Overview The proxy embeddings endpoint provides OpenAI-compatible embedding generation with authentication, rate limiting, and cost tracking.
Endpoint POST {PROXY_BASE_URL}/v1/embeddings
Alternate routes:
POST /embeddingsPOST /engines/{model}/embeddingsPOST /openai/deployments/{model}/embeddings
Authentication Bearer token for authentication. Authorization: Bearer sk-litellm-xxx...
Content-Type
string
default: "application/json"
Content type of the request body.
End-user ID for tracking.
JSON stringified metadata.
Request Body Embedding model to use. Examples: text-embedding-3-small, text-embedding-ada-002
Text to embed. Can be a single string or array of strings. { "input" : "The quick brown fox" }
Or: { "input" : [ "First text" , "Second text" , "Third text" ] }
Format of the embeddings. Options: "float", "base64"
Number of dimensions for the embedding (text-embedding-3 models only).
Unique identifier for end-user.
Response Success Response (200) Object type, always “list”.
Array of embedding objects. Object type, always “embedding”.
The embedding vector as array of floats.
Index of the embedding in the list.
Model used for embeddings.
Token usage information. Number of tokens in the input.
Examples Basic Request curl -X POST http://localhost:4000/v1/embeddings \ -H "Content-Type: application/json" \ -H "Authorization: Bearer sk-litellm-xxx" \ -d '{ "model": "text-embedding-3-small", "input": "The quick brown fox jumps over the lazy dog" }'
Python Request import openai client = openai.OpenAI( api_key = "sk-litellm-xxx" , base_url = "http://localhost:4000" ) response = client.embeddings.create( model = "text-embedding-3-small" , input = "The quick brown fox" ) print (response.data[ 0 ].embedding)
Batch Embeddings import openai client = openai.OpenAI( api_key = "sk-litellm-xxx" , base_url = "http://localhost:4000" ) texts = [ "First document" , "Second document" , "Third document" ] response = client.embeddings.create( model = "text-embedding-3-small" , input = texts ) for i, embedding in enumerate (response.data): print ( f "Document { i } : { len (embedding.embedding) } dimensions" )
With User Tracking import openai client = openai.OpenAI( api_key = "sk-litellm-xxx" , base_url = "http://localhost:4000" ) response = client.embeddings.create( model = "text-embedding-3-small" , input = "Sample text" , extra_headers = { "x-litellm-user-id" : "user-123" } )
Custom Dimensions import openai client = openai.OpenAI( api_key = "sk-litellm-xxx" , base_url = "http://localhost:4000" ) response = client.embeddings.create( model = "text-embedding-3-small" , input = "Sample text" , dimensions = 512 # Reduce from default 1536 ) print ( len (response.data[ 0 ].embedding)) # 512
Error Responses 401 Unauthorized { "error" : { "message" : "Invalid API key" , "type" : "invalid_request_error" , "code" : "invalid_api_key" } }
429 Too Many Requests { "error" : { "message" : "Rate limit exceeded for key" , "type" : "rate_limit_error" } }
400 Bad Request { "error" : { "message" : "Invalid input format" , "type" : "invalid_request_error" } }
Proxy Features Cost Tracking The proxy automatically tracks embedding costs:
Per-key spending
Per-team spending
Per-user spending
Rate Limiting Keys can have TPM limits for embeddings:
Requests throttled automatically
429 error when limit exceeded
Model Routing Proxy can route to different embedding providers:
OpenAI
Azure OpenAI
Cohere
Bedrock
Vertex AI
And more!