How to take AI meeting notes without a bot joining your call

In 2020, "AI meeting notes" meant a bot joined your call. By 2026, it does not have to. The shift is technical (system-audio capture on consumer operating systems became reliable) and cultural (clients started noticing the bot and pushing back). This post is the practical version: why a bot is now a liability, how on-device recording actually works, and the exact stack to make AI meeting notes happen without a third-party joining your meeting.

Why the meeting bot stopped being free

• Optics — "Otter has joined the meeting" is your prospect's first impression. It is a brand awareness moment they did not opt into and a quiet vote against your product being thoughtful.
• Security — every meeting bot is a SaaS recorder of your live conversation. Regulated industries (finance, healthcare, defense, legal) increasingly disallow it. Even unregulated buyers ask why your "AI assistant" is a third-party participant in their pitch.
• Platform locks — Zoom Phone, MS Teams in some tenants, and many enterprise Google Workspaces actively block third-party recorders.
• Awkwardness — the bot is on the participant grid. People stop joking, stop pushing back, and the meeting gets more performative.
• Consent — the bot recording is on the recorder vendor's servers, which means consent is a vendor question, not a yours-and-theirs question.

What "on-device" actually means

On-device meeting capture records two audio streams from the host machine: system audio (everything you hear through your speakers) and your microphone (your voice). The two streams are mixed and streamed to a transcription endpoint. Nothing joins the meeting itself — the recording exists on your computer, the way a screen-recording of your laptop exists on your computer.

The mechanism varies per OS. On macOS 13+, Apple's ScreenCaptureKit exposes system audio via a documented API (with the user's explicit permission). On Windows 10+, WASAPI loopback capture exposes the same. Both require a single permission grant the first time, and both are inherently a single-participant operation — only the device running the capture has the audio, nobody else in the meeting does.

The trade-offs nobody mentions

You can only record meetings you are actually in

A bot can join meetings the bot owner is invited to but not attending. On-device cannot — your laptop has to be in the call. For most workflows this is correct (you should not be capturing meetings you did not attend), but it does rule out the "send my AI to listen" pattern that Fireflies markets heavily.

Bandwidth + storage live with you

You pay the audio storage cost locally (uncompressed: ~10MB/min; compressed: ~1MB/min). On a long week of meetings that is a few hundred megabytes. Most tools (Reline included) sync to cloud for backup, but the raw audio lives on your disk first.

Speaker attribution is harder without identity signal from the meeting platform

A bot can read the participant grid: "this is Sarah talking." On-device has to do speaker diarization from the audio alone. Diarization quality has gotten very good (we use; competitors use Deepgram or AssemblyAI) but it is not free — it adds ~200ms of latency and requires the speakers to differ enough in pitch/timbre.

The on-device stack you actually want

1. OS audio capture via the documented API (ScreenCaptureKit on macOS, WASAPI loopback on Windows). Avoid kernel extensions / VB-Audio Cable hacks — they break on every OS update.
2. Streaming transcription provider —, Deepgram, AssemblyAI, or Whisper running on-device for paranoid setups. Streaming gives you live captions; batch gives you final transcripts.
3. Speaker diarization — most providers ship this; the latency cost is ~200ms.
4. Summarization layer — Claude 4.x or GPT-4.x for the actual summary. Whichever you pick, demand citation-backed output (links back to the transcript segment that justifies each claim).
5. Storage + sync — cloud sync for the note, S3-class storage for the audio if you need long-term retention.
6. Sharing layer — explicit permissions, not "share link" (that is what creates the team-of-N problem).

Why Reline shipped the no-bot category

We built Reline because every other tool in the AI notepad category had committed to one of two choices: (a) ship a bot and accept the costs above, or (b) ship a notepad for solo users only. We thought there was a third option — bot-free, but with the team primitives (permissions, collaboration, RAG chat) that bots historically enabled by being a SaaS. The infrastructure exists in 2026 to do both: capture on-device, sync the note to a team workspace, keep the audio under the recorder's control.

When you should still use a bot

Honest answer: when you have to. If your workflow requires recording a meeting you cannot attend, a bot is your only option. If your meeting platform is one of the few that supports official cloud recording with API access (Zoom enterprise, certain Teams tenants), and your security team has signed off, a bot can be the easiest path. For everyone else — and especially anyone selling to security-conscious buyers — bot-free is the better default in 2026.

FAQ

Common questions

How does Reline record without joining the meeting?

We use the documented system-audio API on macOS (ScreenCaptureKit) and Windows (WASAPI loopback). One participant grants Reline permission once; from then on, Reline captures whatever the machine plays. The meeting platform never sees a Reline participant.

Will my client know I am recording?

That is up to you. Reline does not announce itself to the other side, but recording consent rules vary by jurisdiction (two-party consent states in the US require disclosure). We recommend telling your counterparties anyway — most appreciate the heads-up and it is good practice.

Does Reline work on Zoom, Google Meet, Microsoft Teams, and phone calls?

Yes — Reline records what your computer plays through the speakers and what your microphone hears. The meeting platform is irrelevant. Same for browser-based phone systems (Aircall, Dialpad, RingCentral).

Where is my audio stored?

Local first (on the machine that recorded it), then synced to encrypted cloud storage so you can access notes from other devices. You can configure regional data locality on Professional and Enterprise plans.

Can I record meetings I am not attending?

No — bot-free recording inherently requires you to be in the meeting. If you need to record meetings you cannot attend, a bot-based tool is your only option.