Extractor¶

The Extractor node uses an LLM to extract structured data from unstructured text input. It parses the LLM response into a JSON object according to a schema defined by the connected Output Parser.

Component type: extractor

Ports¶

Inputs¶

Outputs¶

Sub-Components¶

Configuration¶

System Prompt¶

The system prompt should instruct the LLM on what information to extract and in what format. The Output Parser defines the target schema, but the system prompt provides the LLM with context about the extraction task:

Extract the following information from the user's message:
- Full name
- Email address
- Phone number (if mentioned)
- Company name (if mentioned)

Return the data as a JSON object. Use null for fields not found in the text.

The system prompt supports Jinja2 template expressions for dynamic context:

Extract order details from the message.
Customer tier: {{ identify_user_abc123.user_context.tier }}

Output Parser¶

The Output Parser sub-component defines the schema for the extracted data. It controls how the raw LLM response is parsed into the structured extracted output. Connect an Output Parser node to the Extractor via the slate-colored diamond handle.

The parser's configuration specifies the expected fields, their types, and any validation rules. The Extractor sends the LLM response through the parser, which returns a validated JSON object.

Usage¶

Extraction Flow¶

1. The Extractor receives input messages containing unstructured text.
2. The system prompt and Output Parser schema are combined into instructions for the LLM.
3. The LLM analyzes the text and produces a response conforming to the schema.
4. The Output Parser validates and structures the LLM response into a JSON object.
5. The structured object is returned as the extracted output port.

Accessing Extracted Data¶

Downstream nodes can access individual fields from the extracted object using Jinja2 expressions:

{{ extractor_abc123.extracted.name }}
{{ extractor_abc123.extracted.email }}
{{ extractor_abc123.extracted.order_total }}

Or access the entire extracted object:

{{ extractor_abc123.extracted }}

Example¶

Contact information extraction¶

Extract contact details from a free-text customer message:

flowchart LR
    Chat[Chat Trigger] --> Extractor
    Model[AI Model<br/>GPT-4o-mini] -.->|llm| Extractor
    Parser[Output Parser] -.->|output_parser| Extractor
    Extractor --> Code[Code: Save to CRM]

System prompt:

Extract contact information from the user's message.
Be precise with email addresses and phone numbers.
If a field is not mentioned, set it to null.

Output Parser schema:

{
  "name": "string",
  "email": "string",
  "phone": "string | null",
  "company": "string | null",
  "role": "string | null"
}

Input message:

Hi, I'm Jane Smith from Acme Corp. You can reach me at jane@acme.com
or call 555-0123. I'm the VP of Engineering.

Extracted output:

{
  "name": "Jane Smith",
  "email": "jane@acme.com",
  "phone": "555-0123",
  "company": "Acme Corp",
  "role": "VP of Engineering"
}

Order detail extraction¶

Extract structured order data from conversational input for downstream processing:

flowchart LR
    Telegram[Telegram Trigger] --> Extractor
    Model[AI Model<br/>Claude Sonnet] -.->|llm| Extractor
    Parser[Output Parser] -.->|output_parser| Extractor
    Extractor --> Switch
    Switch -->|above_threshold| Approval[Agent: Manager Approval]
    Switch -->|below_threshold| Auto[Code: Auto-Process]

System prompt:

Extract order information from the customer's message.
Amounts should be numeric (no currency symbols).
Dates should be in ISO 8601 format (YYYY-MM-DD).

Extracted output:

{
  "product": "Enterprise License",
  "quantity": 50,
  "unit_price": 99.00,
  "total": 4950.00,
  "requested_delivery": "2026-03-15"
}

The extracted total field can then be evaluated by a downstream Switch node to route high-value orders through a manager approval flow.