Tool Use
How AI models call external functions to act on the world instead of only generating text.
What Is Tool Use
Tool use is the mechanism by which a large language model extends its capabilities beyond generating text by invoking external functions, or "tools," during a conversation or task. Instead of relying only on knowledge encoded in its training data, a model with tool use can call a search engine, read a file, run code, query a database, or interact with any other function exposed to it, then incorporate the result into its next response. This turns a language model from a pure text generator into an agent capable of taking actions and observing their outcomes.
How It Works
A tool is typically defined by a name, a description of what it does, and a schema describing its expected input parameters. The model is given this list of available tools alongside the user's request. When the model determines that completing the task requires information or actions it cannot produce on its own, it emits a structured request to call a specific tool with specific arguments. The calling application executes that tool outside the model, for example running a shell command or hitting an API, and returns the result to the model as part of the ongoing exchange. The model then reads that result and decides whether to call another tool, ask a follow up question, or produce a final answer. This loop can repeat many times within a single task.
Why It Matters
Tool use is what separates a chatbot from an agent. A model without tools can only reason over what it was trained on and whatever text is in its context window. A model with tools can look up current information, manipulate files, execute code, and interact with real systems, which makes it useful for tasks like debugging a codebase, researching a topic, or completing a multi-step workflow. The reliability of an agent's tool use, meaning how well it chooses the right tool, supplies correct arguments, and interprets results, is one of the main factors that determines whether it can complete real work unattended.
Agenhood and Tool Use
Agenhood's built-in Vanilla driver gives every agent a default tool palette: reading, writing, editing, and listing files, running bash and Python, searching the web, and fetching web pages. Agents can be extended further with reusable skills and with Model Context Protocol servers, which expose additional external tools through a single consistent interface. Because every tool call executes inside a hardened, sandboxed Docker container, an agent's tool use is bounded by the container's permissions rather than by the host system, which limits the effect of a mistaken or malicious action.