Documentation

# Agent Definitions

Explore this guide to learn how ChatboxAI works and how to integrate it into your workflow.

Back to overview

Agent Definitions

Define agent roles, personas, and capabilities for your ChatboxAI-powered AI systems.

What is an Agent Definition?

An agent definition specifies:

Identity — who the agent is and what it does
Instructions — behavioral guidelines and constraints
Tools — capabilities the agent can use
Memory — how context is maintained
Guardrails — safety boundaries

Basic Agent Definition

JAVASCRIPT

const supportAgent = {
  name: "SupportBot",
  model: "chatboxai/gpt-oss-120b",
  instructions: `
    You are a customer support specialist for ChatboxAI.
    
    Your responsibilities:
    - Answer questions about the API and documentation
    - Help debug integration issues
    - Escalate complex issues to human support
    
    Guidelines:
    - Always be professional and empathetic
    - Provide code examples when relevant
    - Never make promises about refunds or account changes
    - If unsure, say "I'll need to check on that"
  `,
  tools: [webSearchTool, documentLookupTool],
};

System Prompt Patterns

Persona Definition

CODE

You are Alex, a senior software engineer at ChatboxAI.
You have 10 years of experience building AI-powered APIs.
You communicate in a friendly but technical way.

Task Boundaries

CODE

You ONLY help with API integration questions.
If asked about unrelated topics, politely redirect:
"I'm specialized in API integrations. For [topic], please contact [resource]."

Output Format Constraints

CODE

Always structure your responses as:
1. Direct answer (1-2 sentences)
2. Code example if applicable
3. Next steps or references

Keep responses under 400 words unless more detail is explicitly requested.

Tool-Augmented Agents

JAVASCRIPT

const researchAgent = {
  name: "ResearchBot",
  model: "chatboxai/gpt-oss-120b",
  instructions: "Research topics thoroughly using web search and provide cited summaries.",
  tools: [
    { type: "function", function: webSearchFunction },
    { type: "function", function: extractUrlFunction },
  ],
};

Multi-Model Routing

Use different models for different subtasks:

JAVASCRIPT

async function routeToAgent(task) {
  if (task.type === "code") {
    return await runAgent({ ...codeAgent, model: "chatboxai/gpt-oss-120b" }, task.input);
  } else if (task.type === "summary") {
    return await runAgent({ ...summaryAgent, model: "chatboxai/llama-3.3-70b" }, task.input);
  } else {
    return await runAgent({ ...generalAgent, model: "chatboxai/gpt-oss-20b" }, task.input);
  }
}