Meeting transcription without a bot (or the cloud)

A recording bot is one way to get a meeting transcript. It's also the way that puts a visible third participant in your call and a copy of the audio on someone else's server. If either of those bothers you — clients ask what the bot is, some organizations block bots outright, and confidential calls don't belong on shared infrastructure — you have options. This guide sorts the field by the two questions that actually matter: how does it capture the audio, and where does processing happen?

Disclosure first: we make QuickSpeak, the last tool in the table. We've kept every claim about the other eight to their own pricing and security pages, linked in the sources. Where a vendor's claim is a marketing claim (language counts, for instance), we say so.

The two questions to ask any notetaker

How does it capture? Three mechanisms exist. A bot joins the call as a participant and records from inside the meeting. An extension or desktop app captures the audio your own machine is already playing — the meeting tab's audio, or system audio. A file uploader transcribes recordings you already have.

Where does processing happen? Either the audio is uploaded and transcribed on company servers, or the model runs on your machine and only the result — a transcript — exists locally.

The two questions are independent, and vendors blur them constantly. "No bot" does not mean "private": a tool can capture audio on your machine and still stream it to a cloud model in real time. Granola is the clearest example — it doesn't join your call and doesn't keep the audio, but its own security page says transcription runs through third-party providers and notes are written by OpenAI or Anthropic models. Your audio takes a round trip through someone's cloud; it just doesn't stay there.

The privacy ladder

Rung 1 — bot in, audio up. Otter, Fireflies, and tl;dv's default mode. Maximum convenience (capture happens to meetings whether or not you remember) and maximum exposure: a visible participant, an uploaded recording, a cloud retention policy, and a training policy you should actually read.

Rung 2 — no bot, but cloud processing. Granola, Tactiq, Fathom's bot-free mode, and Krisp's multilingual path. The meeting looks normal — no robot in the participant list — but the audio still reaches servers for transcription or summarization. A middle rung, not a top one.

Rung 3 — on-device. MacWhisper and QuickSpeak run recognition locally; Krisp does on-device transcription for English only, selling private on-device transcription as an Enterprise feature. Nothing is uploaded because there is nothing to upload to. The trade: no shared team library, few or no mobile apps, and your hardware does the work.

The field, compared

Prices as published on each vendor's pricing page, October 2026. "Paid from" is the cheapest advertised paid tier, usually billed annually.

The tools, honestly

Otter.ai

The category default since 2018, with the most mature team features: shared workspaces, CRM sync, SSO on Enterprise. The free tier is the tightest of the big three — 300 minutes a month, 30 minutes per conversation — and Otter trains its models on de-identified customer meetings by default. A class action filed in August 2025 alleges its bot recorded meetings without all participants' consent; Otter disputes this and the case is ongoing.

Fireflies.ai

Built for sales teams: conversation intelligence, CRM logging, and a bot that joins every calendar meeting unless you stop it. The better-than-average cloud privacy posture — contractually no AI training, 0-day retention marketing — is undercut mostly by the auto-join default itself. Free plan's 400-minute storage cap makes it a demo of the paid product.

tl;dv

Generous transcription on the free tier and a slick clip-and-share workflow, but the bot is default-on, free meetings age out after about three months, and AI notes are metered (ten meetings). The desktop recorder is a decent bot-free escape hatch.

Fathom

The friendliest free tier in the cloud group — unlimited recordings and transcriptions, with AI summaries capped at five a month. Bot-free desktop capture exists (not yet for Teams). Everything lands in a US cloud, calendar connection is mandatory, and in-house model training on de-identified data is on by default with an opt-out.

Granola

The best-known no-bot notetaker, and the clearest proof that "no bot" and "no cloud" are different claims: it captures on your machine, doesn't store audio, and still sends speech to cloud transcription providers and notes to OpenAI/Anthropic models. Loved for its notes-from-fragments editing feel; the free tier's 30-day history is the catch. If you want Granola's no-bot behavior with nothing leaving the machine, that's the rung-3 group below.

Krisp

Famous for noise cancellation, now with meeting notes. On-device transcription exists but only for English; the other 15 languages run server-side, and "private on-device transcription & recordings" is listed as an Enterprise feature. No permanent free tier — 7 days, then paid.

Tactiq

The closest mechanical cousin to QuickSpeak: a Chrome extension that transcribes the meeting tab live, no bot. The difference is what happens next — transcripts sync to your Tactiq account in the cloud and AI features run on OpenAI's API. Ten free transcripts a month. A good fit if you want live captions with a cloud account behind them.

MacWhisper

Genuinely local and genuinely good: Whisper and Parakeet models run on the Mac, no account, works offline, €64 once. Best in class for transcribing files — interviews, podcasts, recordings you already have. It records meetings too; where QuickSpeak differs is running in the browser on any OS, with speaker profiles that persist across meetings and an unmetered free recording engine.

QuickSpeak

Ours, so weigh this paragraph accordingly. A Chrome extension (plus a web app) that records the meeting tab, desktop audio, or mic and transcribes in-browser with NVIDIA's Parakeet model — no account, no bot, no upload, works offline. Speaker profiles persist across meetings on your device. The free tier is the whole recording engine, unmetered; $12/month or $99/year adds live captions, cross-meeting search, on-device AI briefs, and transcript chat. What it doesn't have: mobile apps, team workspaces, integrations, or more than 36 languages.

Five questions to ask before you trust any of them

1. Does a bot join the call — and is that default on or off? Auto-join means your tool decides when other people get recorded.
2. Where is the audio stored, in which region, for how long? "Encrypted" describes data at rest, not who can query it.
3. Is my content used for training, and can I actually turn that off? Read the privacy page, not the pricing page. Otter trains by default; Fireflies contractually doesn't; the rest vary.
4. Does it work with the network disconnected? The only honest proof that audio isn't being uploaded. Try it.
5. What happens when I delete? If transcripts and voice models live on a server, deletion is a request. If they live on your disk, it's a file operation.