Four things stand between you and a working install.

Three things to know before you download it

It is not notarized.

There is no Apple Developer account behind this, so there is no certificate to sign it with and nothing to submit to Apple's notary service. It carries an ad-hoc signature, which proves the bundle has not been altered since it was built and proves nothing at all about who built it. macOS attaches a quarantine flag to anything a browser downloads, and on a quarantined app with no notarization Gatekeeper refuses the first double-click outright and offers you the Trash.

The way through, on macOS 15 and later: double-click the app, let macOS refuse, then open System Settings › Privacy & Security, scroll down to the Security section, and click Open Anyway beside the message naming the app. Confirm once and it launches from then on. Older instructions tell you to Control-click and choose Open. Sequoia removed that override for apps that are not notarized, so it no longer does anything. If vouching for a stranger's binary is not a trade you want to make, build it from source instead and the quarantine flag never gets attached in the first place.

It needs the claude CLI, logged in to a paid subscription.

Transcription is local and costs nothing. The notes are a different matter: they are written by a model, through the claude command line tool, spending whatever Claude subscription you already pay for. One detection pass at the start, a short pass every few minutes during the lecture, one full pass over everything when you stop. That comes out of your normal usage limit; it is not a separate pool and it is not unlimited. Without a login you get a recording and a transcript and no notes at all. This is the requirement people miss.

First launch downloads about 600 MB of models.

The CoreML speech models are fetched once, on first run. Do that at home the night before rather than in the ninety seconds before a lecture starts. There is a whole download screen for it, which exists so you find out early instead of finding out in a lecture theatre.

Getting it running

Download the zip from the release, unzip it, drag LectureNotes.app into /Applications, then read the first section on this page again before you double-click it, because it will not open. If you would rather not go through System Settings, strip the flag yourself:

xattr -dr com.apple.quarantine /Applications/LectureNotes.app

Either way you are the one vouching for the binary. Building it yourself avoids the question entirely, and takes about a minute:

git clone https://github.com/toyeshhm/lecture-notes-app.git
cd lecture-notes-app
brew install xcodegen
make app

And if you do not have the CLI yet:

npm install -g @anthropic-ai/claude-code
claude            # then type /login

What you need

An Apple silicon Mac

M1 or later. The speech models run on the Neural Engine through CoreML. There is no Intel build and no CPU fallback.

macOS 14 or later

Sonoma. On Sequoia and later, the Open Anyway route described above is the only one through the interface.

The claude CLI, logged in

A Pro or Max subscription is what makes this free at the margin. An ANTHROPIC_API_KEY works too, but it bills per token.

An Obsidian vault

Or any folder at all. The notes are plain markdown with YAML frontmatter. Obsidian is where they are pleasant to read, not where they have to live.

About 600 MB of disk

One time, on first launch, for the models.

Expect to grant microphone access more than once if you build from source. macOS keys that permission to a code signature, an ad-hoc signature is regenerated on every build, so a rebuild can produce something macOS considers a different app. It looks exactly like a bug and it is not one.