Aller au contenu principal

Embed

Category: RAG
Node id: embed

Generates text embeddings/vectors using configurable providers (OpenAI, Hugging Face, Cohere, local)

What it does​

  • Multi-provider text embedding (OpenAI, Hugging Face, Cohere, local)
  • Configurable embedding dimensions (1-4096)
  • Unit-length vector normalization
  • Deterministic embedding generation for testing
  • Text length tracking in output metadata

Use cases​

  • Generating embeddings for RAG vector store ingestion
  • Converting clinical text to vectors for similarity search
  • Building semantic search indices for document collections

Runtime​

  • Performance tier: fast

Input ports​

NameTypeRequiredDescription
texttextYesText to embed

Output ports​

NameTypeRequiredDescription
resultjsonNoMain result containing embedding vector, dimensions, text_length, provider, and model
successbooleanNoWhether the operation succeeded

Settings​

Basic​

SettingNameTypeRequiredDefaultAllowed valuesDescription
ProviderproviderenumNoopenaiopenai, huggingface, cohere, localEmbedding provider to use
ModelmodelstringNotext-embedding-ada-002Embedding model name
DimensionsdimensionsnumberNo1536min 1; max 4096Embedding vector dimensions

Auth​

SettingNameTypeRequiredDefaultAllowed valuesDescription
API Keyapi_keysecretNoAPI key for cloud providers (not required for local)

Errors​

CodeDescription
TEXT_REQUIREDNo text input was provided for embedding
INVALID_PROVIDERProvider must be openai, huggingface, cohere, or local
API_KEY_REQUIREDAPI key is required for cloud providers

Runtime notes​

  • result.embedding is a float32 array of length 'dimensions' — wire to a vector-store node for indexing
  • Vectors are L2-normalized to unit length
  • The 'dimensions' config must match your vector store index dimensions exactly
  • Currently uses deterministic embeddings for testing — production requires a valid api_key for cloud providers

See Build any workflow for how to put nodes together and Ports, types and data flow for how port types connect.

Quelque chose n'est pas clair ? Envoyez vos commentaires sur la documentation par courriel.