A team that helped train some of France's best-known open-weight computer-use models has raised $30 million to give enterprises the same trick. Remanence, a Paris-based startup that came out of stealth on November 18 with a seed round co-led by FirstMark Capital and EQT Ventures, wants to turn the logs your AI agents already generate into custom models your company actually owns.
The founders — CEO Pierre-Louis Cedoz, Hamza Benchekroun and Axel Moyal — previously worked together at H Company, the French computer-use AI lab, where they led the post-training of the open-weight Holo models and set what the company describes as state-of-the-art results on computer-use benchmarks. The wider team includes alumni of Google DeepMind, Meta and InstaDeep.
The pitch starts from an observation most agent deployments ignore: every tool call, failure and correction an agent makes is a record of how the work actually gets done, and almost all of it is thrown away. Remanence plugs into a company's existing models, converts those execution traces into tailored training environments, and then post-trains open-weight models on them with reinforcement learning. The platform covers four steps — analyzing traces to find where agents fail, synthesizing new tasks and verifiers in a sandbox that mirrors the customer's application, long-horizon reinforcement learning, and inference — and can run fully managed or inside the customer's own cloud. As new traces accumulate, the model is retrained and keeps improving on accuracy, cost and latency.
Ownership is the point. Each customer ends up with a specialized model whose weights it keeps, with its data staying inside its own infrastructure. The company claims a post-trained open-weight model can match frontier models on a customer's specific tasks at a fraction of the cost — a claim it has not backed with published performance figures or named customers, and one that should be treated as company-reported. It says it is working with companies in legal, customer support and enterprise software across Europe and the US.
AI companies spend more on intelligence every quarter and own none of it, Cedoz said in the announcement. The money will fund research and compute infrastructure, including a dedicated GPU cluster, plus hiring in Paris, London and San Francisco. Alongside FirstMark and EQT Ventures, the round includes Zero Prime Ventures, Emblem, Drysdale Ventures, Global Founders Capital and Yellow, with angels including Datadog CEO Olivier Pomel, Mirakl co-founder Philippe Corrot, Pigment co-founder Romain Niccoli, Synthesia co-founders Victor Riparbelli and Steffen Tjerrild, and Mehdi Ghissassi of AI71. No valuation was disclosed.
Remanence is entering a race that is filling up fast. Applied Compute, founded by former OpenAI researchers, retrains open-weight models with reinforcement learning on customer data and recently raised $350 million at a $3.25 billion valuation. Trajectory, which builds training signal from expert feedback, raised a $45 million Series A led by Sequoia at a $300 million valuation, and New York's Plurai took a $13 million seed led by Team8 to simulate and harden agents before launch. What distinguishes Remanence is its raw material: rather than embedding engineers with customers, collecting expert judgments or simulating users, it starts from what production agents have already done — betting that the most valuable training data of the agentic era is the exhaust everyone is currently deleting.
Comments (0)
Log in to join the discussion
Log InNo comments yet