What H Company does
H (often referred to as H Company or the H Company) is a Paris-based AI lab building agentic foundation models — systems trained from the ground up to plan, decompose, and execute multi-step tasks rather than respond to prompts one turn at a time. H positions itself between traditional LLM labs and pure agent-orchestration tooling: instead of wrapping a third-party model, it trains its own foundation models with action and tool-use baked in.
In 2025 H released Runner H, an agent platform aimed at automating browser-based and enterprise workflows, alongside an SDK for developers. The company has been one of the most-watched names in European AI since its founding, and made a major hire when Gautier Cloix (formerly CEO of Palantir France) replaced founding CEO Charles Kantor in June 2025. H also acquired confidential-AI startup Mithril Security in May 2025 to add privacy-preserving inference to its stack.
Who it's for
H targets enterprises and developers that want autonomous agents capable of handling real workflows — back-office automation, customer-service operations, RPA replacements, and browser-based tasks across regulated industries.
Pricing
H offers usage-based pricing for its hosted Runner H platform plus enterprise contracts; pricing details for the foundation models themselves are not public.
Team & funding
H was founded in 2024 in Paris by Charles Kantor (founding CEO), Karl Tuyls, Laurent Sifre, Julien Perolat, and Daan Wierstra — the latter four all former Google DeepMind scientists. The company raised a $220M seed round announced May 21, 2024 — the largest seed round in European tech history. Investors include Accel, UiPath co-founder Daniel Dines, Eric Schmidt, Xavier Niel, Yuri Milner, Bernard Arnault (via Aglae Ventures), Motier Ventures, Bpifrance Large Venture, Creandum, Elaia, Eurazeo, FirstMark Capital, Visionaries Club, Amazon, and Samsung. Gautier Cloix took over as CEO in June 2025.
Position vs competitors
H competes with frontier labs like Mistral, Anthropic, and OpenAI on foundation models, and with agent platforms like Adept (now Amazon) and various vertical agent startups on the action layer. Its differentiator is training agentic capabilities natively into the model.