Bot-free AI notetakers — four different mechanisms behind one phrase

“Bot-free” is a claim about the participant list. It tells you nothing about where your audio went in the second after it was captured, which is what most people are actually asking.

Four common mechanisms are sold under the phrase. They have four different privacy consequences, four different sets of meetings they cannot reach at all, and four different answers to “is the recording still there tomorrow”. Here they are, with the vendor’s own words wherever a vendor has published them.

The four mechanisms

1. Desktop app, transcription on your machine

A system-audio tap plus your microphone, and a speech model on your own Neural Engine. Nothing about the meeting needs a network connection, so there is no vendor to trust with the audio and no per-minute meter.

This category is small, and Narada is not alone in it. MacWhisper says transcription “is all performed locally on your Mac or iOS device, and transcription data doesn’t leave the local device”, with cloud providers and AI prompts as opt-in extras. Superwhisper’s meeting mode “Records audio straight from meeting apps” and “none of it gets uploaded”. Both are cheaper than a meeting-notes subscription — MacWhisper Pro is €64 once, Superwhisper $8.49/month, $84.99/year or $249.99 once — and neither publishes calendar awareness or unattended start. Superwhisper’s local models are themselves a Pro feature, so its free tier runs on cloud ones. MacWhisper’s docs describe a "Meeting Detected" notification; you still click Record.

The cost: the model runs on your hardware, so it takes memory and time, and the ceiling is whatever fits on a laptop rather than whatever fits in a datacentre.

2. Desktop app, transcription in the cloud

Identical from the outside. No bot, a native app, a menu-bar icon. Then the audio leaves.

Granola is the cleanest example, because Granola documents it rather than blurring it:

“For transcription and AI summarization, we process this in the cloud to provide the best quality - we initially tried doing these locally on device, but the computation was too much and it slowed down your computer.”

And it is not a round trip through Granola’s servers. Their network troubleshooting page, written for firewall admins, names the endpoints your Mac connects to directly: api.deepgram.com, *.api.deepgram.com and streaming.assemblyai.com, over “direct outbound WSS over TCP 443”. In exchange Granola keeps no audio at all — “there’s no way to access audio from your meetings” — which is a genuine privacy property and also why you can never re-listen to a disputed sentence. Longer comparison here.

Jamie does the same thing in a different jurisdiction and says so: audio “is uploaded to a server location in Frankfurt, Germany 🇩🇪, and then processed on a serverless GPU environment from Modal”, after which “we delete all audio files permanently."

Bluedot markets a “Local Processing” option; read what it governs. Its help centre describes it as where “your recording will be stored while it is being processed”, while the privacy page lists Google (Alphabet Inc.) as a subprocessor receiving “video transcriptions” with “Place of processing: United States". Not offline operation, and Bluedot does not claim it is.

Fireflies' bot-free desktop mode is here too, with one property worth knowing first: “System audio mode does not save separate audio or video files.” Its documentation does not say where that mode transcribes, but Fireflies' own storage doc puts everything “in Fireflies' secure cloud infrastructure in the United States (AWS and GCP)". Otter has a botless desktop mode as well, described in its own words as recording “the audio from your device” without Notetaker joining — same cloud afterwards. Fathom’s bot-free capture is here too, footnoted on its own pricing page as “Beta feature for Mac”.

The cost: your meeting audio travels to at least one company you did not choose, and the archive lives in an account rather than a folder.

3. Browser extension

An extension reads the meeting from inside the tab. Nothing joins the call, nothing is installed into your audio graph, and in the strictest implementations no audio is handled at all — Tactiq’s page says it “reads what you see and hear as a participant” and that “no audio is ever captured or saved”.

Then the structural limit: an extension can only see a browser. Tactiq’s help centre states it without hedging:

“Tactiq works only when you join Zoom meetings in a web browser.”

The same article notes its macOS Zoom desktop app beta “is currently paused". Fireflies' extension route is narrower still — its own pricing matrix scopes bot-free extension capture to Google Meet. If your week includes the Zoom desktop client, a Teams client, a Slack huddle or a room with people in it, an extension covers a fraction and covers the rest not at all.

The cost: no audio file, a hard dependency on doing every call in Chrome, nothing for in-person.

4. The platform’s own notetaker

Zoom’s My Notes, Teams intelligent recap and Google Meet’s “Take notes for me” need no bot for a boring reason: they are the platform. The stream is already theirs.

They are good at their own platform and incapable of anything else. The catch that bites hardest is permission, since these features are gated above you. Microsoft, on who can read a Teams recap:

“Anyone who wasn’t invited or isn’t in your org won’t have access.”

Google, in its own launch post for Meet note-taking: “you can use the feature now for meetings that you host.” Teams transcription is “both a per-organizer and per-user policy setting”, so if the organiser’s admin turned it off, nothing you own can produce notes for that meeting. Zoom’s Meeting Summary is host-only — though its My Notes feature is the interesting exception, capturing “the system and microphone audio on the device” across “Zoom, Teams, Google Meet, and other third-party meeting platforms” unless the host toggles it off.

