AI Meeting Notes Without a Bot on Zoom and Teams: The Desktop App Buyer's Guide (Granola, Fireflies, Basil Compared)

Published October 29, 2026

October 29, 2026 · 11 min read

If you want AI meeting notes from Zoom and Microsoft Teams without a bot joining your call, you need a desktop app that captures your computer's system audio directly — not a cloud bot invited via your calendar. The 2026 market has four serious options: Granola (the category leader on Mac and Windows), Fireflies.ai's new "Take Notes" desktop mode, tl;dv, and Fathom's bot-free beta. A fifth, Basil AI, is the only one on this list where both the audio capture and the transcription happen 100% on-device. This guide explains how each works, the trade-offs, and the compliance questions buyers keep missing.

Why "no bot" is suddenly the question everyone is asking

Three things changed in 2026. First, meeting hosts and compliance teams started pushing back on the awkwardness of a "Fireflies.ai Notetaker" or "Otter.ai" tile sitting in the attendee list of a client call. Second, Google Meet began flagging third-party bots as a security risk — tl;dv's April 2026 analysis notes that Fireflies is among the tools now blocked from joining Google Meet calls entirely depending on org settings. Third, every major vendor shipped a bot-free desktop mode in response.

The structural choice is simple. A bot route invites a visible participant into your Zoom or Teams meeting and pipes audio to the vendor's cloud. A desktop route runs an app on your Mac or Windows machine that captures the audio your speakers are already playing — nothing new appears in the roster. As heynatively's 2026 Zoom assistant breakdown frames it, the desktop route is better when the call is private and you'd rather not announce a recorder to the room.

How bot-free desktop capture actually works

Every bot-free desktop app does the same thing under the hood: it taps your operating system's audio output (the sound your speakers would play) plus your microphone, and runs those two streams into a transcription pipeline. There is no API integration with Zoom or Teams. There is no calendar-triggered bot. The app simply listens to what your computer is already hearing. Layer3Labs' explainer on Granola describes the model bluntly: Granola records what your speakers play and what your microphone picks up, then transcribes that stream in real time.

Because there is no bot in the participant list, remote attendees see no visible sign that a note-taker is running — which, as the same analysis points out, is the entire point of the design and also the reason consent questions follow it.

Granola: the benchmark bot-free desktop app

Granola is the tool buyers typically compare everything else to. It runs natively on Windows 10 and later, plus macOS, iOS and Android, capturing Zoom, Teams, and Google Meet audio from the device. Setup is advertised as under five minutes: download, connect your calendar, start your first meeting.

The capture model is explicit. Per Granola's own product page, it "uses your computer audio, so doesn't invite a bot" and works with Zoom, Google Meet, Teams and every other meeting app. In the Windows FAQ Granola spells it out: "There is no meeting bot. Granola runs only on your computer and uses your system audio and microphone."

The architecture is often described as "local-first," but it is important to read what that actually means. According to Granola's engineering post on where it runs, capture happens on your device, but transcription and AI summaries use secure cloud processing, and notes and transcripts are stored encrypted in a US-hosted AWS environment. The audio file is deleted after transcription (zero-audio-retention), but your transcript lives in Granola's US cloud, and Voibe's October 2026 review notes that notes are kept indefinitely by default unless you configure a retention policy.

The BuildBetter privacy policy review from October 2026 adds important context: Granola trains on de-identified data by default (opt-out available), stores data in the US only (no EU residency), and faces a pending wiretap class action. Enterprise plans add SOC 2 Type II certification (achieved July 2025), SSO, SCIM, and HIPAA-compliant workspaces per Tooliverse's 2026 pricing breakdown at $14/user/month on Business.

What Granola does well

• True platform-agnostic capture — same desktop app handles Zoom, Teams, Meet, Webex, Slack huddles, and in-person meetings.
• Human-guided notes model: you jot bullets during the call, AI enhances them from the transcript afterward.
• Audio deleted immediately after transcription (verified by SOC 2 Type II scope).

