Tool use is how an AI agent interacts with the world beyond generating text. The model decides when to call an external capability — an API, database query, search engine, code interpreter, or messaging channel — and receives structured results it can reason about in the next loop iteration.
Without tools, an LLM can only talk. With tools, it can act.
How tool use works
- You define available tools with names, descriptions, and parameter schemas
- The model receives the user goal and tool definitions in its context
- During reasoning, the model outputs a structured tool call (name + arguments)
- Your runtime executes the tool and returns the result
- The model incorporates the result and decides the next step
This pattern is also called function calling when the tools are exposed as callable functions with typed parameters.
Choosing the right tools
Tool selection is an architecture decision, not an implementation detail. Poor choices create compounding errors — an agent that can query any database will eventually query the wrong one. Good tool design:
- Gives each tool a clear, narrow purpose
- Uses descriptive names the model can match to intent
- Returns structured, parseable results
- Includes validation on inputs before execution
Related reading
Browse the AI agent tools catalog for infrastructure mapped to each layer of an agent blueprint.