∞ JSON Query Engine

Documentation

• Class name: LLMJSONQueryEngine
• Category: SALT/Language Toolkit/Querying
• Output node: False

This node is designed to process queries against JSON data using a language model, enabling the extraction of information or execution of tasks based on a JSON schema and data. It leverages a JSON query engine to interpret and respond to queries, making it suitable for applications requiring dynamic data interpretation or manipulation.

Input types

Required

• llm_model
  • Specifies the language model to be used for processing queries. This model is crucial for understanding and generating responses based on the JSON data and schema provided.
  • Comfy dtype: LLM_MODEL
  • Python dtype: Dict[str, Any]
• json_data
  • The actual JSON data on which queries will be executed. This data is parsed according to the provided JSON schema.
  • Comfy dtype: STRING
  • Python dtype: str
• json_schema
  • Defines the structure of the JSON data to be queried. This schema guides the query engine in interpreting the data correctly.
  • Comfy dtype: STRING
  • Python dtype: str
• prompt
  • The query or command to be executed against the JSON data. This prompt drives the interaction with the data.
  • Comfy dtype: STRING
  • Python dtype: str
• human_readable
  • A flag indicating whether the response should be synthesized for human readability. This affects how the query results are presented.
  • Comfy dtype: BOOLEAN
  • Python dtype: bool

Optional

• optional_service_context
  • An optional context that can be provided to customize the service's behavior or processing environment.
  • Comfy dtype: LLM_CONTEXT
  • Python dtype: Dict[str, Any]

Output types

• result
  • Comfy dtype: STRING
  • The direct response from the query execution, containing the information or results extracted from the JSON data.
  • Python dtype: str
• json_path
  • Comfy dtype: STRING
  • Additional data about the query execution, including paths within the JSON data that were accessed to generate the response.
  • Python dtype: str

Usage tips

• Infra type: CPU
• Common nodes: unknown

Source code

class LLMJSONQueryEngine:
    @classmethod
    def INPUT_TYPES(cls):
        return {
            "required": {
                "llm_model": ("LLM_MODEL",),
                "json_data": ("STRING", {"multiline": True, "dynamicPrompts": False, "placeholder": "Enter JSON data here..."}),
                "json_schema": ("STRING", {"multiline": True, "dynamicPrompts": False, "placeholder": "Enter JSON schema here..."}),
                "prompt": ("STRING", {"multiline": True, "dynamicPrompts": False, "placeholder": "Enter your JSON query / question here..."}),
                "human_readable": ("BOOLEAN", {"default": True}),
            },
            "optional": {
                "optional_service_context": ("LLM_CONTEXT",),
            }
        }

    RETURN_TYPES = ("STRING", "STRING")
    RETURN_NAMES = ("result", "json_path")

    FUNCTION = "query_engine"
    CATEGORY = f"{MENU_NAME}/{SUB_MENU_NAME}/Querying"

    def query_engine(self, llm_model, json_schema, json_data, prompt, human_readable, optional_service_context=None):
        try:
            schema = json.loads(json_schema)
            data = json.loads(json_data)
        except json.JSONDecodeError as e:
            error_message = f"JSON parsing error: {str(e)}. Please ensure your JSON schema and data are correctly formatted."
            logger.error(error_message)
            return ("", "")

        query_engine = JSONQueryEngine(
            json_value=data,
            json_schema=schema,
            llm=llm_model['llm'],
            synthesize_response=True if human_readable else False,
            service_context=optional_service_context if optional_service_context else None
        )

        response = query_engine.query(prompt)
        logger.data(response, indent=4)

        return (response.response, response.metadata["json_path_response_str"])