RL infrastructure for training AI agents
Companies that sell the tooling behind reinforcement learning for agents: sandboxes, hosted environments, runtimes and training stacks used to build, run and score RL environments.
Part of RL environments and simulation, which also covers finished environments, task suites and simulations.
These companies sell the plumbing rather than finished datasets, so check what runs where. General Reasoning hosts environments behind an API and HUD runs them as a cloud service; Runloop rents sandboxes for coding agents; Theta Lab installs on premises and builds private environments from a customer's own traces. Few publish pricing, so expect to ask for a quote or a pilot.
Companies (13), by published funding
Figures as published by each company or the source on its profile. n/a means no source found.
| Company | What it does | Based in | Funding | Partner program |
|---|---|---|---|---|
| Prime Intellect | Open stack covering RL environments, evals, hosted training and inference, sandboxes and GPU rental. | United States | $150M+ | No |
| Fleet | Sells simulated practice environments and evaluation tools for AI agents to labs, cloud providers and enterprises. | United States | $60M | No |
| Bespoke Labs | Builds company-scale RL environments and agent training and evaluation infrastructure for frontier labs and enterprises. | United States | $40M | No |
| HUD | Developer kit and cloud service for creating, running and scoring RL environments, plus a marketplace where vendors sell them. | United States | $16M | No |
| General Reasoning | Hosts RL environments behind an API, plus infrastructure to deploy them and to train or test agents at scale. | United Kingdom | $10.9M | No |
| Veris AI | Simulation and RL platform that mimics enterprise systems and the people who use them, for agent training and testing. | United States | $8.5M | No |
| Runloop | Development sandboxes, orchestration and benchmarking tools for building and tuning coding agents. | United States | $7M | No |
| BenchFlow | Builds simulated workflow environments and benchmarks for agents, and also runs post-training as a managed service on its own runtime. | United States | $1M | No |
| Archal | Sells resettable sandboxes that imitate third-party software APIs so teams can train and test agents without live accounts. | United States | $500K | No |
| Cua | Agent infrastructure company offering desktop sandboxes and runtimes for computer use, plus RL environments, benchmarks and trajectory data. | USA | $500K | No |
| Monte | Infrastructure for continual learning and post-training, used to test models on work tasks and improve specialized agents. | United States | $500K | No |
| Vmax | Markets Campaign, a self-play platform that builds training environments from a developer's own data and evals. | United States | $250K | No |
| Theta Lab | On-premises agent training stack that turns a customer's production traces into private RL environments. | United States | n/a | No |