The transcript is the record. The minutes are what you act on.

Nobody reopens a forty-minute transcript to remember what got decided. What actually gets read again — by you, three days later, or by someone who wasn't in the room — is a short list: what was decided, what someone now owes, and what's still an open question. That list is the minutes, and turning a transcript into one is the part of note-taking most worth automating, because doing it well by hand takes almost as long as the meeting did.

Where the minutes come from

Coii AudioNotes transcribes on your own Mac, segment by segment, for as long as the recording runs. Once there's a transcript — whether the meeting has just ended or you're revisiting one from last month — a language model bundled with the app reads through it and produces a structured summary: the decisions that were made, the actions that came out of it and who they belong to if that was said out loud, and the questions that were raised but not answered. That generation happens on the Mac in front of you, the same machine that did the transcription, rather than being sent to a server somewhere to be summarized and sent back.

This is worth pausing on, because it's the part most cloud notetakers can't say about themselves: if the network drops, or you're finishing a summary on a flight, the step that turns a wall of dialogue into minutes still runs. A tool whose summarizing step lives on a server owes you notes later, once it's reachable again. Nothing here is waiting on a connection.

Reading the minutes against the transcript

Automatic summarizing is good at compressing what was clearly stated and bad at guessing what was implied but never said aloud. If a decision was made with a nod and a "yeah, let's do that" rather than a clear sentence, it usually still gets caught — but a decision that was only half-made, left dangling as "we'll figure that out later," correctly shows up as an open question rather than something invented to fill a gap.

The transcript is still there underneath the minutes, in full, so checking the summary against it takes seconds: click through from a decision or action item to the point in the transcript it came from, read the surrounding lines, and either confirm it or add what's missing. This is the opposite of a tool that gives you a summary and discards the source — here, nothing about the original conversation goes away once it's been condensed.

Making the minutes more useful before you even see them

A few habits during the meeting make a real difference in how much editing the minutes need afterward:

Say decisions out loud, even ones that feel obvious in the room. "So we're going with option two" is a sentence a transcript-reading summarizer can act on. A decision that only exists as everyone nodding at the same time leaves nothing for either a transcript or a human note-taker to point to later.

Name the owner when you assign something. "Priya, can you send that over" attaches an action item to a person automatically — and because the app has already told the speakers apart by voice, that name lands against the right line in the transcript too. "Someone should send that over" produces an action item with no owner, which is technically accurate and much less useful three days later.

Type your own flag during the call for anything you already know matters. A short note typed while the meeting is happening — even just "important: pricing change" — is read alongside the transcript when the minutes are generated, which is the fastest way to make sure a specific point survives the summarizing step exactly the way you meant it.

Why the summary can run without a network connection

It's worth being specific about why this particular step works offline, because it's easy to assume any AI-generated summary must be talking to a server somewhere. The language model that reads the transcript and produces the minutes ships as part of the app itself, installed on your Mac along with everything else, rather than being an API call to a service running elsewhere. That's what "bundled with the app" means in practice: the same install that lets you record a meeting on a plane also lets you get minutes out of it before you've landed, because the step that turns a transcript into a summary was never depending on being reachable to begin with.

This is a genuinely different architecture from a cloud notetaker's summary feature, not just a different sales pitch about privacy. A tool whose summarization happens on a server has to send your transcript there and wait for a response, which means the summary is only ever as available as that server is — and it means the transcript, even briefly, existed somewhere other than your own machine on its way to becoming minutes. Here, the transcript never has to leave in the first place for the summary to get generated.

Editing the result

The generated minutes are a draft, not a final document, and editing them is expected rather than a sign something went wrong. Add a decision the summary missed, correct an action item's owner if a name was mentioned but attributed to the wrong voice, or delete something that turned out not to matter once the meeting moved on from it. Because the underlying transcript never changes, you can always regenerate or re-check the summary against the source without worrying that the original record has been altered by your edits.

Two jobs where this matters in a specific way

An engineering manager running four one-on-ones back to back has a particular version of this problem: each conversation produces one or two action items, and by the fourth call of the day it's easy to lose track of which commitment came from which report. Minutes generated per meeting, each with its own decisions and action items attached to a specific person and a specific date, solve exactly that — you're not relying on a single running notes document to keep four separate conversations straight.

