anthropic

Overview

The anthropic node provides access to Anthropic's Claude models. Known for exceptional reasoning, long-context understanding, and nuanced analysis. Ideal for complex data interpretation and decision support in edge computing scenarios.

Properties

Available Models

Inputs & Outputs

// Simple message
{
  "payload": "Analyze this data..."
}

// With conversation
{
  "payload": "Continue analysis",
  "messages": [
    { "role": "user", "content": "Initial..." },
    { "role": "assistant", "content": "Response..." }
  ]
}

{
  "payload": "Claude's response...",
  "usage": {
    "input_tokens": 150,
    "output_tokens": 300
  },
  "model": "claude-3-5-sonnet-20241022",
  "stopReason": "end_turn"
}

Example: Intelligent Anomaly Detection

Use Claude's reasoning to detect and explain anomalies in sensor data.

// Function node: Prepare anomaly analysis request
var historicalData = flow.get("sensorHistory") || [];
var currentReading = msg.payload;

// Add current reading to history
historicalData.push({
    timestamp: new Date().toISOString(),
    ...currentReading
});

// Keep last 100 readings
if (historicalData.length > 100) {
    historicalData = historicalData.slice(-100);
}
flow.set("sensorHistory", historicalData);

msg.payload = `Analyze this time-series sensor data for anomalies:

Current reading:
${JSON.stringify(currentReading, null, 2)}

Historical context (last ${historicalData.length} readings):
${JSON.stringify(historicalData.slice(-10), null, 2)}

Statistical summary:
- Average temperature: ${calculateAverage(historicalData, 'temperature').toFixed(2)}°C
- Std deviation: ${calculateStdDev(historicalData, 'temperature').toFixed(2)}

Identify if the current reading is anomalous and explain why.`;

msg.systemPrompt = `You are an IoT anomaly detection expert. Analyze sensor data
and identify anomalies with clear reasoning. Consider seasonal patterns,
sudden spikes, gradual drifts, and correlation between sensors.`;

return msg;

Example: Natural Language Device Control

Convert natural language to device commands using Claude.

// Function node: Parse natural language command
msg.systemPrompt = `You are an IoT command parser. Convert natural language to JSON commands.
Available devices: living_room_light, bedroom_thermostat, kitchen_fan
Available actions: on, off, set_temperature, set_speed, dim

Output JSON only, no explanation. Format:
{"device": "device_name", "action": "action", "value": optional_value}`;

msg.payload = msg.payload.text; // "Turn on the living room light and set bedroom to 72 degrees"
msg.temperature = 0; // Deterministic output

return msg;

// After Claude response, parse the JSON:
// Function node: Execute commands
var commands = JSON.parse(msg.payload);
commands.forEach(function(cmd) {
    node.send({ topic: cmd.device, payload: cmd });
});

Tool Use (Function Calling)

Enable Claude to call functions/tools for enhanced capabilities.

// Configure tools in the node
msg.tools = [
    {
        name: "get_sensor_reading",
        description: "Get the current reading from a specific sensor",
        input_schema: {
            type: "object",
            properties: {
                sensor_id: {
                    type: "string",
                    description: "The sensor identifier"
                }
            },
            required: ["sensor_id"]
        }
    },
    {
        name: "set_device_state",
        description: "Control a smart device",
        input_schema: {
            type: "object",
            properties: {
                device_id: { type: "string" },
                state: { type: "boolean" }
            },
            required: ["device_id", "state"]
        }
    }
];

msg.payload = "What's the temperature in the living room?";
return msg;