An AI notetaker is software that records a meeting or a call, turns the audio into a written transcript, and generates a summary — usually decisions made, action items assigned, and questions left open — without someone typing any of it by hand. The category covers everything from a bot that joins a video call as a named participant to an app that simply records whatever audio a computer can already hear.
Two different ways of getting the audio
Some AI notetakers, including Otter and Fireflies, add themselves to a meeting as a visible participant, using a separate audio feed from the call platform to produce the transcript and to label who said what. That approach works reliably across whichever video platform a meeting runs on, at the cost of a named entry on the call's participant list — a detail that matters more in a client call or a job interview than it does in an internal standup.
A second approach, used by Granola and by Coii AudioNotes, skips the bot entirely: the app records the computer's own microphone and the audio the computer is already playing, and produces the transcript from that instead of a feed supplied by the meeting platform. Nothing joins the call as a participant either way, but where the recording goes afterward still differs — Granola's recording is processed on a server to produce its notes; Coii AudioNotes processes it on the same Mac that made it.
What the summary is actually built from
The transcript is the intermediate step, not the final product most people want. An AI notetaker's real value is in what it does with that transcript afterward — sorting a rambling hour of conversation into a short list of what was decided, what someone owes another person by when, and what's still unresolved, so nobody has to re-listen to the whole recording to reconstruct it later.
Where the audio goes to get there
This is the detail worth checking before choosing one, and it varies more than the feature list usually makes obvious. Most AI notetakers are cloud services by construction — the audio is uploaded to a server to be transcribed and summarised, and several vendors' own feature lists mention an option to opt out of using that audio for model training, which is a tell that the default is to use it. A local-first tool instead processes the recording, the transcript and the summary entirely on the device that made the recording, with nothing uploaded to produce any of the three.
Where this sits against a plain transcription tool
A transcription-only tool stops at the written-out words; an AI notetaker goes one step further and organises those words into something meant to be read once and acted on, rather than searched through. That extra step — structuring a transcript into decisions and action items — is what separates the category from a dictation app doing the same underlying job of turning speech into text.
Coii AudioNotes is one example of the local-first kind: it records a Mac's microphone and system audio as two separate tracks, transcribes segment by segment while the meeting runs, and writes the summary once it ends — all on the Mac, with a full comparison against Otter and against Granola covering the differences in more detail, and the broader case for an AI notetaker built for Mac and a round-up of notetakers that skip the bot entirely covering the rest of the category.