
Remanence raises $30 million to train enterprise models from agent activity
The AMW Read
The funding introduces a company focused on turning agent records into training and evaluation tasks, with impact concentrated in enterprise model development.
Remanence raises $30 million to train enterprise models from agent activity
Paris-based Remanence has raised $30 million in a seed round co-led by EQT Ventures and FirstMark. Zero Prime, Emblem, Drysdale, Global Founders Capital and Yellow also participated. Three former executives of French AI startup H founded Remanence. The company is developing a platform that uses records of enterprise AI-agent activity to train models for individual businesses.
Remanence sits at the intersection of agent evaluation and model training. It says its platform identifies failed tasks in agent records, generates new tasks and verification tools, and runs them in an isolated environment resembling the live service. It then uses reinforcement learning to train open-weight models. The approach treats operational records as material for improving models on company-specific work. Its value will depend on how well the generated tasks and checks reflect production conditions.
For builders and investors, the practical question is whether the resulting models perform better on real workflows. Buyers can evaluate the platform using representative failed tasks and compare outcomes before and after training. Deployment is another consideration: Remanence says it can host the trained models, or customers can run them in their own cloud, a dedicated virtual private cloud or on internal servers. The source does not report customer results, training costs or performance benchmarks.