Blog ·
How to use an AI assistant in meetings without a bot joining the call
Almost every AI meeting tool works the same way: you connect your calendar, and when the meeting starts, a bot dials in. It appears in the participant list with a name like "Otter.ai Notetaker". Someone has to admit it from the waiting room. On many platforms its arrival triggers a recording notice to everyone present.
That model is a good fit for a team that has agreed to record its meetings. It is a poor fit for a lot of ordinary situations: a client call where a visible recorder changes the tone, a first sales conversation, an interview, or any meeting you joined but do not host.
The two architectures
There are really only two ways to get a machine to hear your meeting, and the difference between them explains almost every other difference in behaviour.
A bot participant connects to the meeting through the platform’s API, as a client. Because it is a real participant, it gets a clean per-speaker audio feed and can keep recording after you close your laptop. Because it is a real participant, it is also visible, admissible, ejectable, and subject to the platform’s recording notifications.
A desktop overlay runs on your own computer and captures the system audio — the same sound that reaches your speakers — plus your microphone. The meeting platform is not involved at all and has nothing to announce. It cannot record without you present, and speaker separation is harder because everyone else arrives on one mixed channel.
What each approach is actually good at
- Bot participant: unattended recording, clean speaker labels, long archives, team-wide meeting libraries. Best when recording is expected and agreed.
- Desktop overlay: nothing appears in the participant list, works identically across every platform (including in-person conversations), and answers you during the call rather than summarising it afterwards.
- Neither removes your obligation to follow the recording and consent laws that apply where you and the other participants are.
Why platform independence follows from the architecture
A bot needs an integration per platform: Zoom, Teams, Meet and Webex each require separate engineering, separate approval, and separate maintenance when APIs change. An overlay captures audio below all of that, at the operating-system level. Zoom, a Slack huddle, a phone call on speaker, a conversation with someone sitting across the table — they are all just sound arriving at the sound card.
That is why overlay tools tend to advertise no integration list. There is nothing to integrate with.
The latency difference matters more than it sounds
Most bot-based tools are built around the summary: the value arrives in your inbox afterwards. That is genuinely useful for records and follow-ups, and useless in the thirty seconds where you needed the number, the objection response, or the name of the thing you half-remember.
Streaming transcription changes what the tool is for. Text appears while the sentence is still being spoken and an answer can begin generating before the speaker finishes. The output is not a record of the meeting. It is a participant in it.
Choosing between them
If you want an archive of team meetings that everyone has agreed to record, use a bot. It is the better tool for that job and it is not close.
If you want help during the conversation — a live transcript you can glance at, an answer to a question that was just asked, the ability to ask about a slide someone is sharing — you want something running on your own machine, above the window, that the meeting never knows about.
Arguium takes the overlay approach: system audio capture, live transcription, and answers on screen during the call, with no bot in the participant list. It is free to download for macOS and Windows.