AI Assistant for Zoom Meetings

There are two ways to add an AI assistant to a Zoom meeting. You can connect a cloud bot that joins the call as a visible participant, or you can run a desktop assistant that captures your computer's system audio and never appears in the attendee list. The bot route is simpler for shared team recordings. The desktop route is better when the call is private and you would rather not announce a recorder to the room.

I am the founder of Natively, an open-source (AGPL-3.0) desktop AI assistant for meetings and interviews. This is the setup I actually use on Zoom every day, plus the honest tradeoffs of each approach and how to get help while the call is live instead of just a summary afterward.

Bot in the call vs desktop capture

Zoom has its own built-in AI Companion, and plenty of third-party bots will join your meeting to record and summarize it. They work by entering the call as a participant, which means everyone sees a new attendee show up, usually with "notetaker" or a product name in the label.

A desktop assistant works differently. It captures the audio your computer is already playing and sending, at the operating-system level, so there is nothing to admit to the call. Natively uses a Rust-based system audio capture that works across any desktop app, including Zoom, Teams, Meet, Slack, and Discord. The people you are talking to see the same participant list they would with no tool running at all.

Setting up Natively on Zoom, start to finish

The whole point is that there is almost no setup inside Zoom itself. You do not install a Zoom app or grant a bot access to your account. You run a desktop program alongside Zoom and it listens to what your machine is already doing.

Open Natively before the meeting starts. Grant the one-time microphone and system-audio permission your operating system asks for. Join your Zoom call the way you always do. That is the setup. The assistant begins transcribing as soon as people start talking, and its window sits on your screen where you can read it but a screen share cannot.

On Zoom specifically, one detail matters: system audio capture picks up the other participants because their voices come through your speakers or headphones. Your own voice comes through the microphone. Natively handles both, so the transcript has both sides of the conversation, whether you are on speakers or a Bluetooth headset.

Zoom AI features that actually matter

A lot of Zoom AI marketing is about summaries. Summaries are table stakes now. The feature that changes how a call feels is help that arrives while you are still in it.

The gap between hearing a hard question on a Zoom call and starting a good answer is usually a few seconds of silence you would rather not have. That is the window a real-time assistant is built for. Natively puts a relevant answer or a next question on screen in under 500 milliseconds on my hardware, which is fast enough to use before the pause gets awkward. Live transcription means you can also scroll back three sentences when someone references a number you half-caught, without asking them to repeat it.

For the after-call half, the notes matter more than the raw transcript. A wall of text is not a deliverable. Decisions, owners, and action items are. If you want the reasoning behind that, the complete guide to AI meeting assistants covers how to tell a real notes feature from a transcript with a summary bolted on.

What about Zoom's own AI Companion?

Zoom ships its own AI Companion now, and for a lot of people it is the obvious first thing to try. It summarizes the meeting, drafts follow-ups, and answers questions about what was said, all inside Zoom with nothing extra to install. If your whole team already lives in Zoom and your admin has it enabled, it is a reasonable default and I would not talk you out of it.

Two things make me reach for a separate assistant anyway. The first is where the data goes. AI Companion processes in Zoom's cloud under Zoom's account controls, which is fine for internal syncs and a harder sell for a confidential negotiation or a candidate interview. A desktop assistant that can run locally or on your own key keeps that conversation off a shared cloud entirely.

The second is that AI Companion is built around the recap, not the live moment. It is good at telling you what happened. It is not sitting on your screen feeding you an answer while the other person is mid-sentence. If the value you want is help during the call rather than a tidy summary after it, that is a different tool, and the two can coexist. I have run both on the same call: Zoom's recap for the shared record, a private assistant for the real-time help only I see.

When capture goes wrong, and how to fix it

System audio capture is reliable, but two setups trip people up on Zoom, and both have quick fixes worth knowing before an important call.

The first is the missing far side. If the transcript only shows your own words and not the other participants, the assistant is getting your microphone but not the system audio playing through your speakers. On macOS this is almost always the one-time screen-and-audio permission that has not been granted yet. Approve it in system settings, restart the assistant, and both sides appear. Do this once, well before the call, not while a client is waiting.

The second is Bluetooth headsets switching audio modes. When some wireless headsets activate their microphone, they drop to a low-quality call profile that also thins out the audio the system can capture. If accuracy dips the moment you connect a headset, wired headphones or your laptop speakers will usually transcribe cleaner. It is a hardware quirk, not a software one, and it affects every capture-based tool the same way.

Consent on Zoom is not optional

Being invisible to other participants is a privacy feature for you. It does not change your obligations to them. Recording and transcription consent laws vary by location and by whether the call is internal or external, and an assistant that no one can see does not exempt you from them.

My rule of thumb: if I would feel odd about someone knowing I kept a transcript, I ask first. For internal calls that is usually a non-issue. For external calls, interviews, and anything regulated, follow your company policy and applicable law. The note-taker privacy guide goes deeper on where transcripts live and how to keep control of them.

Frequently asked questions

Can I use an AI assistant on Zoom without a bot joining?

Yes. A desktop assistant like Natively captures Zoom's audio at the system level, so no bot appears in the participant list. The other people on the call see no extra attendee.

Does an AI assistant work with Zoom on both Mac and Windows?

Natively runs on macOS 12 and later and on Windows 10 and 11. It captures system audio the same way on both, so Zoom behaves identically. Linux is not supported yet.

Will the assistant show up if I share my screen on Zoom?

No. Natively's overlay is designed to stay hidden from screen share, so you can present in Zoom while still reading the transcript and answers yourself. Only you see it.

Is it free to use an AI assistant on Zoom?

Natively is free if you bring your own API key or run a local model through Ollama, paying only for the tokens you use or nothing at all locally. There is also a one-time Pro purchase and a managed API tier for people who want no setup.

Try it on your next Zoom call

Start with one call where the stakes are real, a client meeting or an interview, and see whether live help changes how prepared you feel. That is a better test than a standup where notes barely matter.

If you want live transcription and answers on Zoom without a bot in the room, Natively is free to try. On Google Meet instead? The Google Meet guide covers the same setup for that platform.