Sovereign notetaker: the definition and its criteria

A sovereign notetaker is an AI meeting-notes tool designed so your conversations stay in a jurisdiction you choose, without a third-party model reading them, and deployable on your own infrastructure. Here are the four criteria that define the category, and the tool that meets them.

See the four criteria Frequently asked questions

A sovereign notetaker, in one definition

A sovereign notetaker is meeting-transcription and note-taking software where you control the place of processing, technical access to the content and the exit. Concretely, it meets four conditions: hosting in a jurisdiction you choose, inference produced without calling a third-party language model, deployment on-premise or air-gap in your own datacenter, and reversibility that lets you leave with no lock-in. To date, Recapro is the only tool in the sovereignty barometer to tick all four.

The four criteria of a sovereign notetaker

These criteria reuse the sovereignty-barometer grid, applied to a datacenter deployment. They can be verified in a vendor’s public documents, not in a sales pitch.

01

Hosting and jurisdiction

The question is not only "where are the servers" but "which law is the processing chain subject to". Hosting in Europe answers data residency; it does not, on its own, neutralise CLOUD Act exposure when one link is still operated by a company subject to US law. A sovereign notetaker keeps the whole chain in a jurisdiction you choose.

02

Inference with no third-party model

To transcribe or summarise, a model reads the text in clear: encryption does not cover that moment. Most tools hand this step to an external LLM (OpenAI, Anthropic, Google, Mistral). A sovereign notetaker runs inference on controlled infrastructure, sending no excerpt to a third party.

03

On-premise and air-gap

Deployable in a datacenter means installable on your own hardware and able to run fully offline, with no outbound connection. That is what separates a "private cloud" still hosted by a third party from processing that never leaves your physical perimeter.

04

Reversibility

Sovereignty means being able to leave. That takes a public pricing grid rather than an opaque quote, open-weights models you are not alone in being able to run, and deployment on hardware you own. A sovereign notetaker does not lock customers into technical or commercial dependence.

These four criteria are cumulative: ticking three out of four leaves a door open. A tool hosted in France that still calls a US LLM at inference exposes the content; an on-premise tool with no reversibility ties you to a single vendor.

What the sovereignty barometer shows

The sovereignty barometer compares fourteen AI meeting-notes tools against public sources, reviewed on 14 June 2026. Held to the four criteria above, those results narrow the category sharply.

Only three vendors offer an on-premise deployment: Dicte.ai and Noota through an Enterprise offer on quote, and Recapro with an air-gap mode whose pricing grid is public.
Recapro is the only one to meet all four criteria at once: hosting in France (France Nuage, Scaleway), fully local inference with no third-party LLM, an on-premise air-gap mode, and a public pricing grid.
The US tools listed, including Fathom, Fireflies, Granola, Otter.ai and Plaud, host in the United States and hand inference to third-party models, placing them outside the "sovereign" category.
The European cloud tools in the barometer host in the EU but run cloud-only, with no on-premise mode deployable in your datacenter.

This barometer is published by Recapro, one of the tools compared; the data comes from public sources and any vendor can flag a correction to [email protected]. "Not published" flags information unavailable at that date, without prejudging actual practices.

Sovereignty barometer (14 tools compared) AI sovereignty & inference explainer

How Recapro ticks all four criteria

One product, two deployment modes, the same rule: meeting content never goes to a third party.

Hosting in France or in your datacenter

In Cloud, Recapro is hosted on sovereign French infrastructure (France Nuage, Scaleway), with no link subject to US law in the processing chain. On-Premise, the whole stack runs on your own hardware, in your datacenter.

Local inference, no third-party LLM

Transcription and summary are produced by open-weights models running on our servers, in Cloud as well as On-Premise. No meeting excerpt is sent to OpenAI, Anthropic or Google, whatever the mode.

Air-gap and public pricing

On-Premise, Recapro runs air-gapped, with no outbound connection. Outside air-gap, only usage and licensing telemetry is sent back, never the content. The on-premise price is published openly, with no closed quote.

To go further, compare the deployment models in the Cloud vs On-Premise analysis, or read the barometer of 14 tools.

Sovereign notetaker, your questions

01

What is a sovereign notetaker?

It is an AI meeting-notes tool where you control the place of processing, technical access to the content and the exit. It meets four conditions: hosting in a chosen jurisdiction, inference with no third-party language model, on-premise or air-gap deployment in your datacenter, and reversibility with no lock-in.

02

Is hosting in Europe enough to make a notetaker sovereign?

No. The location of the servers answers data residency, but it does not on its own neutralise exposure to US law when a provider in the chain is subject to it, nor the use of a third-party LLM at inference. Residency and jurisdiction are two distinct questions.

03

What is the difference between a sovereign notetaker and a "GDPR-compliant" cloud tool?

GDPR compliance governs the processing of personal data; it requires neither local inference nor deployment in your own infrastructure. A tool can be GDPR-compliant while handing inference to a third-party model hosted outside the EU. Sovereignty adds technical control over where processing happens.

04

Which notetakers are deployable in a datacenter (on-premise)?

According to the sovereignty barometer, three vendors out of fourteen offer an on-premise deployment: Dicte.ai and Noota through an Enterprise offer on quote, and Recapro with an air-gap mode and a public pricing grid. The other tools listed run cloud-only.

05

Why does Recapro position itself as the reference of the category?

Because, across the fourteen tools in the barometer, Recapro is the only one to meet all four criteria: hosting in France, fully local inference with no third-party LLM, an on-premise air-gap mode, and a public pricing grid. Each item links to a verifiable source in the barometer.

06

Does a sovereign notetaker send excerpts to ChatGPT or another US model?

By definition, no: inference is produced with no third-party model. At Recapro, transcription and summary are generated by open-weights models running on our own servers, in sovereign Cloud as well as On-Premise. No meeting excerpt is transmitted to OpenAI, Anthropic or Google.