An AI note taker for product managers, on your own Mac

Product managers talk to a lot of people: customers, prospects, engineers, designers, sales, the leadership team. The useful part of those conversations is scattered across notebooks, docs and memory, and the customer’s exact words are usually the first thing to go.

An AI note taker keeps the words. This page covers how that fits into product work, what to watch for when the calls are with customers, and how Hearkeeper, a Mac app that is not released yet, is built to do it on your own computer.

Two kinds of meeting, two kinds of note

Most of a product manager’s meetings fall into two groups, and they need different things from notes.

Discovery: user interviews and customer calls. Here the point is evidence. You want the customer’s own words, the workaround they described, the moment they hesitated. A summary is a starting point; the transcript and the recording are the source.

Delivery: planning, reviews and cross-team syncs. Here the point is what was decided and who does what next. A clear list of decisions and to-dos, with owners and dates, matters more than a full transcript. Our guide on action items from meetings covers what makes those stick.

Hearkeeper handles both from the same recording. After the call, a language model on your Mac writes a summary, lists the decisions and pulls out the to-dos, with who and when. The full transcript, with speakers separated, sits alongside it.

User interviews without an extra guest

Many AI note takers join a call as a bot. In a user interview, that is one more name in the participant list, and it can change how openly someone talks about a product they are struggling with. Some people ask for the bot to be removed.

Hearkeeper records your microphone and your computer’s audio directly on the Mac. It works with Zoom, Google Meet, Microsoft Teams, Slack huddles, Webex and FaceTime, and with interviews in a room. The page on recording meetings without a bot explains how.

Removing the bot does not remove the need to tell people. Every Hearkeeper recording starts with a consent step: read a short notice aloud, post it in the meeting chat, or both. A consent receipt is saved with the recording, and a recording pill stays on screen, shows up in screen shares and cannot be hidden. When you share your screen for a prototype test, the participant sees it too. The guide on recording meetings with consent covers the rules.

If your company has its own research consent form or a policy on recording customers, follow it. The consent step supports that process; it does not replace it.

Find the pattern across calls

One interview is an anecdote. Ten interviews that mention the same problem are a reason to change the roadmap. The hard part is finding those ten.

• Full-text search across every recording finds the exact word: a feature name, a competitor, an error message a customer read out.
• Semantic search finds the idea when people describe it differently: “export is painful” and “I copy it into a spreadsheet by hand” turn up together.
• Ask a question across your meetings, such as “What did customers say about onboarding this month?”. Each part of the answer points to the second it was said, so you can play the clip before it goes into a deck.

That last part matters. A model that summarises twenty calls can smooth over the detail or get a quote wrong. Every line in Hearkeeper links to its moment in the recording, so the evidence behind a claim is one click away. You can also ask during a call with Command-Return, for example to check what the customer said about pricing ten minutes earlier.

Search is stored in one SQLite file on your disk, and it works offline.

Customer data stays on your Mac

A customer call can include things that are not yours to pass around: a customer’s revenue, their internal tools, a problem with their own customers, a name they asked you not to repeat. Many AI note takers send that audio to a vendor’s servers for transcription and summaries.

Hearkeeper does the work on your computer:

• Transcription on the Mac with whisper.cpp, accelerated by Metal. Live captions run about 2–4 seconds behind the speaker, and a final pass with Whisper large-v3 turbo replaces them after the call and separates the speakers.
• Summaries and answers from a local model, with no API key.
• Nothing uploaded by default. Nothing leaves your Mac unless you connect a team server or your own API key. With neither connected, the audio, transcript and notes stay on your disk.

That can make a security review or a customer’s question about where their data goes easier to answer. Check your company’s own rules on recording and storing customer conversations either way.

Get the notes into your tools

Product notes are only useful where the team works: the spec, the ticket, the research repository, the roadmap doc.

Hearkeeper’s notes export to Markdown, a plain text format that pastes cleanly into most docs and wikis and that any editor can open. A summary goes into the spec, the to-dos go into your tracker, and a set of quotes with their timestamps goes into the research write-up. Nothing is kept in a proprietary format.

Audio and notes are also ordinary files on your own disk, so they follow your own backup and retention rules.

Calls in more than one language

Transcription covers 99 languages. Icelandic is first-class, and calls that switch between Icelandic and English are detected automatically, with each part kept in its own language. That helps teams that interview customers in several markets. See Icelandic transcription.

What it is not

• Not a research repository. It records, transcribes, searches and writes notes. Tagging, synthesis boards and sharing with your team happen in the tools you already use.
• Not your analysis. The summary is a draft from a small local model. Deciding what the calls mean is still your job, and every line links back so you can check it.
• Not a shared workspace first. It is built for your own Mac; a team server is optional.

When is it available?

Hearkeeper is coming to the Mac first (macOS 14.2 or later, Apple silicon and Intel), with other platforms later. The price has not been announced. It is made in Reykjavík by Drangalia ehf.

If you want customer calls you can search and cite to the second, on your own computer, join the waitlist and we will email you when there is a build for you.