Remanence raises USD 30m seed to help companies train their own AI models
The startup turns the execution traces of a company's AI agents into training environments, then uses reinforcement learning to post-train open weight models the customer owns.

Remanence, a Paris based AI startup founded by three former researchers at French AI lab H Company, has raised a USD 30 million seed round co-led by EQT Ventures and FirstMark, Sifted reported on 17 November. The company helps businesses post-train open weight AI models on their own workflows.
Zero Prime, Emblem, Drysdale, Global Founders Capital and Yellow also took part, alongside angels including Datadog chief executive Olivier Pomel and Synthesia co-founders Victor Riparbelli and Steffen Tjerrild, according to Tech Funding News and Tech.eu.
What Remanence builds
As companies deploy AI agents across more workflows, those agents generate execution data that usually goes unused. Remanence's platform connects to a company's existing AI systems and converts those execution traces into tailored training environments. It then uses reinforcement learning to post-train open weight models in those environments, rewarding successful behaviour and adjusting the model's parameters.
As agents produce more data, the models can be retrained to improve accuracy and adapt to changing workflows. Remanence says the approach is designed to lower latency and operating costs, and that customers keep ownership of the model weights. Businesses can also keep their data inside their own infrastructure and choose where training and deployment take place. The company says it is already working with design partners in legal services, customer support and enterprise software in Europe and the US.
The founding team
Remanence was founded by Pierre-Louis Cedoz, Hamza Benchekroun and Axel Moyal, who worked together at H Company, where they led post-training work on its open weight Holo models for computer use, according to Tech Funding News. Their experience also spans Google DeepMind, Meta and InstaDeep. Other angel investors named in the reports include Philippe Corrot of Mirakl, Romain Niccoli of Pigment and Mehdi Ghissassi of AI71.
"We started Remanence to close that gap and help businesses turn their own agent data into specialised models they control," said Cedoz, co-founder and chief executive. Tom Mendoza, partner at EQT Ventures, said the next phase of enterprise AI "will be about who owns the intelligence". Matt Turck, general partner at FirstMark, said the enterprise AI stack is going hybrid, with closed frontier models continuing to play a critical role alongside models that companies own and control.
Use of funds
The money will pay for research, dedicated compute infrastructure including a GPU cluster, and hiring. Remanence will recruit mainly in Paris and set up an early presence in London and San Francisco, Tech Funding News reported.
A crowded but growing layer of the AI stack
Remanence is entering a fast growing market for tools that help businesses train and evaluate AI systems on their own tasks. Tech Funding News pointed to US companies such as Bespoke Labs, which builds simulated enterprise environments, and Mercor, which builds evaluations and training environments for agents. Remanence's distinction, according to the company, is that it builds training environments from a customer's own agent traces rather than a generic set. Tech.eu noted that some businesses are looking to reduce reliance on proprietary models from companies such as OpenAI and Anthropic because of cost.
What it means for founders and investors
A USD 30 million seed is large by European standards, and the round reflects how investors are pricing teams with direct experience of training frontier class models. It also fits a Paris pattern of researchers from local AI labs, here H Company, spinning out to build infrastructure for enterprise AI, in this case with well known French founders writing angel cheques.
For founders building AI products, the bet behind Remanence is worth noting. If specialised open weight models trained on proprietary data can match general purpose models on narrow tasks at lower cost, the competitive moat shifts toward who owns the workflow data. The same logic is behind the routing and fine tuning strategy at Model ML, which raised USD 250 million today. For more on European open weight models, see our report on Mistral Large 4 and on Nebuly's Series A.
Spotted an error? Email contact@europreneursmedia.com and we'll correct it. See our corrections policy and editorial standards.


