Skip to main content
工作流应用

Parameter Extraction Node

The Parameter Extraction Node leverages large language models (LLMs) to extract structured parameters from unstructured content. The model automatically identifies and extracts corresponding parameter values from input content, outputting structured data for use by downstream nodes. It supports plain-text extraction; when Vision-Language (VL) series models are selected, it also supports parameter extraction from images and videos. Typical use cases include form information extraction, tool parameter preparation, data format conversion, and image information structuring.

Usage Example

The Parameter Extraction Node extracts structured information from users’ natural-language inputs for use by subsequent nodes.

Node Integration

Method 1: Add via the + Button

  • Hover your mouse over any node on the canvas (e.g., the Start node); a + button automatically appears on its right side.
  • Click the + button and select Variable Parameter Extraction from the pop-up node list. The node is automatically added to the canvas and connected to the upstream node.

Method 2: Drag & Drop from the Node Library

  • In the left-side node library, locate and select Variable Parameter Extraction.
  • Drag the node to an appropriate location on the canvas, then drag a connection line from the edge of an upstream node to this node.

Core Configuration

Model Selection

Select the large language model used for parameter extraction. This model interprets the semantic meaning of input content and converts unstructured information into structured parameters. Supported models include Tongyi Qwen series models and third-party models (e.g., DeepSeek). For the latest list of supported models, refer to the console. Click the configuration icon (⚙️) next to the model selector, enable the toggle, and configure the following parameters:
ParameterDefault ValueDescription
Max Response Length1024Maximum number of tokens in the model’s output.
top_p0.80Controls output diversity — higher values yield more diverse results.
temperature0.70Controls output randomness — higher values yield more random results.
enable_thinkingEnabledEnables deep reasoning mode; outputs step-by-step reasoning traces.
thinking_budget4000Maximum number of tokens allowed for chain-of-thought reasoning.
result_formatmessageFormat of the returned result.
enable_searchDisabledEnables web search functionality.
When selecting VL-series models (e.g., Qwen-VL-Plus-Latest, Qwen-VL-Max-Latest), two additional configuration options appear: Vision Understanding Type and Model Input Parameters, enabling structured parameter extraction from images and videos.— Vision Understanding TypeSpecifies how the model interprets visual inputs.
OptionModel Input Parameter NameDescription
Image (default)vlImageUrlProvide one or multiple images; the model treats each as independent content and matches them to questions accordingly.
Video FramesvlImageUrlProvide ≥4 images representing consecutive frames of the same video; the model processes them sequentially as a unified sequence.
Full VideovlVideoUrlProvide a direct URL to a complete video; the model automatically parses and understands its content without manual frame extraction.
— Model Input ParameterUsed to pass image/video data.Parameter name
Applicable Vision Understanding TypeDescriptionvlImageUrl
Image / Video FrameList of image URLs — supports direct entry or referencing upstream variables.vlVideoUrl

Input

The content from which parameters need to be extracted. Supports referencing user input or outputs from upstream nodes. When using VL models, this input field remains for textual input (e.g., user queries or descriptions), while images/videos are passed separately via model input parameters.

Extracted Parameters

Define the list of parameters to extract from the input text. The model identifies and extracts corresponding values based on each parameter’s name, type, and description.
FieldRequiredDescription
NameYesParameter name (e.g., meeting_date, city). Serves as the key name for the output variable.
Type Yes Data type of the parameter. Defaults to String. Description Yes Natural-language description helping the model understand what value to extract. Clearer descriptions improve extraction accuracy. Required No Indicates whether the parameter must be successfully extracted. Used only as a hint during model inference—not enforced as strict validation.

Prompt

Provide additional context, examples, or constraints to improve extraction accuracy and consistency.

Memory

Disabled by default. When enabled, the model retains conversation history—supporting multi-turn dialogue scenarios.
Cache TypeDescription
Node-level CacheThe model remembers only the conversation history within this node. Configure memory turns (1–50, default: 3); one input-output pair counts as one turn.
Custom CacheThe model retains global context. Select a context variable (e.g., built-in historyList).

Node Output

The Parameter Extraction Node outputs all custom-extracted parameters plus built-in status variables, available for reference by downstream nodes.
  • Custom Extracted Parameters
    Each parameter defined in Extracted Parameters generates an output variable with the same name.
  • Built-in Status Variables
Variable NameTypeDescription
_is_completedBooleanIndicates whether parameter extraction completed successfully. true means all required parameters were extracted; false indicates at least one required parameter was missing.
_reasonStringWhen _is_completed is false, explains why extraction failed. Empty string if extraction succeeded.

Common Use Cases

Paired NodeDescription
Conditional NodeUses _is_completed to determine extraction completeness and route logic accordingly (e.g., proceed if complete; prompt user for missing info otherwise).
API NodePasses extracted structured parameters as input to API calls—for example, extracting date and city before invoking a weather API.
LLM NodeFeeds extracted parameters to another LLM for further processing—for example, generating a meeting invitation email based on extracted meeting details.
Variable Processing NodeCombines multiple extracted parameters into a single JSON object for unified downstream consumption.
Image Parsing NodeWhen using VL models: first parse image content via the Image Parsing Node, then extract structured information (e.g., product name, color, specs) from the image using the Parameter Extraction Node.
Video Parsing NodeWhen using VL models: first parse video content via the Video Parsing Node, then extract structured information (e.g., scene, characters, actions) from the video using the Parameter Extraction Node.