Output Parser

Overview

The Output Parser node takes raw text produced by a language model and coerces it into a structured, typed object defined by a JSON Schema. It supports JSON, YAML, and Markdown table source formats. When strict mode is enabled the node halts the workflow on a parse failure instead of forwarding a partial result, making it the right choice whenever downstream nodes depend on a guaranteed shape.

Configuration

Field	Type	Required	Default	Description
schema	object	Yes	—	JSON Schema (draft-07) describing the target output shape. All properties are validated after parsing.
format	string	No	"json"	Expected format of the model's raw text. One of "json", "yaml", or "markdown_table".
strict	boolean	No	false	When true, a parse or validation failure routes to the error port and stops the node from emitting on output. When false, a best-effort partial object is emitted with a _parseWarnings field.
trim_code_fences	boolean	No	true	Strip leading/trailing markdown code fences (```) before parsing. Most LLMs wrap JSON blocks in fences; this option handles that automatically.

Inputs

Port	Type	Description
text	string	Raw text output from an LLM or any prior node.

Outputs

Port	Type	Description
output	object	Parsed and schema-validated object. Shape matches config.schema.
error	object	Emitted only when strict: true and parsing fails. Contains message (string) and raw (the original text).

Example

{
  "nodeType": "output_parser",
  "config": {
    "format": "json",
    "strict": true,
    "trim_code_fences": true,
    "schema": {
      "type": "object",
      "required": ["title", "score", "tags"],
      "properties": {
        "title":  { "type": "string" },
        "score":  { "type": "number", "minimum": 0, "maximum": 10 },
        "tags":   { "type": "array", "items": { "type": "string" } }
      }
    }
  }
}

Notes

• When format is "markdown_table", each table row becomes an object using the header row as keys; only the first table in the text is parsed.
• The schema field is validated at workflow-save time — an invalid schema prevents publishing the workflow.
• In non-strict mode, _parseWarnings is an array of strings describing each validation violation; downstream nodes should check for its presence before trusting the shape.
• Pairing this node directly after a prompt_template or chain node is the most common pattern for extracting structured judgement results from an LLM.