Check the language list too. Google’s note-taking supports eight — “English, French, German, Italian, Japanese, Korean, Portuguese, Spanish” — and “Multiple languages spoken in the same meeting aren’t currently supported”. Zoom lists 32 for AI summaries.

The cost: it stops at the edge of one platform, it needs your admin and your host to agree, and the notes land in the org’s cloud under the org’s retention policy.

The three questions, and how to make a vendor answer them

You can classify any product in five minutes without trusting a comparison page, this one included.

1. Where does transcription run — and can you point to the sentence? Search the docs for “on-device”, “local”, “subprocessor”, or a network allowlist page. Allowlist and troubleshooting pages are the most honest documents any SaaS publishes, because they are written for a firewall admin rather than a buyer. That is how you find out a client opens a socket to Deepgram.
2. Does an audio file survive, and who has it? “We delete the audio” and “the audio is in your folder” are opposite answers and both can be right — but only one lets you check a quote in six months.
3. What does the app need in order to see the meeting at all? A browser tab, a platform account, or your Mac’s audio devices. That decides whether it works for Thursday’s in-person meeting.

A useful fourth: does the vendor also sell a bot? Bluedot advertises capture “With or without bot”. Krisp says “Krisp also offers a Bot option for platforms like Zoom”. Fathom’s free plan offers a “Choice of bot-free (in beta) or bot capture”. None of that is dishonest, but it means bot-free is a setting there rather than the architecture — and settings get changed by defaults, updates and admins.

Where real products sit

Every cell comes from the vendor’s own documentation, checked 4 August 2026.

Where Narada sits, and what that costs you

Narada is in the first category and nowhere else. It records two streams — your microphone and your Mac’s system output — and transcribes both on the machine with Parakeet TDT v3 through Core ML. No bot, no account for the recorder, no audio upload in any mode. Each meeting becomes a folder holding an .m4a and a timestamped transcript — with a Markdown summary and a JSON analysis alongside them once an AI mode is on — unencrypted on purpose. The recorder tier is free; Pro AI is $6.99/month or $69.99/year with a 3-day trial; the on-device AI mode costs nothing.

Now the parts a comparison page usually leaves out.

• Transcription covers 25 European languages, and not equally. The engine’s published benchmark runs from about 4% word error rate on Italian to about 37% on Greek, with Slovenian, Latvian, Maltese and Lithuanian in the mid-twenties. Chinese, Japanese, Korean, Arabic, Hindi and Turkish are not supported at all.
• Speaker labels have no names. The call’s audio is split into Speaker 1, Speaker 2 and so on; the app will not guess who they are. Your own lines are attributed to you only because your microphone is a separate file, so a mic-only or in-person recording gets no speaker separation at all.
• In the cloud AI modes the transcript text leaves. Pro AI and bring-your-own-key send transcript text to OpenAI after you explicitly consent. Audio never leaves in any mode, and in Recorder-only or on-device mode nothing leaves for AI at all.
• On-device AI is Apple Silicon only. One-time 3 GB or 7 GB model download depending on your RAM, several gigabytes of memory while it runs, and lighter notes than the cloud route. On an Intel Mac the recorder and the transcription still work; the local model does not exist. Transcription has a one-time download of its own — about 600 MB for the speech model, on any Mac — after which it is offline.
• macOS 15.6 or newer, and Mac only. No Windows, no iPhone app, no live transcription during the meeting, no translation, no team or sharing features, no CRM integrations. The Claude Desktop connector is read-only and its search is keyword matching rather than semantic.
• Auto-record is opt-in. The shipping default is to ask before each meeting.
• macOS audio can fail mid-recording, and Narada says so. A device reshaped under a live capture, an overloaded audio daemon, a microphone that quietly stops delivering — that happens, on every tool that records a Mac. Narada checks each finished recording against itself and marks it Incomplete with the reason. It cannot recover audio that was never written. A tool that never shows you a warning is not necessarily doing better; it may just not be looking.

One thing several category-2 products do that Narada does not: tell the room for you. Granola can post an automated chat message when transcription starts and can put a watermark on your video; Bluedot can show a consent message. Narada shows a reminder on your own screen and nothing more.

The part no software does for you

None of these tools makes a recording lawful. Rules vary by country and by state, and in some places the standard is that everybody agreed, not that nobody objected. That is the specific irony of this category: the bot in the participant list was, accidentally, a disclosure mechanism, and removing it puts the disclosure back on you.

Four seconds at the start of the call settles it. Say you are recording.

Sources

All read on 4 August 2026.

