Gym helps you improve models using how your agents actually work in production. Agent Hive can learn from workflow traces (inputs, tools used, outcomes, corrections) to fine-tune models for higher accuracy, better consistency, and stronger alignment to your real tasks. Use Gym when you want a native, platform-managed training loop; while still keeping the flexibility to run workflows with your own model keys when preferred. The result: a fine-tuned model that can be selected inside your flows and used immediately.