Four jobs behind your meeting notes

A captured meeting passes through four separate jobs: recording, transcription, speaker separation and note generation. A founder evaluating a first assistant can check what happens at each stage and where the data goes. Each stage produces something you can inspect: audio, text, speaker labels or a written account.
An AI meeting assistant captures speech, converts it into text, assigns speech to distinct speakers and produces a written account. In Noats, the microphone channel is attributed to you, while remote voices are separated into other speakers. Those labels let you follow contributions through the transcript.
In Noats, the path from captured audio to written note stays on your Mac by default, including transcription, speaker separation and the meeting files in your Application Support folder. You can evaluate the results by following one conversation through those stages.
Start with the Mac and its audio
Use an Apple silicon Mac running macOS 14.2 or later. Download the transcription and speaker separation models before your first test. With those models in place, transcription and speaker separation can work offline. You can then evaluate both functions using a conversation captured on that Mac.
Open an app that uses the microphone. Noats notices that microphone use and offers to record. Start the note to begin capturing the Mac’s audio and microphone. The offer and the start are separate events: you choose when capture begins, using the audio sources on your computer.
Transcription converts the captured speech into text. This is the general function of automatic speech recognition. For your evaluation, read a passage in the transcript and compare its wording with the corresponding audio. Check a name, a deliverable and a date that matter to the work.
Speaker separation groups speech by speaker. Speaker diarization assigns labels to distinguish speakers in a transcript. You can give each speaker label a participant’s name. Keep that distinction in mind when checking who owns a commitment in the finished account.
Check the names before the commitments
Read the transcript and name its speakers during or after the meeting. In Noats, those names belong to that meeting. Rename a speaker as part of your evaluation and check that the transcript lines attributed to that speaker update to match the correction.
The microphone rule also gives you a concrete setup check. Your side of the call comes through the Mac’s microphone, while remote audio is split into speakers you can name. Follow those labels when reviewing ownership of a task. Our explanation of speaker naming and corrections covers the per-meeting process.
Note generation produces a written account of the conversation. For a founder, useful details to inspect include a decision, the person responsible for a follow-up and the condition attached to a promise. These details give you something specific to compare between the transcript and the generated result.
Use a commitment as your test: compare who agreed to do what, by when and subject to which conditions. For example, if someone agrees to send a revised proposal after receiving customer approval, check that the generated account preserves that dependency. A deadline attached to the approval is different from a deadline attached to sending the proposal. Read the surrounding transcript lines before treating the commitment as ready for follow-up.
Choose where the written account is produced
For the data-location check, follow the meeting material through the app. Capture uses the Mac’s audio and microphone. The downloaded models handle the transcript and speaker labels locally, and the default model writes up the result there too. The resulting files remain in your own Application Support folder.
Noats stores audio, transcripts, speaker labels, notes, folders and chat history in a database and plain files in that folder. Meetings are not sent to a Noats server. This gives you a specific storage location to inspect alongside the processing choices you make in the app.
Local processing still sits alongside routine network requests. The app checks for updates through updates.noats.ai, retrieves the model catalogue when you open Models and fetches model downloads you approve. You can disable automatic updates in Settings under About. These requests have separate purposes from processing your meeting.
An optional cloud connection changes where the written account is generated. The required note and transcript content goes directly to the connected provider under its terms. Our guide to what leaves your Mac during note generation expands on local defaults and those direct requests.
Keep the model on your Mac for local processing, or connect Anthropic, OpenAI or OpenRouter with your own key. Choose where that final step runs before you use the result to assign work or follow up.
More from the blog
· 4 min read
Write a follow-up your customer can act on
Turn personal sales notes into a customer message that preserves decisions, open questions and commitments while removing internal observations.
Noats·Meetings
· 4 min read
Dictate a client follow-up from your meeting notes
Turn meeting notes into a spoken follow-up draft with Noats, then check names, amounts and commitments before sending.
Noats·Product
· 4 min read
Keep the commitment intact when the meeting becomes a note
Preserve decisions, owners, dates and the words that qualify a commitment. Follow one promise from the transcript to the written-up note.
Noats·Meetings