Built-in Tools

Tools that extend AI capabilities with search, code execution, and more.

What You'll Learn

• Overview - What tools are and when to use them
• Built-in Tools - Available built-in tools
• RAG Search - Search your documents
• Web Search - Search the internet
• Code Execution - Run code securely
• MCP Tools - Create custom tools

Overview

Tools let AI perform actions beyond text generation. Agents use tools to complete complex tasks.

How Tools Work

1. Request - AI determines it needs to use a tool
2. Execute - Tool runs with provided parameters
3. Observe - AI receives tool results
4. Continue - AI uses results to generate response

When to Use Tools

Scenario	Recommended
Simple Q&A	No tool needed
Need current data	Web search
Need internal docs	RAG search
Need calculations	Code execution
Need custom actions	MCP tools

Available Tools

Tool	Description	Use Case
RAG Search	Search uploaded documents	Internal knowledge, documentation
Web Search	Search internet	Current events, real-time data
Code Execution	Run code in sandbox	Calculations, data analysis
MCP Tools	Custom integrations	APIs, databases, services

Tool Usage Example

import requests

# Create agent with tools
response = requests.post(
    "http://localhost/api/v1/agent/configs",
    headers={"Authorization": "Bearer YOUR_API_KEY"},
    json={
        "name": "research_agent",
        "system_prompt": "You are a research assistant. Use tools to find information.",
        "model": "gpt-4",
        "builtin_tools": ["rag_search", "web_search"],
        "tool_choice": "auto",
        "max_iterations": 5
    }
)

agent_id = response.json()["id"]

# Chat with agent
chat_response = requests.post(
    "http://localhost/api/v1/agent/chat",
    headers={"Authorization": "Bearer YOUR_API_KEY"},
    json={
        "agent_id": agent_id,
        "message": "What are the latest AI developments and what do our docs say about them?"
    }
)

result = chat_response.json()
print(f"Response: {result['response']}")
print(f"Tools used: {[t['tool_name'] for t in result.get('tool_calls', [])]}")

Tool Configuration

# Configure tools for an agent
agent_config = {
    "name": "full_stack_agent",
    "system_prompt": "You are a helpful assistant with access to various tools.",
    "model": "gpt-4",
    "builtin_tools": ["rag_search", "web_search", "code_execution"],
    "tool_config": {
        "rag_search": {
            "default_collection": "documentation",
            "top_k": 5,
            "similarity_threshold": 0.7
        },
        "web_search": {
            "max_results": 10,
            "time_range": "week"
        },
        "code_execution": {
            "timeout": 60,
            "allowed_libraries": ["pandas", "numpy", "matplotlib"]
        }
    },
    "tool_choice": "auto",
    "max_iterations": 10
}

Direct Tool Calling

Use tools directly without agent orchestration:

# Call RAG search
curl -X POST http://localhost/api/v1/agent/tool-calling \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "tool_name": "rag_search",
    "parameters": {
      "query": "API configuration",
      "collection": "docs",
      "top_k": 5
    }
  }'

See Direct Tool Calling for more details.

Best Practices

1. Use agents for complex tasks - Let AI decide when to use tools
2. Use direct calls for simple tasks - More control, less overhead
3. Configure appropriately - Set timeouts and limits
4. Monitor usage - Track tool usage and costs
5. Handle errors - Implement retry logic for failed tool calls

Next Steps

• Built-in Tools - All available tools
• RAG Search - Search your documents
• Web Search - Search the internet
• Code Execution - Run code safely
• MCP Tools - Create custom tools