GPTSwarm is an open-source framework that models language agents as computational graphs, enabling the development, integration, and automatic optimization of various LLM-based agents. Each node within the graph represents a function to process multimodal data or query other LLMs, while edges define the information flow between operations and agents. This structure allows for the creation of composite graphs representing hierarchies of inter-agent collaboration. GPTSwarm introduces automatic graph optimizers that refine node-level LLM prompts and improve agent orchestration by modifying graph connectivity, thereby enhancing the efficiency and effectiveness of AI-driven problem-solving.