Entity Recognition

Pull structured values out of unstructured text.

Entity Recognition extracts predefined or custom entities from raw text using a configured recognizer. It turns a sentence a person typed into values a workflow can act on — names, dates, contact details, account numbers.

How it works

• Input resolution — takes an input string. $input (the previous skill's output) and $secret can be used to construct that text or to supply credentials for the recognizer.

• Processing — sends the raw string and the selected entityRecognizerId to the backend recognizer, which applies:

  • Standard entity patterns — SSN, Phone Number, Email, Time, Date, Full Name, Zip Code, Address.

  • Custom regex rules, if configured on the recognizer.

• Execution — blocking. Waits for recognition to complete.

• Response — success returns a result array of matched entity strings; failure returns an error object and a status code.

Worked example: extracting details from a free-text query

A customer types: "My name is John Doe, and my email is [email protected]; I need assistance with loan #L12345." The workflow needs the name, the email, and the loan number.

The problem. Pulling specific values out of unstructured text by hand is tedious and error-prone — and customers don't write in your schema.

The approach. Point an Entity Recognition skill at the query text ($input.customerQuery), and let the recognizer do it.

• Standard entities cover the name and the email out of the box.

• Custom regex covers #L12345. Add a pattern to the recognizer with a name, the expression, and any flags — i for case-insensitive. The loan-number format is yours, so the rule has to be too.

Configuration:

{
  "entityRecognizerId": "a6412cdc-664c-4795-b652-cd3d93659da3",
  "input": "John Doe works at Acme Corp in New York. His contact no is [phone].",
  "name": "Entity Recognition Test",
  "description": "This is a test of the Entity Recognition Skill"
}

Output:

{
  "result": ["John Doe", "Acme Corp", "New York", "[phone]"],
  "error": null,
  "statusCode": 200
}

Entity Recognition processing flow

Configuration reference

Field	Type	Required	Description
entityRecognizerId	string	✅	The recognizer model to use.
input	string	✅	Raw text to scan for entities.
name	string	—	Display name for this skill instance.
description	string	—	Description of this skill's purpose.

Supported entities (default recognizer): SSN, Phone Number, Email, Time, Date, Full Name, Zip Code, Address.

Custom regex. Add patterns to the recognizer's configuration for domain-specific extractions — tax IDs, application numbers. Each takes a name, a pattern, and flags.

Output

The output port (entityRecognitionResult) always conforms to:

Field	Type	Always	Description
result	string[]	✅	Matched entity strings from the input.
error	object | null	—	Error object if recognition failed.
statusCode	number	✅	HTTP-like status code.

Failure — invalid entityRecognizerId:

{
  "result": null,
  "error": { "message": "Missing or invalid config, entityRecognizerId not provided" },
  "statusCode": 400
}

Errors

Code	Message	Cause
400	Missing or invalid config	entityRecognizerId not provided or invalid.
422	Input text is empty or malformed	input is blank or not a valid string.
500	Internal server error	Unexpected recognizer failure.

Security

• Recognizer access tokens — if entityRecognizerId points to an external or secured service, use $secret to inject credentials.

• Log redaction — logs redact raw input text and detected entity values.

• Review your regex. Custom patterns should be checked for ReDoS vulnerabilities, which can degrade performance or take the service down. A regex you wrote in a hurry is running against text a stranger typed.

• Extracted entities are often PII. They're redacted in logs, but downstream skills receive them in the clear — mask with PII Guard, encrypt, or store securely as your policy requires.

Testing in isolation

• Path — /skill-runtime/workflows/nodes/EntityRecognition/execute

• Method — POST

• Body:

{
  "config": { "entityRecognizerId": "your-recognizer-id", "input": "Text to scan for entities." },
  "input": {}
}

---

To add this skill to an agent, see Adding a Skill to the Agent.