HyperAgent Improves LLM Tool-Use with Schema Hypergraphs.

Large language model (LLM) agents are increasingly used for complex real-world tasks, often relying on external tools. However, their ability to reliably plan and execute tool use is hampered by limitations in implicit reasoning and the dynamic nature of execution environments. Current approaches typically depend on LLMs to infer how tools should be combined from textual descriptions, which can lead to inefficient exploration and unreliable task completion. This research introduces HyperAgent, a novel framework designed to address these challenges. HyperAgent models the relationships between tools at a schema level, constructing a "Tool-Schema Hypergraph" where tools are represented as hyperedges connecting input and output schema nodes. This hypergraph guides a dynamic planning and execution process. For any given task, HyperAgent first extracts a relevant tool context graph to build a schema-aware Task Directed Acyclic Graph (DAG). During execution, it dynamically fulfills subtasks by creating a state-conditioned tool support graph, identifying unmet requirements, and retrieving appropriate tools, leading to improved performance and reduced resource usage.