why we built capturall on-device first

When we started building Capturall, the standard architecture for meeting transcription was already settled: a bot joins your call, streams the audio to a cloud service — usually in the US — and you get notes back. It works. It is also the reason many of the people we talked to, especially in Europe, were not using any of these tools at all.

Consultants told us their NDAs made third-party cloud recording a non-starter. Sales people told us prospects went quiet the moment a bot appeared in the participant list. Product managers told us research participants asked, reasonably, where the recording of their voice would end up. The common thread was not a missing feature. It was the architecture itself.

the default should be that audio goes nowhere

Modern Apple silicon changed what is possible here. An iPhone or a Mac can run serious speech-to-text models locally, fast enough to be useful for real meetings. So we made that the default: Capturall records on your own device, and transcription runs on that same device. The audio never leaves it. There is no upload step to secure, no retention policy to squint at, no data processing agreement covering a server you have never seen — because by default there is no server involved at all.

This is not a privacy setting buried in a menu. It is the path the product takes unless you deliberately choose otherwise.

no bot in the call

On-device capture has a second effect that we underestimated at first: there is no bot. Capturall never joins a meeting as a participant. Your calls look like calls. This matters commercially — a recorder bot changes how candid people are — and it matters practically, because plenty of organizations block recorder bots outright.

Recording without a visible bot puts more responsibility on the person recording, not less. That is why Capturall has an explicit consent step before any capture. We think the honest version of this product asks you to tell people they are being recorded, every time.

being honest about the cloud

On-device first does not mean cloud never. Two cases need a server: using Capturall in the browser, where local models are not practical, and high-accuracy mode, where a larger model catches things the on-device model misses. For those cases, processing happens on EU servers only — we use Mistral's Voxtral models with a zero-data-retention posture — and nowhere else.

Rather than blur the line between the two paths, we label it. Every transcript in Capturall carries a badge: processed on-device, or processed in EU cloud. If a tool's privacy story only lives on its marketing site, it is not much of a story. Ours is printed on every transcript, per conversation, where you can check it.

insights you can trace back to the tape

A transcript by itself is homework. What people actually need from a conversation is smaller and sharper:

• action points, with an owner and a deadline
• decisions that were actually made
• customer needs and objections, in the customer's own words
• a summary that does not flatten the disagreements out of the meeting

Capturall extracts these, and — this is the part we care about most — every insight carries a citation to the exact moment in the transcript it came from. If an extracted objection looks off, you click through and read what was really said. Language models are useful and fallible; citations are how we keep them accountable.

dutch as a first-class language

We are building from the Netherlands, and Dutch is where many transcription tools quietly fall apart. Strong nl-NL and nl-BE support is a core requirement for us, not a locale we added at the end. If your meetings switch between Dutch and English mid-sentence — as real meetings here do — that is exactly the case we test for.

where we are now

Capturall is in early access: apps for iPhone and Mac, plus the browser. We are onboarding in small groups because feedback from ten engaged users beats analytics from a thousand silent ones. We will not show you invented testimonials or a wall of customer logos we do not have. What we can show you is the architecture, the badge on every transcript, and the product itself.

If meeting notes without a bot in the call — and without your audio leaving your device — sounds like the tool you have been waiting for, request early access and tell us how you work. It genuinely shapes what we build next.