own Claude Code or Codex account, under your provider’s terms. Choose transcription-only and nothing is sent. The Private plan, coming soon, runs the models on your computer too, so nothing leaves it at all.

A Mac on macOS 13 or later, and Claude Code or Codex installed and signed in for the AI features. Mumble uses your existing plan’s allowance and never asks for or stores an API key.

Yes, in transcription-only mode. You get a local transcript with no AI and nothing sent anywhere. The live map needs Claude Code or Codex.

They can, if you stop and type a prompt afterwards. Mumble builds the diagram while you’re talking, in the shape you picked, and links every box to the sentence it came from. No prompt, no copying anything across.

Six shapes: Decision, Brainstorm, Process, Research, Retro and Architecture. Each map can be drawn in three styles, Business, Sketch or Icons, and you can switch in one click without re-running the AI.

Every box links to the words behind it, with a timestamp. Click a box to see the quote. Anything Mumble infers rather than hears is marked in purple as a suggestion.

Say so. “No wait, scratch that” or “actually…” makes Mumble repair the map in place, so the box shows what you finally decided.

Yes, during and after the recording. Drag, rename, add notes, shapes, frames and images. Your edits always win over later AI updates.

Add them to a vocabulary list, for one workspace or one recording, and transcription gets them right. A small model with a good vocabulary beats a bigger model without one.

Yes. When you stop, Mumble writes an overview and up to 12 key points, each cited back to the transcript or a box on the map.

Follow-ups that someone states go into a Tasks group. Owners and dates appear only when someone actually said them.

Yes. Embedded views for calendar, project, table and notes read the same boxes live, so they stay in sync with the map.

It doesn’t label speakers yet. It captures the whole conversation in order. On calls, headphones give the cleanest audio.

English today.

In plain, readable files in ~/Mumble on your Mac. Export the canvas as PNG, PDF or to the clipboard, and the summary and transcript as Markdown.

Yes. Canvas files are deterministic JSON with readable IDs, so renaming a box is a one-line diff and adding one is a single block. They review well in a pull request.

Not yet. Canvases are local files on your Mac, so for now you export and share the result. Team mode, coming soon, adds sharing, publishing and team collaboration.

Not yet. Mumble is a native Mac app for now. Join the waitlist to hear when that changes.

Nothing while Mumble is in beta. After the beta, using Mumble locally on your Mac stays free, forever. Two paid plans are coming: Private, a one-off price for fully local models and unlimited canvases, and Team, a monthly plan for sharing and collaboration.

Join the waitlist below and click the link in the confirmation email. You’re on the list once you confirm.

tiny space, an independent product lab. Mumble is our first product.