The Recursive Agents Hackathon

Under the PyTorch Conference in San Francisco, create agents that can take a goal and execute it. Allow them to devise the plan, create sub-agents, delegate tasks, communicate with one another, monitor their own progress, recover when issues arise, and continue their work. Move beyond single-use agents into ongoing loops that seldom require your intervention: agents forming teams and managing delegation, memory, and long-running inference that can keep functioning after you step away until the task is finished. Build on Crusoe Managed Inference and deploy leading open models, including NVIDIA’s Nemotron family, on high-performance NVIDIA infrastructure. Compete for a chance to win $25,000 in cash prizes and an NVIDIA RTX 5090. Note: This event is entirely in-person in San Francisco. Teams can consist of up to four members. All participants must apply and gain approval to attend.