recapro.ai

Barometer

Where are your meetings processed? The sovereignty barometer for AI meeting-notes tools

Fourteen AI transcription and meeting-notes tools, compared on what matters for confidentiality: the vendor’s country, the hosting location, the use of a third-party language model for inference, and the availability of an on-premise mode.

This barometer is published by Recapro, one of the vendors compared below: the data comes from public sources, but this context should be taken into account. Each cell is drawn from the relevant vendor’s public documents (privacy policies, processing agreements, security pages), consulted on 14 June 2026, and links to its source. “Not published” means the information is not publicly available at that date; this absence does not prejudge the vendor’s actual practices. Vendors can change these elements at any time and report any correction to [email protected]. Recapro, the publisher of this barometer, is pinned at the top as a comparison reference; the other tools are listed alphabetically. The “On-premise pricing public” column indicates whether the price of an on-premise or enterprise deployment is publicly shown on the vendor’s site (“No on-premise offer” = no on-premise server deployment identified; “on quote” = an offering exists but no public price). The table covers meeting tools; for the healthcare case, see the sovereignty explainer.

Recapro (France)

Publisher of this barometer

Hosting
France Nuage / Scaleway (France)
Third-party LLM at inference
None (local inference)
On-premise / air-gap
Yes (air-gap)
On-premise pricing public
Yes (public grid)

Dicte.ai (France)

Hosting
Scaleway (Paris)
Third-party LLM at inference
Not specified (EU / open-source models per the vendor)
On-premise / air-gap
Yes (Enterprise)
On-premise pricing public
No (on quote)

Fathom (United States)

Hosting
AWS (United States)
Third-party LLM at inference
OpenAI, Anthropic
On-premise / air-gap
No
On-premise pricing public
No on-premise offer

Fireflies (United States)

Hosting
United States
Third-party LLM at inference
OpenAI, Anthropic
On-premise / air-gap
Not published by the vendor
On-premise pricing public
No on-premise offer

Gladia (France)

Hosting
OVH / Shadow (France) + United States
Third-party LLM at inference
OpenAI, Together, Meta, Mistral
On-premise / air-gap
Not published by the vendor
On-premise pricing public
No (on quote)

Granola (United States / United Kingdom)

Hosting
AWS (United States)
Third-party LLM at inference
OpenAI, Anthropic
On-premise / air-gap
No
On-premise pricing public
No on-premise offer

HeyPocket (United States)

Hosting
Not published by the vendor
Third-party LLM at inference
Not published by the vendor
On-premise / air-gap
No
On-premise pricing public
No on-premise offer

Leexi (Belgium)

Hosting
AWS (Ireland)
Third-party LLM at inference
Azure OpenAI; Mistral optional
On-premise / air-gap
Not published by the vendor
On-premise pricing public
No on-premise offer

MeetGeek (Romania)

Hosting
AWS (Ireland)
Third-party LLM at inference
OpenAI
On-premise / air-gap
No
On-premise pricing public
No on-premise offer

Modjo (France)

Hosting
AWS (France)
Third-party LLM at inference
OpenAI, Microsoft, Google
On-premise / air-gap
No
On-premise pricing public
No on-premise offer

Noota (France)

Hosting
Google Cloud (EU)
Third-party LLM at inference
Not published by the vendor
On-premise / air-gap
Yes (Enterprise)
On-premise pricing public
No (on quote)

Otter.ai (United States)

Hosting
AWS (United States)
Third-party LLM at inference
OpenAI, Anthropic
On-premise / air-gap
No
On-premise pricing public
No on-premise offer

Plaud (United States / Hong Kong)

Hosting
United States
Third-party LLM at inference
OpenAI, Google
On-premise / air-gap
No
On-premise pricing public
No on-premise offer

tl;dv (Germany)

Hosting
Google Cloud / Hetzner (EU)
Third-party LLM at inference
Anthropic, Mistral (user choice)
On-premise / air-gap
No
On-premise pricing public
No on-premise offer

Why hosting alone does not guarantee confidentiality: see the AI sovereignty & inference explainer.

Sources (consulted 14 June 2026)

Frequently asked questions

01

Which AI meeting-notes tools host your meetings outside the European Union?
Of the fourteen tools compared, several host in the United States, notably Fathom, Fireflies, Granola, Otter.ai and Plaud; Gladia keeps part of its stack there. The others are EU-hosted, except HeyPocket whose hosting is not published. European hosting, however, says nothing about whether a third-party model is called at inference time.
02

Which tools send inference to a third-party language model?
Most tools in the barometer call an external LLM (OpenAI, Anthropic, Google or Mistral) to transcribe or summarise. Recapro is the only one to state fully local inference on its own models. Dicte.ai mentions EU or open-source models without naming the operator; HeyPocket and Noota do not publish the information. “Not published” does not prejudge the actual practices of those vendors.
03

Which AI transcription tools offer an on-premise mode?
Three vendors advertise an on-premise deployment: Dicte.ai and Noota through a quote-based Enterprise offer, and Recapro with an air-gapped mode whose pricing is public. The other tools in the barometer run in the cloud only, with no option to install in your own infrastructure.
04

What is the best AI meeting-notes tool in France?
The right choice depends on your sovereignty requirement. Based on the public privacy policies reviewed in the barometer, US-headquartered tools (including Otter.ai and Fireflies) are subject to US jurisdiction, which exposes them to data-disclosure requests under the CLOUD Act. For processing in France with deployment inside your own infrastructure, Recapro and Dicte.ai meet the criteria of France-based hosting and an on-premise mode; European cloud tools host in the EU without an on-premise mode. Each barometer row links to its public source.
05

What is a French or sovereign alternative to Otter.ai, Fireflies or tl;dv?
These three tools host outside France. Based on their public privacy policies referenced in the barometer: Otter.ai and Fireflies host in the United States, tl;dv hosts in Germany. On inference, the public information per vendor is detailed in the barometer table; some rely on third-party language models, others have not published this information. A « not published » entry does not prejudge a vendor’s actual practices. To stay in France with controlled inference, Recapro (local inference and air-gap mode) and Dicte.ai are among the vendors listed.
06

Which AI meeting-notes tool suits a regulated sector such as banking or legal?
In a regulated sector, a key question is where content is processed and who can technically access it. Per the barometer, tools with an on-premise mode deployable in your own infrastructure keep processing within your physical perimeter, without transit to a third party. Recapro offers such an air-gap mode. Depending on your security policy and regulatory requirements, other factors such as end-to-end encryption, certifications and data residency also matter; the banking, insurance and legal briefings in the barometer cover these dimensions in detail.
07

How is this barometer compiled?
Each cell comes from the vendor’s public documents (privacy policy, data-processing agreement, security page), reviewed on 14 June 2026, with a link to the source. “Not published” means the information is not publicly available, without prejudging actual practices. A vendor can report a correction to [email protected].

The only row where inference never leaves your perimeter

Sovereign French Cloud or air-gapped On-Premise: with Recapro, your meetings are processed by models running on our servers, never by a third-party LLM.

See plans and pricing Read the explainer