The Recursive Agents Hackathon

During the PyTorch Conference in San Francisco, build agents that can take a goal and run with it. Let them make the plan, spin up sub-agents, delegate work, communicate with each other, check their own progress, recover when things break, and keep going. Push beyond the one-shot agent into persistent loops that rarely need you involved: agents building teams and handling delegation, memory, and long-running inference that can keep operating after you walk away until the work is complete. Build on Crusoe Managed Inference and put leading open models, including NVIDIA’s Nemotron family, into production on high-performance NVIDIA infrastructure. Compete to win $25,000 in cash prizes and an NVIDIA RTX 5090. Note: This event is fully in-person in San Francisco. Teams may have up to four members. All attendees must apply and be approved to attend.