The same shape applies to anyone running structured one-on-ones as a standing part of the job rather than an occasional meeting. When the format of the conversation repeats every week, having a consistent minutes structure — decisions, actions, open questions — makes it easy to scan last week's minutes for a two-second refresher before this week's conversation starts, rather than re-reading a full transcript to remember where things were left.

A third job with the same pattern

A product manager sitting in customer calls has a version of this problem that looks different on the surface but is the same underneath: a single call might surface a feature request, a complaint about a specific workflow, and an offhand comment about a competitor, and none of those three things are equally important to act on. Minutes that separate "decisions" from "open questions" give a natural place to put "the customer wants X" as a question worth raising internally, rather than losing it in a transcript alongside forty other sentences from the same call. Read across a week of customer calls, four or five sets of minutes are a much faster way to spot a pattern than four or five full transcripts would be — the compression is what makes the pattern visible in the first place.

Common mistakes when relying on generated minutes

Treating the minutes as the only record you keep. They're the fast read, not the only one. If a dispute ever comes up about what was actually said — a client insisting they never agreed to a change order, a report saying a commitment was phrased differently than the action item suggests — the transcript is the document that settles it, not the summary. Delete the transcript and keep only the minutes, and you've thrown away the part that was actually evidence.

Not correcting a wrong owner once you notice it. An action item attributed to the wrong person because of a speaker-labeling slip is easy to miss on a quick skim and awkward to discover later, when someone is asked about a commitment they never made. It takes seconds to fix once you've noticed it, so noticing it during the first read-through — while the meeting is still fresh — is worth the extra care.

Expecting the minutes to capture something nobody said. If a decision was made through a shared document or a follow-up email rather than out loud on the call, no summarizer working from audio can know about it. The minutes are a record of the conversation, not of everything that happened around it.

Minutes for a recurring meeting versus a one-off

A weekly status meeting and a one-time negotiation call want slightly different things from their minutes, even though the underlying mechanism is identical. For a recurring meeting, the value is in the pattern across weeks — being able to open last week's minutes for a fifteen-second refresher, and this week's minutes six days from now for the same reason. For a one-off call, the value is closer to a receipt: a clear, quotable record of what was agreed, generated the same afternoon rather than reconstructed from memory a week later when someone asks what was decided. The app doesn't need to know which kind of meeting it's summarizing — the structure of decisions, actions and open questions works for both — but knowing which one you're in changes how soon you're likely to reread it, and how carefully worth double-checking it is against the transcript before you rely on it.

How long this takes in practice

The generation itself happens in the background while you're already moving on to the next thing — there's no separate export step where you wait for a summary to come back, and no queue behind other people's meetings on a shared server, because there is no shared server involved. The practical time cost is entirely on your side: the minute or two it takes to skim the result against the transcript, catch the one thing worth adding, and move on. For a short daily standup that's often nothing more than a glance; for a long negotiation call, it's worth the full read-through before anyone acts on what it says was decided.

What minutes leave out on purpose

A good summary is not a shorter transcript — it's a different document with a different job. Small talk, tangents that didn't lead anywhere, and the back-and-forth of someone thinking out loud before reaching a decision are all things the minutes are supposed to compress away, not preserve. If you find yourself wanting one of those details back, that's what the full transcript underneath is for; the minutes were never meant to replace it, only to sit on top of it as the version you actually reread.

Minutes as a shared artifact, without a shared account

There's no cloud workspace behind this — no shared folder, no team seat, no link you send someone that opens on their end. What you have instead is a document on your own Mac: the transcript, the minutes generated from it, and anything you've edited into either. If you need to hand it to someone else, you copy the text or the file the same way you'd hand over any document you wrote yourself. That's a real limitation against a tool built around a shared workspace, and worth knowing before you count on this for a team that lives inside shared meeting notes — but it's also the same reason nothing about what was discussed in that meeting ever left the room to get there.

What good minutes actually save you

The value isn't the minutes themselves — it's not having to be the person in the meeting who's simultaneously listening and typing a summary in real time, and not having to be the person three weeks later scrolling through a forty-minute transcript looking for the one sentence where somebody agreed to something. A transcript answers "what was said." Minutes answer "what do I do now." Getting from one to the other in the time it takes to read four lines is the entire point, and it's worth trying against a week of your own real meetings during the 30-day trial rather than a single demo call, since the pattern of what the summarizer catches and what it needs a nudge on tends to show up after the third or fourth meeting, not the first.