Every AI notetaker comparison asks the wrong question.

Most 2026 notetaker roundups argue about summary quality, integrations, and price. Those converge every quarter. The one axis that survives a security review is where your call audio goes — and on that axis, five of the six leading tools give the same answer.

The axis that actually separates them

By 2026 the summaries are largely a solved problem. Every serious tool produces a clean recap, action items, and a follow-up draft. If the output is comparable, the differentiator is the pipeline behind it, and it reduces to two questions:

• Does raw audio leave the device it was captured on?
• Is that audio ever written to a disk you do not own?

Answer those and the marketing falls away. Here is the whole field on that one axis:

                       AUDIO LEAVES      WRITTEN TO A DISK     A BOT JOINS
                       YOUR DEVICE?      YOU DON'T OWN?        THE CALL?
Fathom                 yes               yes                   yes
Otter                  yes               yes                   yes
Fireflies              yes               yes                   yes
tl;dv                  yes               yes                   yes
Granola                yes (to ASR)      transient             no
Fazit                  no                no                    no

Four tools are cloud recorders with a bot. One — Granola — is genuinely better and is the only real comparison. Fazit is the only row where every answer is no. The rest of this post is why each cell reads the way it does.

Fathom

Where it beats Fazit. A genuinely generous free tier, a polished experience across Zoom, Meet, and Teams, and enterprise trappings — SOC 2 Type II, HIPAA, SSO. It is cross-platform; Fazit is Apple Silicon only.

The fatal weakness. Fathom attends as a visible bot named “Fathom Notetaker,” and the recording lands in its cloud, stored in AWS data centres in the US and Canada. De-identified meeting data is used by default to improve Fathom’s own in-house models unless you go into settings and opt out. The recording is the asset, and keeping it out of training is your job, not the default.

Where Fazit wins. There is no bot in the room, no server-side recording, and no opt-out to find. Nothing was stored, so there is no corpus to train on, breach, subpoena, or hand to whoever owns the company next.

Otter.ai

Where it beats Fazit. Fast real-time transcription, a long track record, aggressive pricing, and browser-based ubiquity.

The fatal weakness. Otter is the defendant in a federal class action (Brewer v. Otter.ai) alleging it recorded conversations without all-party consent and trained its speech models on user recordings; its policy describes training on “de-identified” recordings, and by the complaint’s account it seeks consent only from the meeting host. A motion to dismiss was argued in May 2026 and is under submission; analysts have floated exposure in the $50–150M range. This is simply the litigated version of the risk the whole cloud category carries.

Where Fazit wins. You cannot train a model on a recording that was never created. Fazit’s audio lives in a RAM ring buffer and is consumed by an on-device model; there is no recording to subpoena, leak, or feed to a training set, and a one-line consent disclosure ships with the app.

Fireflies.ai

Where it beats Fazit. A deep integration ecosystem — CRM sync, Slack, unlimited-transcription tiers — that makes it sticky for sales orgs.

The fatal weakness. Fireflies faces biometric-privacy litigation in Illinois (including Cruz v. Fireflies.AI Corp., filed December 2025) alleging voiceprint collection without consent, alongside reports of unauthenticated access to meeting data. Its recording bot is now flagged by Google Meet as a third-party security risk and can be blocked from joining outright. A centralized store of voiceprints is precisely the asset that BIPA statutes were written to govern.

Where Fazit wins. No voiceprint is extracted, no bot needs to be admitted (or gets flagged), and there is no central store of meeting data to be accessed with or without authentication — because there is no central store.

tl;dv

Where it beats Fazit. Strong recording and playback UX, coaching and conversation analytics, and EU-headquartered positioning that reads well to European buyers.

The fatal weakness. It is still a visible-bot cloud recorder. tl;dv is the strongest of the cloud tools on paper here: German company, EU data centres, SOC 2, and a stated policy of not training on customer data. Grant all of it. The video file still exists on their infrastructure, and free-tier recordings auto-delete after three months, which is a retention limit rather than a privacy guarantee. Cloud recording under an EU flag is still cloud recording.

