How to Build AI Agents in PHP Using Laravel Tool Calling Pattern

A developer has shared a production-tested pattern for building AI agents in Laravel that goes beyond simple chatbots by enabling models to call real functions. The approach uses three plain Laravel components — an AgentController, an AgentRunner loop, and a ToolRegistry — to let an AI model request data from actual services like databases instead of fabricating answers. Each tool is a single PHP class with a defined name, description, parameter schema, and a handle method that returns results as strings fed back into the conversation. The runner enforces a maximum number of rounds to prevent infinite loops, while authorization logic is scoped inside each tool's code rather than relying on prompts. The author emphasizes that every tool must be treated like a public API endpoint, since the model — and potentially malicious users — can influence which tool is called and with what arguments.