• Granola: docs.granola.ai — security, privacy & data FAQs (no bot joins), granola.ai/security (cloud processing, named providers), docs.granola.ai — network troubleshooting (Deepgram and AssemblyAI endpoints), docs.granola.ai — transcription (no audio saved), docs.granola.ai — transparency solutions (chat message, watermark).
• Jamie: docs.meetjamie.ai — data (Frankfurt upload, Modal, audio deleted).
• Bluedot: bluedothq.com (“With or without bot”), bluedothq.com/privacy (subprocessors and processing locations), help.bluedothq.com — recording settings explained (what “Local Processing” governs).
• Krisp: krisp.ai — security for AI Meeting Assistant (virtual mic and speaker, on-device transcripts, Azure summaries, US-only storage), krisp.ai/ai-meeting-assistant (on-device English plus 15 server languages, optional Zoom bot), krisp.ai/pricing (storage caps).
• Fireflies: guide.fireflies.ai — recording without a bot on the desktop app (no separate audio or video files), guide.fireflies.ai — how Fireflies joins and records (“always visible in the participant list”), guide.fireflies.ai — data storage and transfer (US AWS and GCP).
• Otter: help.otter.ai — Otter Desktop App (botless recording of device audio, and the consent burden it puts on the user), help.otter.ai — stop Otter Notetaker from automatically joining (the bot mode and its calendar auto-join).
• Fathom: fathom.ai/pricing (“Beta feature for Mac”, capture choice), help.fathom.video — is Fathom secure (US storage).
• Tactiq: tactiq.io — AI note taker that doesn't join meetings (browser capture, no audio saved), help.tactiq.io — how to use Tactiq with Zoom (browser only, macOS beta paused).
• Zoom: support.zoom.com KB0083849 (My Notes device audio, host toggle), KB0058013 (host-only Meeting Summary), KB0078144 (32 languages), KB0080354 (AI Companion joining third-party meetings).
• Microsoft: support.microsoft.com — Recap in Microsoft Teams (guest exclusion), learn.microsoft.com — meeting transcription (per-organizer policy), learn.microsoft.com — intelligent recap privacy (where notes are stored).
• Google: blog.google — Take notes for me (“meetings that you host”), support.google.com/meet/answer/14925782 (eight languages).
• MacWhisper: macwhisper.com (€64 one-time), docs.macwhisper.com — keeping transcriptions private (local by default), docs.macwhisper.com — record meetings (Meeting Detected notification).
• Superwhisper: superwhisper.com/meeting-transcription (“none of it gets uploaded”), superwhisper.com/docs — Pro (pricing, local models are Pro).
• Narada: prices and the macOS 15.6 minimum from its App Store listing, checked 4 August 2026. Word error rates per language from the FLEURS benchmark table published in the documentation of FluidAudio 0.13.4, the transcription package Narada ships. https://raw.githubusercontent.com/FluidInference/FluidAudio/main/Documentation/Benchmarks.md

Questions people ask about this

What does bot-free actually mean?

That no extra participant appears in the meeting. It is a claim about the attendee list, not about where the recording is processed. Several bot-free products upload your audio to a transcription vendor within seconds of capturing it, which is a perfectly reasonable design but not the same as keeping the meeting on your machine.

Is a bot-free notetaker more private than a bot?

Not automatically. A bot-free desktop app that streams audio to a cloud transcriber sends the same audio to the same kind of server as a bot does — it just does not announce itself in the call. The privacy difference comes from where transcription happens, not from the absence of a participant.

Which bot-free tools transcribe on the Mac itself?

On the evidence of vendor documentation: Narada (Parakeet TDT v3 through Core ML), MacWhisper and Superwhisper for their local models — Superwhisper’s local models are a Pro feature — and Krisp for English only, since Krisp’s own security page says summaries are generated via Microsoft Azure and all stored data lives in the United States. Granola, Jamie, Bluedot, Otter and Fathom transcribe in the cloud. Fireflies does not say where its bot-free desktop mode transcribes; its own data doc places all processing in US AWS and GCP.

Does bot-free mean I do not have to tell anyone?

No. Recording rules vary by country and by state, and in some places every participant has to agree. No app makes a recording lawful, and most bot-free tools remove the one signal that used to announce it. Say it out loud at the start of the call.

Why do so many bot-free vendors also sell a bot?

Because a bot reaches meetings a local capture cannot — someone else’s call you are not attending, or a platform whose desktop client the tool cannot read. Bluedot advertises capture “With or without bot”, and Krisp says “Krisp also offers a Bot option for platforms like Zoom”. For those products, bot-free is a mode you select rather than the architecture.

Try it on your next call

Narada records your Mac's audio and your microphone, transcribes on-device, and hands you the decisions and follow-ups. Free to use, no account, no bot in the meeting.

Download for MacFree · macOS 15.6+ Try the live demo

Keep reading

Answer How do I record a meeting without a bot joining the call? Bot-free meeting recording means recording your own computer's audio instead of inviting an assistant into the… Alternatives Granola alternatives for Mac, sorted by the reason you're leaving Why people leave Granola — no audio file, forced Google or Microsoft sign-in, $14 a seat monthly — and which M… Hub Narada compared with other meeting notetakers Head-to-head comparisons of Narada against Granola, Otter, Fireflies, Fathom and tl;dv, with pricing and priva…