RunRL›RunRL› offers Reinforcement Learning as a Service

RunRL sells reinforcement learning as a service. A customer supplies a model, a prompt, and a reward function, and the service runs hosted reinforcement learning post-training against that configuration until the model's reward on the target task goes up, replacing manual prompt iteration with managed training runs.

The company pitches the offering at teams that already spend on observability and want that data to drive self-improvement rather than monitoring alone, and at teams that need a model or agent to get better at a specific task without building a reinforcement learning pipeline themselves. The service is operated by a San Francisco company that went through Y Combinator's X25 batch.