Embedding

Overview

The Embedding node calls an embedding model API and converts a text string into a fixed-dimensional float vector. The resulting vector is attached to the workflow data object under output_field so that downstream nodes — most commonly vector_search or a custom code node — can consume it without making a second API call. Use this node when the same embedding needs to feed multiple branches, or when you want explicit control over which model produces the representation.

Configuration

Field	Type	Required	Default	Description
model_key	string	Yes	—	Identifier of the embedding model to use (e.g. "text-embedding-3-small", "text-embedding-3-large", "text-embedding-ada-002"). Must be registered in the platform's model registry.
input_path	string	No	"$.text"	JSONPath expression pointing to the text field within the incoming data object. Defaults to the top-level text property.
output_field	string	No	"embedding"	Name of the field added to the output object that holds the embedding vector (array of floats).
encoding_format	string	No	"float"	Encoding format for the returned vector. One of "float" (array of 32-bit floats) or "base64" (base64-encoded binary, smaller payload).
dimensions	integer	No	—	Desired output dimensionality. Only supported by models that allow variable dimensions (e.g. text-embedding-3-*). Omit to use the model's default.

Inputs

Port	Type	Description
input	object	Data object containing the text to embed. The field referenced by input_path is extracted and sent to the embedding API.
text	string	Shorthand input: a bare string to embed. When connected, input_path is ignored.

Outputs

Port	Type	Description
output	object	The original input object with output_field added (or overwritten). The embedding value is an array of floats (or a base64 string when encoding_format is "base64").
embedding	array of numbers	The raw embedding vector, emitted separately for easy wiring into vector_search's query_embedding port.
usage	object	Token usage information: { prompt_tokens: number, total_tokens: number }.

Example

{
  "nodeType": "embedding",
  "config": {
    "model_key": "text-embedding-3-small",
    "input_path": "$.query",
    "output_field": "query_embedding",
    "encoding_format": "float",
    "dimensions": 512
  }
}

Notes

• Embedding calls count against your connected AI provider's token quota; monitor usage via the usage output port or the platform's cost dashboard.
• When feeding the result directly into vector_search, wire the embedding output port to vector_search's query_embedding port to avoid a second embedding call.
• The dimensionality of the stored vectors in a collection must match the dimensionality produced by this node; mismatched dimensions cause the search query to fail.
• For batch processing, prefer the loop_map node to embed multiple texts in parallel rather than serialising calls through a single Embedding node.