Configuration

Creating an AI Agent

1. In the Handit.AI Dashboard, go to Agents

2. Click Create Agent

3. Add Model Nodes and Tool Nodes

4. Click Save

Understanding AI Agents

Handit.AI uses a wrapper-based approach to track AI agents. Each agent can have multiple nodes that represent different processing steps.

Agent Structure

• Model Nodes: AI models like GPT-4, BERT, etc.

• Tool Nodes: External services, APIs, databases

Basic Agent Setup

Node.js

import { startAgentTracing, traceAgentNode } from '@handit.ai/node';

// Define a simple agent
async function simpleAgent(input) {
  const processNode = traceAgentNode({
    agentNodeId: agentsTrackingConfig.mariaManager.documentCompressor,
    callback: async (data) => {
      // Your processing logic here
      return processedData;
    }
  });

  return await processNode(input);
}

// Wrap with tracing
const tracedAgent = startAgentTracing(simpleAgent);

Python

from handit import start_agent_tracing, trace_agent_node

class SimpleAgent:
    @start_agent_tracing
    async def process(self, input_data):
        process_node = trace_agent_node({
            "agent_node_id": agent_config["mar_i_a_manager"]["document_compressor"],
            "callback": self.process_data
        })
        
        return await process_node(input_data)

Advanced Configuration

Error Handling

The wrappers automatically handle errors without breaking your application:

Node.js

const errorHandlingNode = traceAgentNode({
  agentNodeId: agentsTrackingConfig.mariaManager.taskAssigner,
  callback: async (data) => {
    try {
      return await riskyOperation(data);
    } catch (error) {
      // Error will be tracked but won't break your app
      return fallbackResponse;
    }
  }
});

Python

error_handling_node = trace_agent_node({
    "agent_node_id": agent_config["mar_i_a_manager"]["task_assigner"],
    "callback": async_error_handler
})

async def async_error_handler(data):
    try:
        return await risky_operation(data)
    except Exception as e:
        # Error will be tracked but won't break your app
        return fallback_response

Setting Up AI Tracking

Each AI Agent has a tracking slug used to monitor interactions.

Node.js

const { captureAIEntry } = require('@handit.ai/node');

captureAIEntry({
  ai_tracking_slug: "my-agent",
  request_body: { query: "What's the weather?" },
  response_body: { answer: "It's sunny in Bogotá!" }
});

Python

tracker.capture_ai_entry(
    ai_tracking_slug="my-agent",
    request_body={"query": "What's the weather?"},
    response_body={"answer": "It's sunny in Bogotá!"}
)