AI Agents

AI Agents are autonomous entities that reason, use tools, and maintain conversation memory to handle complex AI workflows.

AI Agents are autonomous entities that can reason, use tools, and maintain conversation memory. Inspired by LangChain agents but "Boxified" for simplicity and productivity, agents handle complex AI workflows by automatically managing state, context, and tool execution.

🎯 What are AI Agents?

An agent is more than a simple chat interface — it's an intelligent entity that:

• Maintains Memory — Remembers conversation history across interactions

• Uses Tools — Can call functions to access data, perform calculations, or interact with systems

• Applies Skills — Injects domain knowledge into the system context (v3.0+)

• Runs Middleware — Intercepts lifecycle events for logging, retry, and guardrails (v3.0+)

• Reasons and Plans — Determines when and how to use tools to accomplish tasks

• Delegates to Sub-Agents — Can orchestrate specialized sub-agents for complex tasks

• Integrates with Pipelines — Works seamlessly as a stage in BoxLang AI pipelines

🏗️ Agent Architecture

🔄 Agent Decision Flow

📖 Sub-Pages

Page	Description
Getting Started	Creating agents, configuration, model selection, return formats
Class-Based Agents	Build encapsulated reusable agents by extending AiAgent
Memory Management	Memory types, per-call identity routing, resume/suspend, multi-tenant
Tools & MCP	Tool Registry, ClosureTool, MCP server seeding
Skills	Always-on and lazy-loaded skills for domain knowledge injection
Middleware	Lifecycle hooks for logging, retry, guardrails, human-in-the-loop
Sub-Agents & Hierarchy	Parent-child agent delegation patterns
Streaming	Real-time streaming responses
RAG & Document Loading	Knowledge base integration and document retrieval
Transformers	Input/output transformation and structured data extraction
Advanced Patterns	Pipelines, event interception, best practices

Quick Start

// Minimal agent
agent = aiAgent(
    name: "Assistant",
    description: "A helpful AI assistant",
    instructions: "Be concise and friendly"
)

response = agent.run( "What is BoxLang?" )
println( response )

// Full-featured agent (v3.0)
agent = aiAgent(
    name           : "SupportBot",
    description    : "Customer support specialist",
    instructions   : "Help customers with product questions",
    model          : aiModel( "openai" ),
    tools          : [ searchTool, ticketTool ],
    memory         : aiMemory( "cache" ),
    skills         : aiSkill( ".ai/skills" ),
    availableSkills: aiSkill( ".ai/advanced-skills" ),
    middleware     : [ new LoggingMiddleware(), new RetryMiddleware() ],
    mcpServers     : [ { url: "http://tools-server/mcp", toolNames: ["search"] } ]
)

response = agent.run( "My order hasn't arrived" )

Related Documentation

• AI Skills — Full skills reference

• Tool Registry — Global tool registry

• Middleware — Full middleware reference

• Memory Systems — Memory configuration

• Pipelines — Using agents in pipelines