---
title: "Mumr 1.13.0: Every voice on the call, told apart — Changelog"
description: "Mumr 1.13.0 labels and names the speakers in a transcript, can write summaries with a local model powered by Mirai, and has a redesigned Settings window. All of it runs on your Mac."
canonical: "https://mumr.app/changelog/1-13-0"
language: "en-GB"
last-updated: "2026-11-14T23:08:24+00:00"
source: "https://mumr.app — Markdown rendition of the HTML page at `canonical`"
---

[v1.13.0](https://mumr.app/changelog/1-13-0) macOS November 9, 2026

# Every voice on the call, told apart

Mumr 1.13.0 has four parts. After a meeting, Mumr now tells the voices on the call apart, labels them Speaker 1, Speaker 2 and so on, and lets you give each one a name. Summaries can come from a local model [powered by Mirai](https://trymirai.com), an open model that runs on your Mac and keeps working with the network off. Mumr can open when you log in, so it is there when a call starts, and it can record the call apps you choose by itself. And Settings has a new design, with every summary engine side by side. Recording, transcription and speaker labels all happen on your Mac, as before.

## Speaker labels

![Three folded white name cards on a dark slate table, reading Anna, Mark and You, lit from behind by a violet light.](https://mumr.app/changelog/img/1-13-0-speakers.webp)

On a Mac with Apple silicon, Mumr downloads the speaker-label model (about 221 MB) by itself once the speech model is ready, and Settings → Transcription → **Speaker labels** shows its progress with Cancel; Cancel or Delete stops automatic downloads for good. On an Intel Mac the row does not appear. The files come from Hugging Face, and Mumr checks each one against a fixed size and checksum before it installs them.

From then on, when a meeting ends and its transcript is final, Mumr separates the voices on the call on your Mac. The transcript shows **Separating speakers…** while it works. The finished transcript reads Speaker 1, Speaker 2 and so on, numbered in the order they first spoke, with your own lines still marked You. A meeting that ended before the download finished gets its labels once the download does.

-   **Naming**: click a speaker's name in the transcript to type a name. If the meeting has a calendar event, its organizer and attendees are offered as suggestions
-   **Corrections**: in the same popover, **Merge into…** joins two voices that are the same person, and **This line was…** moves one line to someone else. Changes apply to that meeting only
-   **Where names go**: exports, Copy Transcript, PDF, Send to Participants and the Claude and GPT hand-off use the names. The summary uses only names you gave, never "Speaker 2"
-   **Your voice**: Settings → Transcription → **Your voice** learns your voice from meetings on headphones, or from **Record 1 Minute**. Once it has, the people your microphone hears on speakers or in the room get their own speaker numbers, and meetings in person get labels too. While it learns, the row shows how far it has got, as in **Learning · 1 of 3 meetings**, and **Record 1 Minute** gives you a text to read aloud. **Reset** makes Mumr forget it
-   **If labelling fails**: Mumr tries once more by itself. If the speaker-label model did not load, it does not retry; Settings → Transcription → **Speaker labels** shows **Repair** instead. After that, and once the model is ready, **Separate speakers** in the meeting's Regenerate menu runs the step again from the meeting's audio, without re-transcribing

**Tip:** names survive **Regenerate transcript**, so you can re-transcribe a meeting without naming everyone again.

## Local model

![A space-black laptop open on dark slate, its screen dark with a soft violet glow, a pair of white wired earbuds beside it.](https://mumr.app/changelog/img/1-13-0-local-model.webp)

[Powered by Mirai](https://trymirai.com). Until now, a summary written on your Mac came from Quick highlights, which picks sentences from the transcript, or from Apple Intelligence on macOS 26. The local model writes the same kind of summary as Claude or GPT, from the same prompts, including any you customised, and it gives each meeting its title too. The transcript never leaves your Mac.

Open Settings → AI and click the **Local model** tile. Before the first download, a sheet names the two companies involved and what each one sees, and nothing happens until you click **Agree and Download**. When the download finishes, Mumr uses that model for the summary and the title of each meeting, with the network off if you like.

