Module 7: Tools & APIsLesson 1 of 5

What Tools Actually Are

4 min

Video coming soon

What Tools Actually Are

Tools let your agent affect the real world.

The Limitation of Pure LLMs

Without tools, an LLM can only:

  • Generate text
  • Answer questions
  • Have conversations

It cannot:

  • Read files
  • Browse the web
  • Execute code
  • Send messages
  • Control anything

Tools Bridge the Gap

Tools are functions the agent can call:

Agent thinks: "I need to check the weather" Agent calls: weather_tool("San Francisco") Tool executes: HTTP request to weather API Tool returns: "72°F, sunny" Agent responds: "It's 72°F and sunny in San Francisco"

Tool Categories

Information Tools

  • web_search — Search the internet
  • web_fetch — Read a webpage
  • read — Read local files
  • memory_search — Search memory

Action Tools

  • write — Create/modify files
  • exec — Run shell commands
  • message — Send messages
  • browser — Control a browser

System Tools

  • cron — Schedule tasks
  • sessions_spawn — Create sub-agents
  • gateway — Manage the system

How the Agent Decides

The agent sees tool descriptions and decides:

Available tools: - web_search: Search the web for information - read: Read contents of a file - exec: Execute a shell command User: "What's the weather in Tokyo?" Agent thinks: "This needs current data from the web" Agent action: web_search("Tokyo weather")

The agent matches the task to the right tool.

The Agent Loop

1. Receive input 2. Think about what to do 3. Decide: Respond directly OR use a tool 4. If tool: Execute, get result, continue thinking 5. Generate final response

Tools extend what's possible. Without them, agents are just chatbots.