Tool use
When an LLM calls a tool to act in the world (send email, query DB) rather than just produce text. This turns the model from a passive "eloquence trap" into an active assistant.
Tool use refers to an LLM's ability to interface with external APIs or software functions to execute tasks or fetch real-time data, effectively shifting the model from a static text generator to an active agent capable of interacting with the outside world.
Tool use is the architectural integration of an LLM with external execution environments, where the model generates structured calls to APIs or functions to perform side-effect-producing operations, thereby transitioning the system from a passive language model into an autonomous agentic workflow capable of state-changing interactions.
evolution
- 2020-06 · historyGPT-3 API Release
The introduction of the OpenAI API allowed developers to programmatically integrate LLMs into external software environments.
- 2022-11 · historyChatGPT Plugins
OpenAI introduced a formal framework allowing the model to interact with third-party APIs like Expedia and WolframAlpha.
- 2023-03 · historyGPT-4 Function Calling
OpenAI released a dedicated API feature that reliably structures model output to trigger specific external functions.
- 2023-07 · historyLangChain Agent Framework
The maturation of agentic frameworks enabled models to autonomously plan and execute multi-step tool-use sequences.