AgileRL, a startup accelerating the development of Reinforcement Learning for training AI models, today announces their seed, led by Fusion Fund, along with Flying Fish, Octopus Ventures, Entrepreneur First, and Counterview Capital, bringing their total funding to £6 million.
Its platform reduces the time and cost of RL by 10x with performant end-to-end tooling. AgileRL plans to use the funding to open a San Francisco office and hire more than a dozen roles across engineering and go-to-market.
Building an RL program in-house means effectively creating a small AI research lab. You need teams of expensive PhDs, months of trial runs, and big compute budgets. Companies must assemble everything from scratch each time, including simulators, reward design, data collection, hyperparameter search, distributed training, evaluation suites, monitoring, safety guardrails, and deployment pipelines. Every new use case breaks the old setup and starts the work over. It’s slow, expensive, and brittle, and all but the largest technology companies can afford to effectively do it at scale.
AgileRL offers both a free open-source RL platform and Arena, a managed full-stack RLOps platform that handles all the difficult engineering work. Its approach enables training on-policy, off-policy, offline, multi-agent, contextual multi-armed bandits and large language models, alongside evolutionary hyperparameter optimisation, distributed training with multi-GPU support, environment validation, and one-click deployment. The result is streamlined development, a 10x improvement in training speed, and superior AI model performance compared to standard approaches, supported by a community shaped by academic citations and more than 300,000 downloads.
The company’s framework is already being used by labs at institutions including MIT, Roblox, Carnegie Mellon and University of Waterloo, for applications that span defence, robotics, finance, and others.
