Remanence

Turning agent execution into compounding intelligence

Built by engineers and researchers from

  • H Company
  • BenevolentAI
  • Meta
  • Google DeepMind
  • InstaDeep
  • Upstart

Remanence is pioneering the next era of enterprise AI, building intelligent systems that continuously learn, adapt, and improve through real-world experience.

We turn complex enterprise workflows into interactive learning environments, combining high-fidelity simulations with frontier reinforcement learning to train specialized models.

Our mission is to turn company’s knowledge and every agent interaction into specialized agents and models that continuously learn and get measurably better over time.

The continual-learning platform for companies running agents in production.

Every run makes
the next one better.

  1. Trace Analysis

    Connect your observability stack. Find where your agent fails.

  2. Environment Synthesis

    New tasks and verifiers from your traces, run in a sandbox that mirrors your application

  3. Long-horizon RL

    Any open-weight model, trained against your live baseline.

  4. Inference

    Fully-managed or in your own cloud.

“AI companies spend more on intelligence every quarter and own none of it. We turn the experience their agents already generate into a model they own, one that improves every time it’s retrained”

Pierre-Louis Cedoz

Co-Founder & CEO Remanence

Pierre-Louis Cedoz, Co-Founder & CEO of Remanence

Why work with us?

  1. Evals that look like production.

    Environments built from your real data, to evaluate and improve your agents on the work they actually do.

  2. Beyond harness optimization.

    After tuning prompts and tools, specializing the model is the next hill to climb, for better accuracy and lower latency.

  3. Frontier quality, lower cost.

    Long-horizon agents on frontier models get expensive. A post-trained open-weight model matches them on your tasks at a fraction of the cost.

  4. Own your AI, don’t rent it.

    Your models, your data, no vendor lock-in. Your data moat compounds into a model you own.

  5. Built for regulated industries.

    Deploy in your VPC or on-prem, with training and serving inside your boundary.

  6. Better with every run.

    Production traces become training tasks, so your model keeps improving as your business evolves.

From consuming, to owning
your intelligence.