European AI-first firms

Some firms build many kinds of software and add AI where it helps. Others put AI models, data and AI agents at the centre of what they sell, and this guide calls those AI-first firms. An AI agent is a program that uses an AI model to carry out a series of steps by itself. After Reveneau and Fluxon, six European AI-first firms appear here: ML6, deepsense.ai, Unit8, Alexander Thamm, dataroots and Fuzzy Labs.

Every fact was read on 21 October 2026, and each sentence says where it comes from. A register is an official public list of companies kept by a government body. Four of the six firms have an entry in a register read for this guide: the Belgian KBO register for the company numbers ML6 and dataroots print on their own sites, the Swiss commercial register for Unit8, and UK Companies House for Fuzzy Labs. For the other two, every fact comes from the firm itself.

Reveneau and Fluxon

Reveneau comes first. It has no office in Europe, and its offices are in San Francisco, at 435 Utah Street, and in Vancouver, on Floor 33, 909 Mainland Street. Reveneau says AI writes 100 percent of its code, and that every change must pass a large eval suite, a set of automated tests written from the project's specification, before release. It says Jev, the model that grades the suite, makes it run ten times faster than its previous grader. Reveneau says it puts the money it saves by using AI in place of more engineers into clients' prices, and that it takes responsibility for the whole project through production and after release. You can read about its AI development service.

Fluxon comes second. Its AI development page carries the heading "Forward-deployed by default." [17] It hires in Poland and Portugal, by its own careers page, and no page read names a European office.

The six firms at a glance

Firm	Head office, and who says so	What its pages describe	Size, by its own count	Anthropic tier
ML6	Ghent (Belgian KBO register)	AI agents, vision, voice and robotics; its own AI platform	100+ AI experts	Select
deepsense.ai	Warsaw (own site)	AI agents, knowledge systems, AI infrastructure	120 AI experts	Select
Unit8	Lausanne (Swiss commercial register)	Data and AI for industrial companies; generative AI on Azure	150+ and 180+ data experts	Select
Alexander Thamm	Munich (own legal notice)	Multi-agent AI systems for companies in Germany, Austria and Switzerland	Not used here	Not listed
dataroots	Leuven (Belgian KBO register)	Generative AI, machine learning, data platforms, in general terms	Not published	Not listed; Talan is Select
Fuzzy Labs	Manchester (own site)	Putting AI models into use with open-source tools	Not published	Not listed

ML6

The Belgian KBO register, the country's register of enterprises, lists Skyhaus BV under enterprise number 0502.515.626, active, with its seat at Esplanade Oscar Van de Voorde 1 in Ghent and a start date of 27 December 2012 [2]. BV marks a Belgian private limited company. ML6's contact page prints that company name and number beside its Ghent office, Skyhaus B.V. Nederland beside its Dutch offices and Skyhaus GmbH beside its German ones [3]. It names offices in Amsterdam, Eindhoven, Berlin and Munich. Its history page says it has worked since 2013, counts "100 + AI Experts at ML6" and shows "400 + Projects delivered" [4]. The project count is ML6's own, and the page lists no projects behind it.

Its homepage menu lists agentic AI solutions, vision AI, physical AI and robotics, voice AI, computer use AI and integration AI [5]. Computer use AI is AI that operates computer programs the way a person would. ML6 also sells its own platform, which it calls Unum Enterprise Superintelligence. So ML6 offers two things: build work for clients, and a product of its own. Its own text in Anthropic's directory names DACH (Germany, Austria and Switzerland) and France and Benelux (Belgium, the Netherlands and Luxembourg) as its office regions. Anthropic lists ML6 at the Select tier [1].

deepsense.ai

deepsense.ai lists its headquarters at al. Jerozolimskie 44 in Warsaw, and its privacy policy names deepsense.ai Sp. z o.o., a Polish limited liability company, at the same address [6]. Its pages print no court register or tax number, so no register entry could be matched to it by an exact number. Its about page says "Providing AI Solutions Since 2014", and its homepage counts 120 AI experts and says "200 commercial AI projects completed" [7]. Both figures are its own.

Its own text in Anthropic's directory calls it an "Applied AI consultancy and Anthropic Service Partner, helping enterprises move Claude from prototype to secure, production-grade deployment." Its homepage shows case studies titled "MCP Server for Conversational Analytics in Beverage Manufacturing", for a client it leaves unnamed, and "From Manual Accounting Workflows to Repeatable AI Automation", for a client it calls "an American real estate developer". MCP, the Model Context Protocol, is a standard way to connect an AI model to a company's own tools and data. Its footer names case studies for Nielsen, AdaCore, NOAA and Volkswagen. Anthropic lists deepsense.ai at the Select tier [1]. Apart from that tier, every fact about deepsense.ai on this page comes from its own pages.

