Skip to content

Tools

One of the most important parts of any agentic system is the set of tools that the agent can use. While an LLM on its own can understand and generate language, it's the ability to use tools that transforms it from simple conversational bot into a true agent capable of taking action, solving problems, and achieving real-world goals.

What Is a Tool?

A tool provides the agent with the ability to interact with external systems, perform actions, or access information that is not contained within the LLM's training data. Tools can be anything from APIs, databases, or even simple functions that perform specific tasks.

At a technical level, tools are often defined as functions that the agent can call. Large Language Models (LLMs) like OpenAI’s or Anthropic’s support something called tool_calls or function_calls. These allow the LLM to output structured JSON.

Common Tools

  • Search the web
  • Execute code in a sandbox
  • Send a message to Slack
  • Call a REST API
  • Read or write to a file system

Think of tools as the hands and eyes of your agent: they allow it to act beyond just thinking and talking.

Example: Connecting Tools to an Agent

Here’s a conceptual diagram of an LLM agent using tools to interact with different systems:

graph TD
    A[LLM Agent] -->|Tool Call| B[SlackTool]
    A -->|Tool Call| C[GitHubTool]
    A -->|Tool Call| D[CodeExecutionTool]
    B --> E[Send Message]
    C --> F[Create Comment]
    C --> G[List Issues]
    D --> H[Run Python Code]

Each tool defines:

  • Name: the identifier used by the agent (e.g. post_to_slack)
  • Description: a brief explanation of what the tool does (e.g. "Posts a message to a Slack channel")
  • Parameters: the inputs the agent must provide (e.g. message, channel)
  • Function: the logic the tool actually executes (e.g. send HTTP request)

Why Tools Matter

Without tools, LLMs are limited to just generating language.

With tools, agents can:

  • Automatically debug code
  • File support tickets
  • Summarize incoming Slack messages
  • Generate and send reports
  • Chain actions across different services