noum.ai

All features

Only what already works. Items marked come with the next version. Everything else is on the roadmap.

Recording meetings

No bot on the call
Microphone and call audio are captured on your Mac. Nobody extra appears in the participant list.
Automatic recording
The app notices when a call starts and ends, and starts and stops recording. Recordings that are too short are thrown away.
Asks first
An offer to record appears at the top right for 20 seconds. No answer and that call is not recorded. You can also record without being asked.
Recording from the calendar
Recording can start on schedule. You choose the calendars, and they give the meeting its title and participants.
A strip at the screen edge
While recording, a narrow strip with a waveform and a stop button stays at the right edge, even over a full-screen call.
Clean audio
After recording, the far side’s echo is removed from the microphone. The source audio stays as it was.
Notes and marks
During a meeting, mark a moment with ⌃⌥M or write a note. The summary takes them into account, and clicking a mark plays from there.
Meeting prep
Open an upcoming meeting to see who is coming, when you last met, and that meeting’s decisions and action items.
Recording from a phone
Record a conversation in Voice Memos on your iPhone and save it to the Noum folder in iCloud Drive. The Mac collects the file and treats it as a normal meeting. Turn it on in Settings.

Dictation

Hold, speak, let go
Text is pasted wherever the cursor is, in any app. Use a key or a button at the screen edge. Turned on during setup.
On this Mac only
The audio is recognised locally and lives in memory for one phrase.
History and dictionary
All phrases are in the main window. Add names and terms to the dictionary. You can clear or turn off the history.
Text clean-up
An AI adds punctuation and drops filler words. It gets the phrase’s text, never the audio. While clean-up is off, the text goes nowhere.

Transcription

On this Mac
Speech is recognised on the Mac itself, with no internet. Russian has its own model, trained on Russian speech.
In the cloud, if you need it
Connect a cloud service such as OpenAI with your own access key. It receives the meeting audio and is better at telling apart several people on one microphone.
A choice of models
Three run on the Mac. Parakeet is the default and knows 25 European languages. GigaAM is made for Russian, and Whisper understands the most languages. In the cloud you can pick AssemblyAI, OpenAI or ElevenLabs. Switch any time in Settings.
Text during the meeting
Read the live transcript as the conversation goes. This also happens on the Mac.
Import files
Bring in an existing recording and it is processed like any other meeting.
Transcribe again
Transcribe a recording again another way, in the cloud for example. You can see which step the work has reached.
Correcting the text
Replace a word across the whole transcript. The original is kept.
No lines made up in silence
Speech recognition sometimes writes “Thank you.” into a pause. Lines with no voice under them on their track are dropped. You can turn this off in Settings.

Summaries and names

Who said what
You and the others are on separate tracks, and the app fills in names from the conversation and the calendar’s participant list.
A voice for each person
When several people are on the other end of a call, their voices are told apart on this Mac, so each gets a name of their own.
Fix a name with one click
Got a name wrong? Click it right in the transcript and correct it.
Decisions and action items
Key points, decisions, action items and open questions, written by the AI you chose.
Summaries from Noumy
No key, no subscription and nothing to install, free during the beta. The meeting text goes through our server to OpenAI. Our server keeps none of it. Pick it during setup or in Settings.
Or your own AI
A Claude or ChatGPT subscription through their Claude Code and Codex apps, your own Anthropic or OpenAI key, another OpenAI-compatible service, or any model on the Mac itself through Ollama.
Your own instructions
Change the summary instructions and language in Settings.
Templates
Five ready-made: one-on-one, sales, interview, team meeting, lecture. Add your own and set one per folder. The previous summary is kept.
Catches up later
If the AI could not be reached, say with no network, the recording waits and one button finishes the work.

Asking your recordings

Chat with a meeting
Ask one meeting, or all of them, what was agreed. Answers link to moments in the recording, and clicking plays from there.
Saved questions
Save a frequent question as a command and ask it with one button.
Recordings in Claude
The notetaker mcp command serves your recordings to Claude and other MCP clients. No more copying transcripts by hand.

The recordings window

Home
Recordings over time: agenda, list, day, week, month and upcoming meetings from your calendar.
Library
Folders and subfolders. The folder is stored in the recording’s data, so the files on disk stay where they are.
The recording’s page
A player that skips 15 seconds and changes speed, with Summary, Transcript and Participants tabs. Click a timestamp to play from there.
Search
Covers titles and the text of meetings.

Files and sharing

A folder per meeting
Audio, transcript, speaker names and summary are ordinary files. The audio takes roughly 30–60 MB an hour, and keeping it can be turned off.
Share
Copy, save as Markdown or PDF, or send through the system.
Sending results
The summary goes to Notion, Telegram or your webhook by itself, if you set that up.
Your script after a meeting
The on_stop setting runs your command with the meeting folder as its argument.
Command line
notetaker checks your setup, starts and stops a recording, lists recordings and repeats processing.

The app itself

Set up in three steps
Access, transcription, then summaries and recording. One button asks for every permission.
Updates itself
A new version downloads by itself and the app offers to restart. Otherwise it updates at the next quit. Nothing happens during a recording.
Two languages, two themes
English and Russian interface, light and dark.

Try it on your next call

Installing takes a couple of minutes. After that the app keeps itself up to date.

Download for Mac

Free for now, version 0.8.13, 25 MB, macOS 14.2 or laterHow to install