The Paris-based startup Remanence announces on 17 November a $30 million seed round led by EQT Ventures and FirstMark. Founded by three former H Company employees, it aims to enable companies to turn the data produced by their AI agents into specialised models that they own.

From agent traces to specialised models

The starting point is a field observation. Companies deploy agents based on generalist models, token consumption rises with task complexity, and the data these agents produce while working often remains unused. Open-weight models offer another path, as they can be adapted to an organisation's tools and processes. But post-training-the phase where an already trained model is specialised-requires rare experts, computing power and custom-built training environments. Few companies have the means.

Remanence automates this step. Its platform connects to the models a company already uses, converts their execution traces into training environments, then uses them to specialise open-weight models through reinforcement learning. Each new batch of traces enables retraining, from which the company expects models that are more accurate on targeted tasks, faster and cheaper to run, as the company's processes evolve. The customer keeps the model weights. Its data can remain in its own infrastructure, and it chooses where training and deployment take place.

Three former H Company employees

Remanence was founded by Pierre-Louis Cedoz, its CEO, Hamza Benchekroun and Axel Moyal. All three worked together at H Company, the French lab specialising in agents capable of controlling a computer, where they led post-training of the open Holo models. Their backgrounds also include Google DeepMind, InstaDeep and Meta.

"We have seen first-hand the power of post-training, and how difficult it remains to access for most companies," explains Pierre-Louis Cedoz. For Matt Turck, partner at FirstMark, enterprise AI "is becoming hybrid": closed models from major labs will coexist with models that companies own and control.

A GPU cluster and hiring in Paris

The round also includes Zero Prime, Emblem, Drysdale, Global Founders Capital and Yellow. Several entrepreneurs are participating personally, including Olivier Pomel (Datadog), Philippe Corrot (Mirakl), Romain Niccoli (Pigment), Victor Riparbelli and Steffen Tjerrild (Synthesia) and Mehdi Ghissassi (AI71).

The funds will go to research and computing infrastructure, with a dedicated GPU cluster, as well as the development of reinforcement learning environments and evaluation tools. The company, which will hire mainly in Paris and plans an initial presence in London and San Francisco, is already working with design partners in legal, customer support and enterprise software, in Europe and the United States.

The topic resonates with French executives. In a TCS study published in early November, 94% of French leaders consider independence from models essential. For a seed round, $30 million remains a high amount. It reflects investor interest in enterprise AI where the model belongs to the user.

Quotes translated from English by ActuIA.

Our articles will then appear first in Google Top Stories.