Where Granola stops short

• Transcription still happens in US AWS cloud — "local" refers to audio capture, not AI processing.
• Notes kept indefinitely by default; data residency is US-only.
• Model training is opt-out, not opt-in, on Free and Business tiers.

Microsoft's answer: "Recap without saving transcript" (and why it isn't bot-free capture)

If you work inside a tightly governed Microsoft 365 tenant, you may have heard about Microsoft's new Copilot feature. It is important to understand what it actually is. Per Microsoft Learn's official documentation, Intelligent recap without saving transcripts lets Copilot-licensed organizers generate AI meeting summaries without saving a recording or persistent transcript. The AI summary is stored as a read-only Word document in the organizer's OneDrive.

The rollout timeline from Microsoft Message Center ID MC1275312 (updated July 27, 2026): general availability began rolling out mid-October 2026 and completes by late October 2026, for M365 Copilot (Premium) license holders.

Three things to understand:

1. This is not a bot-free desktop app — it's a cloud Copilot feature inside Teams.
2. Apps4Pro's analysis notes that under stricter compliance frameworks, the distinction between "no transcript saved" and "no AI processing" matters: AI is still processing live meeting context in real time in Microsoft's cloud.
3. It only works for Teams. For Zoom and Google Meet calls, you still need a desktop app or a bot.

For a deeper compliance analysis of this specific rollout, see our prior piece on Microsoft Teams Copilot's recap-without-transcript compliance reality.

Zoom AI Companion: the native option (and its limits)

Zoom's answer is AI Companion, which is also not a bot. Per Fazit's August 2026 breakdown, AI Companion is not a bot that joins your call — it is a server-side productivity layer that processes the meeting's audio, video, transcript, and shared-screen content inside Zoom's cloud under the host's account.

The gotcha for buyers who want local control: Zoom's own community confirms that AI Companion's transcription only works with cloud recordings — local recordings stored on your computer get no AI summary, because Zoom cannot process files it doesn't have. If you want to keep a Zoom meeting entirely local AND get AI notes, you need a desktop app running alongside Zoom.

Fireflies, Fathom, tl;dv, Otter: the "we added bot-free in 2026" tier

Every major cloud note-taker shipped a bot-free desktop mode this year, largely in response to Granola. Dymesty's 2026 comparison of Zoom notetakers documents the shift: Fathom, Fireflies, Fellow, MeetGeek, and tl;dv now offer bot-free modes alongside their traditional bot capture.

The trade-off is consistent across all of them. Fireflies' 2026 desktop "Take Notes" mode provides live transcripts and AI notes without a bot, but speaker labels become generic and separate audio/video files are not saved. Dymesty's analysis is explicit: "The Fireflies bot restores named speakers and saved media." You pick: clean roster, or named speakers plus a replayable recording.

Fathom changed substantially in 2026 by introducing a bot-free desktop experience that captures Zoom locally without connecting a Zoom bot — while still supporting the bot route when you want a replayable recording. YAPS' 2026 Fireflies alternative roundup confirms Fathom's bot-free capture is still in beta, and that Otter's desktop bot-free route means "the decision is no longer simply Fireflies has a bot and Otter does not" — buyers now have to check capture mode, OS, speaker attribution, and sharing behavior on each tool.

The comparison table: bot-free desktop options for Zoom and Teams in 2026

Where Basil AI fits in this landscape

Every tool above solves the "no bot in the attendee list" problem. What they don't all solve is the deeper question: once the bot is gone, where does the audio go?

Granola is explicit that audio is streamed to transcription providers, transcribed, then deleted — the transcript ends up in Granola's US AWS cloud. Fireflies, tl;dv, and Fathom all run transcription in their own clouds. That's a meaningful improvement over the old bot-everywhere model, but it still means your meeting content leaves your machine and lands on someone else's infrastructure before it comes back as a summary.