Where Fazit wins. For a European regulated professional, a pipeline that transcribes in volatile memory and never writes an audio file to a sound carrier is a materially stronger position than a promise of compliant storage. The distinction matters under German law (§201 StGB) precisely because “never created” and “created then deleted” are different legal facts. That argument is laid out in Why “Never Records” Is Not Marketing — It Is an Invariant.

Granola — the only serious comparison

Granola is the tool worth taking seriously, because it shares Fazit’s instinct: no bot in the room, capture the audio your own device already hears, and don’t retain it. It is SOC 2 Type II, the notes are excellent, and on privacy it is genuinely better than every bot-based tool above. If Fazit did not exist, Granola would be the right answer.

The fatal weakness. Granola is bot-free, but it is not on-device transcription. To turn audio into text and text into a summary, it streams your call audio to third-party services and deletes the raw audio afterwards. its security FAQ names OpenAI and Anthropic, and AssemblyAI lists Granola as a customer on its own site. The full subprocessor list stays behind a Trust Center request. “Bot-free” is not the same claim as “audio never leaves the machine.” Your conversation still transits the public internet to four subprocessors. Deletion is a promise about what happens after the audio arrives on someone else’s server — not a guarantee that it never arrived.

Where Fazit wins decisively. Fazit’s ASR runs on the device. Audio is captured into a fixed-size RAM ring buffer, a local model consumes a snapshot, and the note is written to your vault. The audio never leaves the machine — so there is no subprocessor to list, no transfer mechanism to disclose, and no “we delete it after transcription” step to take on trust, because there is no transcription server to delete it from.

What the whole table reduces to

Two distinctions do all the work. Cloud recorders live on created then deleted. Granola improves that to sent then deleted. Fazit is never created, never sent. Everything else — pricing, integrations, summary polish — is a tie-breaker that only matters after you have decided how much of your client’s voice you are willing to put on someone else’s disk.

That is why every note Fazit writes carries audio_retained: false in its frontmatter. It is not a compliance checkbox; it is a description of the execution path. The line is true because there is no code path that writes audio to disk, and source access for independent security review is available on request.

What Fazit gives up (so you don’t find out later)

The on-device design is a trade, not a free lunch, and it is worth naming the costs before you choose:

• Apple Silicon only. The pipeline runs on the Mac’s neural hardware. If your team is on Windows, Fazit is not for you today.
• No bot means no proxy attendance. A cloud bot can join and record a call you are not on. Fazit only captures what your own machine hears. If your workflow depends on recording meetings you don’t attend, that is exactly what Fazit refuses to do.
• A smaller integration surface than Fireflies or Otter. Fazit writes Markdown into your vault; it is not a CRM autopilot.

Those are deliberate. The moment a tool records calls you didn’t attend, or buffers to a server for reliability, it inherits the exact risk profile this comparison is about. Fazit declines the features that would require it.

FAQ

How is Fazit different from Otter or Fireflies?

Otter and Fireflies capture with a bot and process in the cloud, so a recording and a transcript exist on vendor infrastructure. Fazit captures without a bot, transcribes on the Mac, generates notes with a local model, and writes a Markdown file into your own vault. No recording is created at any point.

Is Granola a good alternative to Fazit?

Granola is the closest of the cloud tools and genuinely good: no bot, excellent notes, SOC 2 Type II. The difference is that its audio still reaches its systems and third-party transcription services before being deleted, and the transcript is retained on its servers. If you want the notes in files you own with no vendor copy, that is the gap.

Does Fazit work with Zoom, Teams and Google Meet?

Yes, and with anything else that plays call audio on the Mac, including Webex, Slack, Discord, FaceTime and calls in a browser tab. Capture happens at the operating system rather than through a per-app integration, so there is no picker and no per-platform support list to check.

What does Fazit send to the internet?

During a call, nothing. The only endpoint in use is localhost. Models are downloaded once from public CDNs on first run, and account, licensing and payment use our servers, which is a separate lane from call content and is described on the security page.