Once voice coding has given your agent a voice, the terminal stops being the only room it can be useful in. Meetings are the obvious next one, and there are two genuinely different ways to handle them.
- Recording locally: nothing joins the call, audio never leaves your machine, free and unlimited.
- Sending your agent in: it appears as a participant or a silent notetaker. This one costs money, because it has to run somewhere that isn’t your laptop.
- Which you want depends on whether you’re in the meeting.
You’re in the meeting: record it locally
If you’re attending anyway, you don’t need a bot in the call, you need the audio that’s already reaching your machine.
The local recorder captures your microphone plus system audio from any meeting app, transcribes it on-device with speaker tagging, and hands the transcript to your agent to write the summary and action items. No participant appears in the call. Nobody is prompted to admit a bot. Nothing is uploaded.
That last point is the one that matters in practice. Plenty of meetings are ones you’d never be allowed to put a third-party notetaker into, anything with customer data, anything under NDA, anything involving your own compensation. A local recorder sidesteps the question entirely, because there is no third party.
It’s free and unlimited, for the same reason the rest of the on-device loop is: it isn’t costing anyone server time.
The cheapest bot is the one that never had to join.
You can’t make it: send the agent
The other case is genuinely different. The meeting is happening, you can’t be there, and there’s no audio arriving at your machine because you aren’t in the room.
For that, paste the meeting URL into the widget and your agent joins the call itself, Zoom, Meet or Teams, as a visible participant. You choose which kind:
- Agentic: it speaks, answers questions and takes part.
- Transcript only: it sits silently and takes notes.
Either way a diarized transcript comes back, tagged by who said what, and your agent writes it up afterwards.
This is the one metered piece of SKI, and the reason is structural rather than commercial: a bot in a live cloud call has to run on a machine that stays connected for the duration, and that machine can’t be your laptop. There’s a free allowance and metered credits beyond it. Everything on-device stays free regardless.
Choosing between them
The decision is simpler than it looks:
- Are you attending? Record locally. Free, private, no bot etiquette to negotiate.
- Are you not attending? Send the agent, and pick silent notetaker unless it genuinely needs to speak.
- Is the content sensitive? Local recording, and check your own policy before any bot joins anything.
Why this is worth setting up at all
The tooling question is the boring half. The reason people bother is what it does to the meetings themselves.
The recurring meeting you attend for four minutes of relevance. Every team has one: a standing sync where nine tenths of it does not touch your work, but you go because the tenth might. Sending a silent notetaker and reading what mattered afterwards gets that hour back, and you stop deciding between being informed and being productive.
The handover that gets lost. Decisions made out loud have a short half-life. Someone says “we will keep the old endpoint until Q3”, three people nod, and six weeks later nobody can remember whether that was agreed or floated. A speaker-tagged transcript that your agent can read turns that from memory into a lookup.
The timezone problem. If half your team is asleep during the call, the notes are the meeting for them. Written by an agent that also has your repository in context, they arrive as something actionable rather than a wall of text somebody has to translate.
The meeting you genuinely cannot attend. Two calls at once, a school run, a flight. The choice used to be reschedule or lose the thread. Now it is neither.
The best outcome is not a better transcript. It is fewer meetings you have to physically sit through.
There is a collaboration effect underneath all of that. When the record is reliable and everyone can query it, meetings stop being the place where information is stored and become the place where decisions are made. That tends to make them shorter, because the ritual of recapping last week for whoever missed it stops being necessary.
A note on the social part
Worth saying because the tooling conversation usually skips it: a bot appearing in a call is a visible act. People notice, some people mind, and in several jurisdictions recording participants without telling them is not merely rude.
Name the bot something honest, it joins under whatever name you give it, and say what it’s there for. “I can’t make this, my notetaker is joining” costs you one sentence and removes the entire ambiguity.
The local recorder avoids that conversation, which is a real reason to prefer it when you have the choice.
Both live in the same widget as the rest of the workflow, and the transcript ends up somewhere your agent can already read it, which is the actual point. A meeting summary you have to go and fetch is just another tab.
