Extending LLM outputs by integrating external tools, APIs, or functions the model can call to solve problems.