Unit8

The Swiss commercial register, read through its public search service Zefix, lists Unit8 SA with the company number CHE-182.119.429, its seat at Avenue d'Ouchy 4 in Lausanne and a first notice in the Swiss Official Gazette of Commerce dated 20 November 2017 [8]. The same search shows a second company, Unit8 Holding SA, also in Lausanne. A holding company is one whose main job is to own other companies, and how the two are related was not read. Unit8's contact page names offices in Zürich, Bern, Kraków, Wrocław, Warsaw, Munich and Frankfurt [20], and gives Avenue d'Ouchy 4 in Lausanne as its registered office.

Its homepage describes Unit8 as a Swiss data and AI services company. It says its mission is to help companies that did not start as digital businesses get value from their data, using "a mix of data science, analytics and AI" [9]. Its generative AI page says, "Unit8 GenAI Accelerator deploys the infra and code on Azure that powers your GenAI use-cases." [10] Infra is short for infrastructure, the servers and services a system runs on, and Azure is Microsoft's cloud. Its homepage says it has been a "Certified Palantir Partner since 2020", which Palantir was not asked to confirm. Its about page gives two staff counts, 150+ and "180+ data experts and counting" [11]. Its directory text names the United States, Canada and DACH as office regions. Anthropic lists Unit8 at the Select tier [1]. Switzerland is outside the EU, and Eurostat's figures on AI use do not cover it.

Alexander Thamm

Alexander Thamm, which writes its name as Alexander Thamm [at], gives its head office at Sapporobogen 6-8 in Munich on its legal notice page, with the Munich register court and register number 199036 [12]. The German commercial register works through a search form that could not be read by script, so that number was not checked. It names offices in Berlin, Frankfurt, Leipzig, Nuremberg, Essen, Innsbruck in Austria and Zagreb in Croatia.

Its homepage describes its work as agentic and forward deployed AI, and says it builds and runs multi-agent AI systems in production for large companies in the DACH region. A multi-agent system is a set of AI agents that pass work between them. Its about page claims "over 3,500 projects in AI, data science, and data engineering" since 2012 [13]. The page gives no list behind the count. Every fact about Alexander Thamm on this page comes from its own pages, and Anthropic's directory had no entry for it on 21 October 2026 [1].

dataroots

dataroots' site footer prints "dataroots BV" and the VAT number BE0668591405 [15]. The Belgian KBO register lists that number, 0668.591.405, as active, at Arnould Nobelstraat 32 in Leuven, with a start date of 30 December 2016 [14]. The register gives the name TALAN BELGIUM for it, shown since 31 July 2026. The homepage says, "We are part of the Talan Group, a network of digital experts". It names offices in Ghent and Utrecht.

Here is what its pages show. The homepage describes generative AI used to "simplify, understand and extract complex information", predictive models, MLOps and cloud data platforms. The pages read for this guide name no AI client and show no AI project in detail. Anthropic's directory has no entry for dataroots and lists Talan at the Select tier [1]. If you ask dataroots for a proposal, ask which company will sign it, dataroots BV or TALAN BELGIUM, and ask for a description of one AI system it has put into use.

Fuzzy Labs

UK Companies House lists Fuzzy Labs Limited, company 11762819, as active, incorporated on 11 January 2019, with its registered office at 123 Wellington Road South in Stockport [16]. Its own site gives its office at 61 Oxford Street in Manchester. Its homepage says, "Fuzzy Labs use open source MLOps to help build and productionise AI models." MLOps is the work of putting machine learning models into use, then watching and updating them. Open-source tools are tools whose code anyone may read, change and share.

Its case study titled "Building an AI Assistant to Halve Police Case File Prep Time" names South Yorkshire Police as the client and says the work was "Backed by the Police Science and Technology Fund (STAR Fund)" [18]. The word "halve" is the firm's own claim, and the police force's own statement was not read. Fuzzy Labs was not in Anthropic's directory on 21 October 2026 [1].

How to read a project count

Three firms on this page print a count of their projects: ML6's "400 +", deepsense.ai's 200 and Alexander Thamm's "over 3,500". None of the three pages lists the projects behind its number, and no outside source read for this guide checks them. A count shows how a firm describes itself. Ask each firm for two projects like yours, with the client's name, and ask to speak to that client.

Why buyers bring in an AI-first firm

Eurostat, the statistics office of the European Union, asked EU enterprises that had considered AI why they did not use it in 2025. A lack of relevant expertise came first, at 70.89 percent. A lack of clarity about the legal consequences came next, at 52.52 percent, and concerns about data protection and privacy third, at 48.83 percent [19]. An AI-first firm sells the first of these. Ask it how it will handle the other two for your system, in writing.

An AI-first firm fits when the model, the data or the agents are the hard part of the project. If you also need the screens your customers will use, read European product studios that build AI. If you are a large company that wants one supplier across many countries, read large European IT groups with an AI practice or Reveneau's page for enterprise buyers. To describe a project to the first firm on the list, contact Reveneau.