OrchDAG: Complex Tool Orchestration in Multi-Turn Interactions with Plan DAGs

This research presents OrchDAG, a novel framework for complex tool orchestration in multi-turn interactions. The authors, Yifu Lu, Shengjie Liu, and Li Dong, propose using Plan DAGs (Directed Acyclic Graphs) to model the dependencies and execution order of multiple tools in AI systems. OrchDAG enables more flexible and efficient tool usage in conversational AI systems, allowing for parallel execution of independent tools while maintaining proper sequencing of dependent operations. The framework is evaluated on various complex tasks requiring multiple tool interactions, demonstrating improved efficiency and success rates compared to sequential approaches.