AI & ML Nodes

I nodi AI & ML permettono di integrare modelli di machine learning, API AI e automazioni intelligenti nei workflow.

ML Inference

Esegue inferenza su modelli ML locali o remoti.

Configurazione

Provider supportati

OpenAI Integration

Workflow Esempio: Chatbot Support

type: mlInference
provider: openai
model: "gpt-4-turbo"
apiKey: "{{secrets.OPENAI_API_KEY}}"
input:
  messages:
    - role: "system"
      content: |
        Sei un assistente clienti per Example Corp.
        Rispondi sempre in italiano.
        Se non sai qualcosa, suggerisci di contattare support@example.com
    - role: "user"
      content: "{{trigger.data.message}}"
  temperature: 0.7
  max_tokens: 500
timeout: 30000
label: "GPT-4 Customer Support"

Output

{
  "provider": "openai",
  "model": "gpt-4-turbo",
  "response": {
    "id": "chatcmpl-abc123",
    "choices": [
      {
        "index": 0,
        "message": {
          "role": "assistant",
          "content": "Ciao! Sono qui per aiutarti..."
        },
        "finish_reason": "stop"
      }
    ],
    "usage": {
      "prompt_tokens": 150,
      "completion_tokens": 85,
      "total_tokens": 235
    }
  },
  "latency": 1234
}

Sentiment Analysis

Analizza il sentiment di testo.

Workflow Esempio: Review Analysis

type: mlInference
provider: huggingface
model: "cardiffnlp/twitter-roberta-base-sentiment-latest"
apiKey: "{{secrets.HF_API_KEY}}"
input:
  text: "{{trigger.data.reviewText}}"
label: "Sentiment Analysis"

Output

{
  "provider": "huggingface",
  "model": "cardiffnlp/twitter-roberta-base-sentiment-latest",
  "response": [
    { "label": "positive", "score": 0.92 },
    { "label": "neutral", "score": 0.06 },
    { "label": "negative", "score": 0.02 }
  ],
  "latency": 234
}

Image Generation

Genera immagini con AI.

Workflow Esempio: Social Media Automation

type: mlInference
provider: openai
model: "dall-e-3"
apiKey: "{{secrets.OPENAI_API_KEY}}"
input:
  prompt: "{{gptPrompt.output.response.choices[0].message.content}}"
  size: "1024x1024"
  quality: "hd"
  style: "vivid"
label: "Generate Social Image"

Output

{
  "provider": "openai",
  "model": "dall-e-3",
  "response": {
    "data": [
      {
        "url": "https://oaidalleapiprodscus.blob.core.windows.net/...",
        "revised_prompt": "A vibrant digital illustration..."
      }
    ]
  },
  "latency": 15000
}

Local Models (Ollama)

Esegui modelli localmente con Ollama.

Workflow Esempio: Document Processing

type: mlInference
provider: local
endpoint: "http://localhost:11434"
model: "llama2"
input:
  prompt: |
    Riassumi il seguente documento in 3 punti chiave:
    
    {{transform.output.extractedText}}
  options:
    temperature: 0.3
    num_predict: 500
label: "Local LLM Summary"

Embeddings

Genera embeddings per ricerca semantica.

Workflow Esempio: Semantic Search

type: mlInference
provider: openai
model: "text-embedding-3-small"
apiKey: "{{secrets.OPENAI_API_KEY}}"
input:
  input: "{{trigger.data.query}}"
  dimensions: 512
label: "Generate Query Embedding"

Output

{
  "provider": "openai",
  "model": "text-embedding-3-small",
  "response": {
    "data": [
      {
        "embedding": [0.0023, -0.0045, 0.0123, ...],
        "index": 0
      }
    ],
    "usage": {
      "prompt_tokens": 8,
      "total_tokens": 8
    }
  }
}

OpenAPI Importer

Importa specifiche OpenAPI per generare automaticamente nodi API.

Configurazione

Workflow Esempio: API Discovery

type: openApiImporter
specUrl: "https://api.example.com/openapi.json"
baseUrl: "https://api.example.com/v2"
auth:
  type: bearer
  token: "{{secrets.API_TOKEN}}"
label: "Import Example API"

Output

{
  "title": "Example API",
  "version": "2.0.0",
  "baseUrl": "https://api.example.com/v2",
  "endpoints": [
    {
      "path": "/users",
      "method": "GET",
      "operationId": "listUsers",
      "summary": "List all users",
      "parameters": [
        { "name": "limit", "in": "query", "type": "integer" }
      ]
    },
    {
      "path": "/users/{id}",
      "method": "GET",
      "operationId": "getUser",
      "summary": "Get user by ID"
    }
  ],
  "schemas": {
    "User": { "type": "object", "properties": {...} }
  }
}

API Test Suite

Esegue test automatici su API importate da OpenAPI.

Configurazione

Workflow Esempio: API Testing Pipeline

type: apiTestSuite
source: openapi
openApiSpec: "{{openApiImporter.output}}"
mode: sequential
endpoints:
  - operationId: "listUsers"
    expectedStatus: 200
    validateSchema: true
  - operationId: "getUser"
    pathParams:
      id: "user_123"
    expectedStatus: 200
  - operationId: "createUser"
    body:
      name: "Test User"
      email: "test@example.com"
    expectedStatus: 201
timeout: 10000
label: "API Integration Tests"

Output

{
  "mode": "sequential",
  "totalTests": 3,
  "passed": 2,
  "failed": 1,
  "duration": 1234,
  "results": [
    {
      "operationId": "listUsers",
      "status": "passed",
      "responseStatus": 200,
      "responseTime": 145,
      "schemaValid": true
    },
    {
      "operationId": "getUser",
      "status": "passed",
      "responseStatus": 200,
      "responseTime": 89
    },
    {
      "operationId": "createUser",
      "status": "failed",
      "responseStatus": 500,
      "error": "Internal Server Error",
      "responseTime": 234
    }
  ]
}

Workflow Completo: AI Content Pipeline

Pipeline completa per generazione e moderazione contenuti.

Rate Limiting e Costi

Gestione rate limits

type: mlInference
provider: openai
model: "gpt-4-turbo"
rateLimit:
  maxRequests: 60
  perMinutes: 1
  onLimitReached: queue  # queue, error, skip

Monitoraggio costi

type: transform
expression: |
  {
    timestamp: new Date().toISOString(),
    model: mlInference.output.model,
    tokens: mlInference.output.response.usage.total_tokens,
    cost: calculateCost(
      mlInference.output.model,
      mlInference.output.response.usage
    )
  }

Best Practices

type: mlInference
provider: openai
model: "gpt-4-turbo"
fallback:
  provider: anthropic
  model: "claude-3-sonnet"
retries: 3
retryDelay: 1000

cache:
  enabled: true
  ttl: 3600  # 1 ora
  keyFields: ["messages", "model"]