Mumr.app's answer

Mumr records both sides of a call on your Mac. Your microphone and the audio the Mac is playing are captured separately with Core Audio process taps, so Zoom, Microsoft Teams, Google Meet, a browser tab or any other app that makes sound all record the same way. No bot joins the meeting. Transcription runs on the Mac itself, using Parakeet, Whisper Small or Apple Speech, whichever you select.

The AI that writes the summary is a setting rather than a decision Mumr made for you, because the right answer differs per meeting: a client call under a confidentiality clause and a standup do not need the same model. The built-in on-device summary needs no account, no API key and no network. Apple Intelligence on macOS 26 also stays on the machine. A cloud key, or the claude or codex CLI already installed on your Mac, is there when you want a stronger model.

Every meeting lands in a folder you picked, as an m4a file plus JSON for the transcript and the summary, with a mumr-meetings.json index at the folder root. That layout is why Claude Code or Codex can read the folder directly, and Mumr has a button that opens one of them on the folder with the question ready to send. Nothing is sent until you press Return.

Mumr.app's answer

Granola, Otter.ai and Fireflies.ai handle the meeting end to end in their own cloud. Mumr handles the same meeting on your Mac: capture and transcription run locally, and the only thing that can leave is transcript text, and only if you point the summary at a cloud model with your own key. In that case Mumr calls Anthropic, OpenAI or OpenRouter directly with the key you supplied, and that provider bills your account. Mumr is not in that transaction.

Mumr is also free on the Mac App Store with no in-app purchases, so there is no per-seat price and no plan to choose. Choose Apple Intelligence, the built-in on-device summary, or your own local claude or codex CLI, and no meeting content leaves the machine at all. Separately, Mumr sends anonymous usage events to PostHog. They carry no meeting titles, transcripts, summaries, attendees or file paths.

MacWhisper, superwhisper and Aiko already cover local transcription on a Mac. Mumr overlaps with them there and adds the meeting around it: it captures the far side of the call as well as your microphone, reads your calendar so it can offer to record the meeting that just started, and writes a structured summary with key points, and with decisions and action items once you point it at an AI backend.

Mumr.app's answer

People who take a lot of calls on a Mac and want the notes without handing the audio to a service. In practice that is two groups: people who record client calls under a confidentiality clause and cannot upload the audio to a vendor, and people who take enough calls that a per-seat notetaker subscription is the thing they are trying to avoid.

A second reason people arrive is the choice of AI. Someone already paying Anthropic or OpenAI would rather spend the key they have than buy another subscription. Someone who wants no cloud model involved at all can stay on the built-in on-device summary or on Apple Intelligence. Developers already running Claude Code or Codex get a meeting folder those tools can read directly.

Mumr is a single-user Mac app. There is no shared workspace, no CRM sync, no admin console, and no Windows, web or iPhone version, so a sales team that needs call data pushed into a pipeline is not the audience.

Mumr.app's answer

I struggled in meetings. Following the conversation and writing anything useful down at the same time never worked for me. One of the two always lost, and it was usually the notes.

Every tool I found solved that by uploading my calls to someone else's servers, and charging me every month for it. The upload was the part I could not get past. A meeting recording is not only my data, it belongs to everyone who was on the call, and none of them agreed to a vendor. Paying a subscription for that trade made it easier to keep saying no.

So I built the one I wanted and put it on the Mac App Store for free. Recording and transcription happen on your Mac and stay there. The summary is whichever AI you pick, including none. There is no account, no subscription and no Mumr server. The version of this I needed did not exist, and there was no reason to keep it to myself once it did.

Mumr.app's answer

macOS app: Swift and SwiftUI on macOS 15 or later, sandboxed and hardened for the Mac App Store. System audio is captured with Core Audio process taps and the microphone with AVAudioEngine using Apple's voice processing. Transcription runs on-device through FluidAudio (NVIDIA Parakeet TDT 0.6B v3, the default), WhisperKit (OpenAI Whisper Small, compiled to Core ML), or Apple's Speech framework in on-device mode.

Summaries come from one of four places, whichever you select: Apple's NaturalLanguage framework on-device, Apple Foundation Models on macOS 26, the Anthropic / OpenAI / OpenRouter HTTP APIs called directly with your key, or the claude and codex CLIs already on your Mac, run as a local subprocess through NSUserUnixTask. Calendar integration is EventKit. Meetings are stored as plain files and JSON, with no database, and API keys live in the macOS Keychain.

Backend and web: mumr.app is TypeScript on Hono and Node 22, styled with Tailwind CSS v4, bundled with esbuild and deployed on DigitalOcean App Platform. It serves the landing page, the blog and the comparison pages. It holds no user data, because the app never sends it any.

The only third-party service the app talks to on its own is PostHog, for anonymous, content-free product analytics.