Basil AI is built differently. The audio is captured by your Mac's or iPhone's microphone, transcribed by a model running on the device's Neural Engine, and summarized locally. Nothing streams to Azure, AWS, or any vendor cloud. There is no recap document in someone else's OneDrive, because there is no "someone else." If you want AI meeting notes from your Zoom and Teams calls and you want the audio and transcript to never leave your device, that is the specific gap Basil fills.

The trade-off is honest: Basil is Apple-ecosystem only (Mac + iPhone + iPad), and it does not have Granola's calendar-triggered auto-launch or Fireflies' CRM integrations. For buyers with Windows fleets, Granola is the right pick. For privacy-constrained Apple-ecosystem workflows — legal, healthcare, M&A, executive 1:1s — Basil is the stricter architecture.

For a deeper look at how Basil's local transcription differs from Granola's model, see our comparison of Granola alternatives and bot-free vs on-device privacy on Mac.

The consent question nobody is answering well

Bot-free capture is a feature for the person running it and a consent gap for everyone else on the call. Because no bot appears in the participant list, remote attendees see no visible sign that a note-taker is running. This isn't just an etiquette problem — in two-party-consent jurisdictions it can be a legal one.

Granola's own 2026 how-to guide is clear about this: it tells users to "let the other participants know you are using Granola to transcribe the call" before focusing on the conversation. That advice applies to every bot-free app in this guide, including Basil. The absence of a visible bot does not relieve you of disclosure obligations.

For regulated-industry buyers, the broader compliance picture matters too. See our analysis of what "compliant AI meeting notes" actually means and AI meeting data retention and training policies compared.

How Basil AI solves this: on-device-vs-cloud, specifically for Zoom and Teams

Basil AI runs on your Mac or iPhone and captures the audio of your Zoom, Teams, Meet, Webex, or in-person meeting the same way every other bot-free app does — via system audio and microphone. There is no bot, no participant in the roster, no calendar integration pinging your meetings to a vendor server.

The difference is what happens next. On every other tool in the comparison table, that captured audio stream is encrypted and shipped to the vendor's cloud for transcription. On Basil, the audio is handed to Apple's on-device Speech Recognition and Neural Engine via the Apple Developer frameworks for speech and ML. The transcript is generated locally. Summaries, action items, and speaker diarization run on-device. The resulting notes write to Apple Notes via Apple's privacy-architected iCloud — your iCloud, not ours.

Because the audio never leaves the device, there is no AWS region question, no "audio deleted after transcription" promise to verify, no vendor retention toggle to configure. The GDPR Article 5 data minimization principle is satisfied structurally rather than contractually: Basil can't retain or process what it never received.

This is also materially different from Microsoft's "Recap without saving transcript." Microsoft deletes the transcript but keeps the AI processing in Azure and stores the Recap document in the organizer's OneDrive for 120 days by default. Basil's architecture doesn't produce a document in anyone else's cloud because the processing was never in anyone else's cloud.

Decision guide: which bot-free desktop app should you pick?

Based on the trade-offs above:

• Pick Granola if you need Windows + Mac coverage, want calendar-triggered auto-launch and cross-platform mobile apps, and are comfortable with transcripts in US AWS cloud.
• Pick Fireflies' Take Notes mode if your team already uses Fireflies for CRM automation and you want to toggle between bot and bot-free per meeting.
• Pick tl;dv or Fathom if a video recording of the call is important to you and you accept cloud storage.
• Pick Microsoft's Recap-without-transcript only if you are a pure Teams shop with M365 Copilot Premium and your governance team has signed off on tenant AI toggles.
• Pick Basil AI if you are Apple-ecosystem, if attorney-client, PHI-adjacent, M&A, or board-level conversations are in scope, and if "audio never leaves the device" is a hard requirement — not a preference.

FAQ: the questions buyers keep searching

See the FAQ block above for direct answers to the most common 2026 questions on bot-free desktop capture, Granola on Windows, Fireflies alternatives, and whether Microsoft's intelligent recap counts as bot-free. The short version: "bot-free" is now table stakes, and the real differentiator is whether the audio stays on your device.