Key findings
- Bots exist for accuracy, not convenience. Joining the call gives a tool separated per-speaker streams, which is why bot-based speaker labels beat room-mic ones.
- Three bot-free approaches: record the room with a phone, capture system audio on desktop, or transcribe the platform's own recording afterwards.
- Bot-free does not mean private. It describes capture, not destination. A bot-free app can still upload to four vendors.
- Only two options here keep audio entirely on your machine: on-device phone transcription, and local Whisper on a Mac.
- Removing the bot removes the notice. The bot was telling the room it was being recorded. Now that is your job.
There is a specific moment that made us build a recorder that does not join calls. You are in a first meeting with a client, everyone is being slightly formal, and a participant appears in the list called "Notetaker." Somebody asks what it is. The next ten minutes are about your software instead of their problem.
Bots are not bad. For an internal standup or a sales call where everyone already expects recording, they are the best option available and the accuracy is genuinely better. But there are plenty of conversations where you want the transcript and you do not want a fourth entity in the room.
Why bots exist at all
Worth understanding before you give one up, because you are trading something real.
When a note taker joins a call as a participant, it receives the same audio streams every other attendee gets: one separate stream per speaker. That makes speaker attribution close to trivial. The tool knows Dana said the thing because Dana's stream carried it.
A room microphone gets one mixed signal with everybody in it, plus HVAC, plus the person tapping a pen. Separating speakers from that is inference rather than fact, and it is meaningfully harder. Any bot-free tool that records a room is doing a harder job, and its speaker labels will be less reliable. Mine included.
So the question is not which approach is better. It is which failure you would rather have.
The six, compared
| Tool | How it captures | Audio leaves device | Works for remote calls | Cost |
|---|---|---|---|---|
| Voxxli | iPhone mic, records the room | No | No, in-person only | Free, paid tier for AI features |
| Granola | System audio on Mac and Windows | Yes, then discarded | Yes | Paid, free trial |
| Apple Voice Memos | Device mic | No | Not really | Free, built in |
| MacWhisper and similar | Local file or system audio | No | Yes, with capture set up | One-time purchase |
| Otter.ai mobile | Phone mic, without the Assistant bot | Yes, and retained | No, in-person only | Free tier, paid plans |
| Native platform recording | Zoom, Teams or Meet built-in | Yes, to that platform | Yes | Included in most plans |
Approach one: record the room
Put a phone on the table. This is the oldest method and it is still the only one that works when the meeting is not on a computer at all.
Voxxli is ours, so discount accordingly. It records on an iPhone and transcribes using Apple's on-device speech framework, which means the audio file never leaves the phone and transcription works with the network off. There is no account. If you want a summary, action items, flashcards or chat, the transcript text is sent to a third-party AI provider, which is named in the privacy policy. The audio never is. It cannot join a Zoom call, which is the whole point but also a real limitation.
Apple Voice Memos is the honest free baseline and more people should consider it. Recent iOS versions transcribe recordings natively, on-device, at no cost. You get no summaries, no speaker labels and no search across recordings, but if all you need is "what did they actually say," it is already on your phone.
Otter's mobile app records in person without deploying the Otter Assistant bot, which surprises people who think of Otter as a bot product. Worth knowing it is an option. Worth also knowing that the audio still uploads to Otter's cloud and is retained there, which we covered in detail in this comparison of where transcription tools send your audio.
Approach two: capture system audio
The desktop answer. Your computer is already playing everyone's voices through its speakers, so a local app can listen to that output directly without ever joining the meeting.
Granola is the best-known version and is worth taking seriously. It captures audio locally on macOS and Windows, sends it to cloud transcription providers, and then discards the recording once the transcript exists. Transcripts persist, audio does not. Granola has said publicly that it tried on-device transcription and moved to the cloud for quality reasons, which we find both credible and clarifying about the tradeoff everyone in this category is making.
Local Whisper apps such as MacWhisper run OpenAI's Whisper model on your own machine. Nothing uploads at all. Accuracy is very good, you pay once rather than monthly, and the cost is setup effort plus your laptop fans. For anyone whose threat model is genuinely "no vendor should have this," this is the strongest option on the list and it is not close.
Approach three: use the platform's own recording
Zoom, Teams and Meet all record natively. No third party involved, the notice is built into the platform, and everyone already understands what the red dot means. Record there, then run the file through whatever transcription you like afterwards.
The tradeoff is that it is not automatic and the recording lives with your meeting platform, which may or may not be an improvement depending on who you are worried about.
No bot, no account, no upload
Voxxli records in-person meetings, lectures and interviews on your iPhone. Apple's on-device speech engine does the transcription, so the audio file never leaves the phone. Works in airplane mode.
The privacy trap
This is the part most articles on this keyword get wrong, so it is worth being blunt about.
"No bot" and "private" are unrelated claims. One describes how audio is captured. The other describes where it goes afterwards. A tool can absolutely avoid your participant list and still ship your recording to a transcription vendor, a model vendor and a storage bucket in three different jurisdictions.
Look at the table again. Four of the six options here send audio off your machine. Removing the bot did nothing about that.
If what you actually want is privacy rather than discretion, the test is simple and we keep repeating it because it works: turn off networking and try to record. If a transcript still appears, processing is genuinely local. If it spins, it is not, whatever the landing page says.
Nobody designed the bot to be a consent mechanism, but that is partly what it became. A visible participant called "Notetaker" told the room something was recording. Take it away and that notice has to come from you, out loud. Twelve US states require every participant to consent before you record, and being discreet is not a defence. See recording laws by state for which twelve.
Which one to pick
- Mostly in-person meetings, iPhone in your pocket: a room recorder with on-device transcription. Voxxli, or Voice Memos if you do not need summaries.
- Mostly remote calls, want it automatic: Granola. Accept that audio touches the cloud briefly and is then discarded.
- Nothing may leave your machine, ever: local Whisper on a Mac. More setup, no compromises.
- Occasional need, no budget: your platform's native recording plus any transcriber afterwards.
- You need reliable per-speaker attribution: honestly, use a bot. It is the right tool for that job and every option on this page is worse at it.
Questions people actually ask
How do I take meeting notes without a bot joining the call?
Three ways. Record the room with a phone app, which works on any platform and for in-person meetings. Capture system audio with a desktop app such as Granola, which hears the call without joining it. Or use the meeting platform's own recording and transcribe the file afterwards. None add a participant.
Why do AI note takers join meetings as a participant?
Accuracy. Joining gives the tool separated per-speaker audio streams, which makes speaker labelling far more reliable than inferring speakers from one mixed room recording. The cost is visibility, and often needing the host to admit the bot.
Is a bot-free note taker more private?
Not inherently. Bot-free describes capture, not destination. Several bot-free tools still upload your audio to multiple cloud vendors. The question that matters is whether audio leaves your device at all.
Can Voxxli join a Zoom or Teams call?
No. It records the room it is physically in through the iPhone microphone, which suits in-person meetings, lectures and interviews. For remote calls, a desktop tool that captures system audio is a better fit.
Do I still need consent if there is no visible bot?
Yes, and the absence of the bot makes it more important. Consent law turns on whether recording is happening, not on whether it is visible. The bot was providing incidental notice, so without it you need to say so yourself. Twelve US states require every participant to agree.