When you set up Mumr on a Mac that can run it, the AI page now leads with **Qwen, free on this Mac**, with Claude CLI, Codex CLI or an API key below as the alternatives. The local model's card, tile and download sheet use [Mirai](https://trymirai.com)'s own look, in setup and in Settings.

-   **The model**: Qwen 3.5 4B by Alibaba, about 2.9 GB, on Macs with 16 GB of memory or more. A Mac with 8 GB is offered Qwen 3.5 2B, about 1.1 GB, which is smaller and faster but can get names, owners and dates wrong, and Settings says so beside it
-   **The Mac**: Apple silicon and macOS 26.4 or later. On other Macs the tile does not appear
-   **Switching**: click another tile whenever you like. **Delete…** removes a model's files from your Mac, and the summaries it already wrote stay. Delete the one you use, and new summaries go back to Quick highlights
-   **Fallback**: if a meeting is too long for the model, or the model cannot finish, Mumr writes Quick highlights instead and says why on the summary
-   **Licences**: the engine is MIT-licensed and the Qwen 3.5 models are Apache 2.0. **Open-source licences**, under the model in Settings, shows the full texts

[Mirai](https://trymirai.com)

The model runs on [uzu](https://github.com/trymirai/uzu), [Mirai](https://trymirai.com)'s open-source engine for Apple silicon. Two outside services take part, and neither sees your meetings: the model files come from Hugging Face, and each time the engine starts it sends [Mirai](https://trymirai.com) your macOS version, chip and memory size so [Mirai](https://trymirai.com) can pick the right build for your Mac. The [privacy policy](https://mumr.app/privacy) lists both.

## Ready when the call starts

Mumr can only offer to record a meeting while it is running. It now has a few ways to be ready, and each one stays off until you choose it.

-   **Open at login**: the last page of setup offers **Open Mumr at login** or **Not Now**. If you already use Mumr and it does not open at login, a card in the main window asks once, **Open Mumr when you log in?**, with **Turn On** and **Not Now**. You can change it any time in Settings → General
-   **A quieter prompt**: when another app is using the microphone, the prompt to record comes after a minute, or after 10 seconds for a call app or while a calendar event is under way. When its countdown ends, it shrinks to a small **Record** button. Its menu has **Always record** for Zoom, Microsoft Teams, Slack, Discord and FaceTime, and **Don't ask for** the app
-   **Always record**: the second time you press **Record** for a call app, Mumr asks, for example, **Always record Zoom meetings?** Choose **Always record**, and Mumr starts recording 10 seconds into each call in that app, plays the start chime and shows a **Recording Zoom meeting** panel with **Stop**, **Discard** and **Hide**. **No thanks** leaves it off
-   **Your rules**: Settings → Recording → **Meetings** → **Apps** lists the call apps on your Mac and every app you have set a rule for, **Always record**, **Don't ask** or **Ask**, with **Reset**

To notice a call, Mumr looks only at which app is using the microphone, never at its audio.

## Improvements

-   **Settings**: The Settings window has a new design, with a sidebar like System Settings. Each pane has its own coloured badge and groups its options on cards. Switches, progress bars and your chosen engine take Mumr's violet, or the accent colour you picked in System Settings. A status reads in green or orange, always with a word beside the colour
-   **Settings**: Settings → AI shows every summary engine as a tile, with one line on where it runs and what it writes, such as **On this Mac · Full summary** or **Anthropic · Full summary**. The engine Mumr uses is marked **In use**
-   **Settings**: Turning automatic summaries off is now the **Summarize after each meeting** switch in Settings → AI. Turn it back on, and Mumr goes back to the engine you used before
-   **Transcript**: Your lines and the call's lines now appear in the order they were spoken, where they used to read as two blocks
-   **Regenerate**: The menu lists the summary engines by name, the local model included, with a checkmark on the one in use, a download size only where one is needed, and nothing this Mac cannot run. A model that finishes downloading while a meeting is open shows up straight away
-   **Downloads**: Progress for the speech, speaker-label and local models moves in steady whole-percent steps

If you skipped 1.12, its [notes](https://mumr.app/changelog/1-12) cover the speech model downloading during setup and the recording card in the sidebar.
