Abstract
We introduce tulip agent, an architecture for autonomous LLM-based agentswith Create, Read, Update, and Delete access to a tool library containing apotentially large number of tools. In contrast to state-of-the-artimplementations, tulip agent does not encode the descriptions of all availabletools in the system prompt, which counts against the model's context window, orembed the entire prompt for retrieving suitable tools. Instead, the tulip agentcan recursively search for suitable tools in its extensible tool library,implemented exemplarily as a vector store. The tulip agent architecturesignificantly reduces inference costs, allows using even large tool libraries,and enables the agent to adapt and extend its set of tools. We evaluate thearchitecture with several ablation studies in a mathematics context anddemonstrate its generalizability with an application to robotics. A referenceimplementation and the benchmark are available atgithub.com/HRI-EU/tulip_agent.