"AI notetaker" usually means a bot joins your call to write the notes.

Search for an AI notetaker and most of the results share one design decision: a bot logs into the meeting as its own participant, sits in the call alongside everyone else, and produces a transcript and a summary from inside it. That works, and for a team that needs the notetaker to schedule itself onto a shared calendar without anyone touching it, it is a real convenience. It is also the part of the category that a growing number of people specifically want to avoid — the bot shows up by name on the participant list, a client sometimes asks who it is mid-call, and every transcript it produces exists first on that vendor's own servers before anyone on the call sees it.

What goes wrong with the bot-based version

The bot itself is the recurring friction: it has to be invited or given calendar access, it occasionally fails to join and the meeting goes unrecorded, and it is visibly, namedly present in a way a human recording their own screen never is. The other friction is where the audio goes once captured — with Otter, Fireflies and Zoom AI Companion, the recording and the summary are both produced by a server the vendor operates, which is also where the model that writes the summary lives. Neither of those is made up: it is simply what "AI notetaker" has meant by default since the category started, and it is worth knowing before assuming every option in it works the same way.

The Mac's own audio, instead of a participant in the room

Coii AudioNotes is an AI notetaker built the other way: it reads what the Mac's own microphone and speakers already hear, as two separate tracks, rather than joining the call as anything. Nothing shows up on the participant list, and nothing has to be invited or granted calendar access — it runs the moment the meeting starts on the same machine that is already in it.

The AI part still happens — just on the Mac

"AI notetaker" describes what turns a transcript into a summary, and that step is still here: a language model bundled inside the app reads the transcript, along with anything typed alongside it, and writes minutes — decisions, action items, open questions — while the meeting is still fresh. The difference from Otter or Fireflies is not whether that step exists, but where it runs. A cloud notetaker's summary is written by a server that has to be reachable; this one is written on the same Mac that recorded the call, which also means it still works with no connection at all.

Speakers told apart without a bot doing the listening

A bot-based notetaker separates speakers because it is inside the call, hearing each participant's individual audio feed directly. Reading the Mac's own microphone and system audio instead means speaker separation has to work a different way — by voice print, matching each voice against the others in the same recording rather than against a named feed from a video platform. In practice this produces the same labelled transcript a bot-based tool does, without the bot itself.

A call where the bot has actually mattered

Picture a first call with a new client: the sales lead is talking, the client is asking pointed questions about cost, and everyone in the room is paying attention to the conversation rather than to who else is on the call. A named bot logging in a few seconds late, sitting on the participant list under its own name, is a small thing — until the client pauses and asks what it is. The call recovers, but the moment cost something: a beat of the client's attention spent on the software instead of the pitch. None of that happens when the notetaker is reading the laptop's own audio rather than sitting in the meeting as a guest; there is simply nothing on the roster to ask about, and the conversation stays the conversation.

What it deliberately doesn't do

Worth stating plainly, since "AI notetaker" gets used to describe some fairly different products. There is no calendar integration that joins a scheduled call automatically — recording starts when you start it, which trades a small manual step for nothing logging into a meeting uninvited. There is no shared team workspace, no link to send a colleague so they can read the notes from their own account — each recording lives on the Mac that made it, across up to three of your own machines per licence, not on a server other people can be granted access to. And it runs on macOS only; a team split across Windows and Mac laptops will not get a consistent experience from this specific tool, which is a fair reason to look at Otter or Fireflies instead if cross-platform access is the actual requirement.

Setting it up is smaller than the decision to switch

The first launch asks for two macOS permissions — microphone access, and permission to record system audio — the same prompt any Mac app requesting audio capture triggers, and that is the whole of it. No account to register, no team workspace to configure, no calendar to connect before the first call can be recorded. The bigger part of switching away from a bot-based notetaker is usually not the software at all — it is remembering to actually press record instead of relying on a bot that used to join automatically, a habit that takes a week or two of calls to fully replace.

What it costs, what it runs on

Coii AudioNotes is $19, paid once, for three Macs, with a 30-day trial that needs no card and no account. It runs on macOS 13 Ventura or later, on Apple Silicon or Intel — there is no Windows, Linux, iOS or Android version, which rules it out for a team not already on Macs. Against Otter's $8.33–$19.99-a-month tiers or Fireflies' comparable subscription, six people on Coii AudioNotes costs $114 once rather than roughly a thousand dollars a year; against Granola, the price gap narrows but the underlying architecture stays different — Granola shares the no-bot design but is still a cloud subscription, where this one is not. Full comparisons here, here and here.

Why "AI" and "cloud" got bundled together in the first place

It's worth asking why an AI notetaker running in the cloud became the default assumption at all, rather than treating it as inevitable. Most of this category started as a hosted product because running a capable language model used to require server-grade hardware nobody expected a laptop to have — so the summary step landing on a vendor's infrastructure was simply where the technology lived at the time. That constraint has loosened enough that a language model can now run bundled inside a Mac app, which is the actual reason "AI notetaker" and "sends your audio to a server" no longer have to mean the same thing, even though most of the category still assumes they do by default.

What switching away from a bot-based notetaker actually involves

Moving from Otter or Fireflies means giving up whatever integrations were already built around the bot — a shared workspace teammates could browse, a calendar connection that scheduled the bot automatically, an export into a CRM or a project tool. None of those exist here, and pretending otherwise would be dishonest about the trade. What is gained in exchange is a notetaker that never needs an invite, never shows up as a named guest, and produces its transcript and summary without a step depending on a server being reachable. Whether that trade is worth making comes down to whether the missing integrations were actually being used, or whether the team just inherited them by default the same way the cloud assumption got inherited.

The label matters less than the two questions underneath it

"AI notetaker" is a marketing category more than a precise technical description, which is exactly why it's worth breaking apart rather than shopping by the label alone, the way a search for "a smartphone" would be too broad to actually decide anything by itself. Two separate design decisions get bundled under that one phrase — whether something joins the call, and where the transcript gets turned into a summary — and a tool can land on either side of each one independently, as Granola's bot-free-but-cloud approach shows. Asking the two questions directly, rather than searching for "best AI notetaker" and trusting whatever ranking comes back, is the more reliable way to land on whichever combination of bot-or-not and local-or-cloud actually matches what went wrong with the last tool, rather than picking whatever wins the label this year.

Who this genuinely isn't built for

A team already living inside a shared workspace, with transcripts browsed by people who weren't on the original call and exported into a CRM automatically the moment it ends, gets real day-to-day value out of that infrastructure — value this tool does not attempt to replace, and would be dishonest to claim it does. This page is written for the person who searched "AI notetaker for Mac" specifically because the bot or the cloud step was already a problem, not as a claim that every team should want the same trade. If the shared workspace is what actually gets used every day, that is a reasonable point in Otter's or Fireflies' favour, not a gap to argue around — the honest comparison names what each side is actually trading away, rather than pretending one option wins on every axis at once.

Two questions worth separating before choosing one

"Does a bot join my call" and "does the recording leave my Mac" are different questions, and the four names above answer them differently: Zoom AI Companion and the bot-based tools answer no to neither, Granola answers no to the first and yes to the second, and Coii AudioNotes answers no to both. None of those is a wrong answer in every situation — a team that wants a notetaker to sit in a meeting nobody on the host side remembered to record manually has a real reason to prefer a bot that joins on a schedule. But for someone who searched "AI notetaker for Mac" specifically because a bot in a client call already went badly once, the distinction is the entire point. What a meeting recorder without a bot actually changes about that experience gets the fuller answer. The broader case for a Mac-native meeting notes app covers the rest of what the app does beyond the notetaker label. Full comparison against Zoom's own version here.