The ability to use tools, also known as function calls, has become a valuable mechanism for enabling LLMs to expand their capabilities beyond out-of-the-box pre-trained capabilities. This is especially important in bridging the abilities that LLMs are known to lack. Tools also link existing APIs and systems to expand, automate, optimise or enable a broader range of tasks.

Many models, including OpenAI GPTs and Anthropic Claude 3, support tool use out of the box, and third-party frameworks specifically designed to excel in this area are available for models that do not natively support tools and function calls. For example, LangChain has built-in tools to run SQL queries, perform calculations, and perform vector searches. When the built-in tools don’t do what you need, you can define custom tools to expand LangChain’s capabilities.