Youlisten. Ternlowrites.

Ternlo is a meeting notetaker for the Mac. It records both sides of the call, tells the speakers apart and writes the summary. Your own keywords go in too, if you type any.

Download for Mac

or install with Homebrew

brew install --cask ternlo/tap/ternlo

Free · v0.1.0 · macOS 14 or later · Apple silicon

No bot in the call. No account needed. You decide which models run: on your Mac or with a provider you choose.

No bot joins your call. Ternlo listens on your Mac: your microphone on one track, the call on another. You stay in the conversation and don't have to write it all down. Afterwards, each point of the summary links back to the moment it was discussed.

One meeting, start to finish.

Scroll through a weekly call the way Ternlo sees it.

  1. Press Record. 00:00

    Ternlo captures your microphone and your Mac's sound as two separate tracks, so your voice is never mixed with the other side. You can type a few keywords, but you don't have to.

  2. The transcript is live. 02:14

    The text is there while people still talk. So the chat, and your agent over MCP, can already read along and answer during the meeting.

  3. Afterwards, Ternlo works out who said what. 31:48

    It tells the voices apart on your Mac and uploads no audio for it. A model can suggest names, and you accept each one.

  4. Keywords and transcript become one summary. 32:10

    For meetings: a TL;DR, decisions and action items. Each statement links back to its time in the recording.

  5. Later, just ask. 3 weeks later

    Chat searches your notes by keyword and by meaning and cites the notes it used. Ask in one note, one folder, or across all of them.

Illustration of the app with an invented meeting.

More than a transcript.

Six things that set Ternlo apart from a plain recorder.

What was on screen is part of the notes.

Ternlo can take screenshots during the meeting, reads the text on them and matches it to what was said. So the summary knows which slide people meant. You release which screens it may use.

Your agents know your meetings.

Ternlo ships its own MCP server. Claude Code, Codex or Claude Desktop search and read your notes, even while a recording is still running.

  • 12 read-only tools
  • No audio
  • Off until you turn it on

Ask, and get the source.

The built-in chat searches one note, one folder or everything, reads what it finds and names the places it relied on.

AI tidies up after the meeting.

A model checks who really spoke, suggests names, gives the note a title and proposes a folder. You accept what fits and undo what does not.

The agenda comes from Outline.

Link a note to an Outline page. The summary and the name suggestions read the agenda, and you can append the meeting notes to the page.

Illustrations with an invented meeting.

And the small things.

A model per language
Choose which engine transcribes each language: Whisper or Parakeet on your Mac, or OpenAI, Deepgram and Groq with your own key.
Summary presets
Meeting, 1:1, lecture, interview, brainstorm, voice memo, or your own prompt.
Custom vocabulary
Add names, jargon and acronyms, and the transcription expects them.
Import and export
Import audio and video files or watch a folder. Export notes as PDF, Markdown or text.
Keyboard first
A command palette on ⌘K and a full keyboard map on the ? key.
Transcribe later
For a quiet Mac during the meeting, record now and transcribe on-device afterwards.
Too long for the model?
A long meeting is split and summarized in steps. Nothing is cut silently.
Light and dark
Follows macOS, or set it yourself.

This is the app.

Screenshots of Ternlo 0.1 with sample meetings. Drag the handle to compare light and dark.

A meeting note. Time stamps in the summary jump to the transcript. (Dark)
A meeting note. Time stamps in the summary jump to the transcript. The story above is a rebuilt illustration. These are screenshots.
The library: folders, badges and full-text search.
Home: recent notes and one button to record.

You decide what leaves your Mac.

Ternlo needs no account and sends no telemetry. Recording, transcription, speaker labels, search and summaries can all run on your machine. If you prefer a cloud model for a step, you add your own key. Below you see what each step sends.

Fully local

This Mac

  • RecordMicrophone and call audio
  • TranscribeWhisper or Parakeet
  • Tell speakers apartCommunity-1 or Nemotron 3
  • SummarizeOllama, LM Studio, llama.cpp
  • Search and askBuilt-in search, local chat model

No meeting content leaves your Mac.

With cloud models

This Mac

Providers you choose

  • RecordStays on this Mac
  • TranscribeAudio, your word list, the last lines → OpenAI, Deepgram or Groq
  • Tell speakers apartStays on this Mac
  • SummarizeNotes, transcript, agenda, released screens → Anthropic, OpenAI or Claude Code
  • Search and askYour question and what it reads → A chat model with your key

These steps can go out, with your own key. You choose for each step. Other steps can use a cloud model too, if you set one up: name suggestions read the transcript, and screen reading, with an optional pre-check, sees the screenshots. If you set no model for screens, the summary model reads them, when it can take images.

No account, no telemetry
Ternlo never contacts a Ternlo server. Notes and transcripts live in one database in your Library folder.
Keys stay in the Keychain
API keys go into the macOS Keychain, in one encrypted entry. They are never stored in plain text.
You release what was on screen
Ternlo can take screenshots during a meeting. The screen model reads them and suggests which matter. By default the summary uses only the ones you release; at the YOLO level the model picks them. Stricter levels keep them on this Mac until you release them.
Keep the audio, or don't
Recordings are kept compressed, about 37 to 51 MB per hour. Turn off Keep audio and new recordings keep none. Delete stored audio removes the rest. Transcripts stay.
Updates come from GitHub
A downloaded Ternlo asks GitHub for updates. The check carries no meeting data, and you can turn it off.
Models come from Hugging Face
Local models are downloaded once, when you choose them. After that they run on your Mac.

Questions

Is Ternlo free?

Yes. You can download and use the app free of charge, for personal and professional work, on any number of your own Macs. If you choose a cloud provider for transcription or summaries, that provider bills you directly.

Does a bot join my meetings?

No. Ternlo records your microphone and the sound your Mac plays. It works with Zoom, Google Meet, Teams or any other app, and with people in the same room. Please tell people before you record them.

Do I need an API key?

Not for recording, transcription or speaker labels: these run on your Mac. For summaries you need one of three things: a local model server such as Ollama or LM Studio, Claude Code on your own plan, or an API key from Anthropic or OpenAI. Chat works with a local server or an API key.

Which Macs does it run on?

Macs with Apple silicon and macOS 14 or later.

Where is my data?

In one SQLite database under ~/Library/Application Support/com.ternlo.macos/. Kept audio is stored next to it.

How do updates work?

Ternlo checks GitHub for a new version, downloads it in the background and installs it when you choose Restart to update. You can turn the check off in Settings. With Homebrew: brew upgrade --cask --greedy ternlo.

Is Ternlo open source?

No. The source code is not public. The app is free to use, and it works without any Ternlo service.

Install Ternlo.

Free for personal and professional use on your own Macs.

Download the DMG

v0.1.0 · 51 MB · notarized by Apple

brew install --cask ternlo/tap/ternlo
  1. Open the DMG and drag Ternlo to Applications.
  2. Open Ternlo. A short setup asks for permissions, language and models.
  3. Press Record in your next meeting.

Requires macOS 14 or later on Apple silicon. Release notes

Verify the download

SHA-256 of the DMG

e6be9b0c8da4c515b9d65357dbe54ba9b6fb64aca9c51481e762ae21ce68e5af