Last updated 2026-05-17

Package llmrouter — Embeddings

Every public symbol declared in embeddings.go. One interface — Embedder — plus the request and response types it speaks.

Import path: 

import "github.com/elloloop/llmrouter"

Embedder

The embedding contract. Implementations are concurrency-safe. Source: embeddings.go#L11. 

type Embedder interface {
    Embed(ctx context.Context, req EmbedRequest) (*EmbedResponse, error)
}

Embed is a single-round-trip call — there is no streaming, no Stream handle. A non-nil error is returned on synchronous failures (bad config, non-2xx HTTP status, malformed response body).

EmbedRequest

The polymorphic embedding-request shape. Source: embeddings.go#L17. 

type EmbedRequest struct {
    Model          string          // model id; e.g. "text-embedding-3-small"
    Inputs         []string        // texts to embed; index-aligned with response
    Dimensions     int             // optional lower-dim output (OpenAI v3, Voyage, ...)
    TaskType       string          // task hint, normalised cross-vendor
    EncodingFormat string          // "float" (default) or "base64"
    User           string          // OpenAI-style end-user id
    Raw            json.RawMessage // byte passthrough; Model is overlaid
}
Model
The model identifier. The literal string is sent to the upstream; providers don't alias.
Inputs
The list of strings to embed. EmbedResponse.Embeddings is index-aligned with this slice.
Dimensions
Requests a lower-dimensional output. Supported by OpenAI text-embedding-3-*, Voyage, and a few others. Zero means "use the model default."
TaskType
Canonical task hint (RETRIEVAL_QUERY / RETRIEVAL_DOCUMENT / SEMANTIC_SIMILARITY / CLASSIFICATION / CLUSTERING / QUESTION_ANSWERING / FACT_VERIFICATION). Providers translate to their native vocabulary (Cohere input_type, Voyage input_type). See cross-vendor task-type mapping.
EncodingFormat
"float" (default) or "base64". Only affects wire format; the response always contains [][]float32.
User
OpenAI-style end-user identifier for telemetry and abuse tracking. Forwarded by OpenAI / Azure; ignored elsewhere.
Raw
Byte passthrough escape hatch. The library overlays Model; everything else (Cohere's embedding_types, Voyage's truncation, Titan's normalize) survives.

EmbedResponse

The embedding-response shape. Source: embeddings.go#L51. 

type EmbedResponse struct {
    Model      string          // resolved model id echoed by the provider
    Embeddings [][]float32     // index-aligned with EmbedRequest.Inputs
    Usage      *Usage          // token counts; may be nil
    Raw        json.RawMessage // original wire-format JSON
}
Model
The model id the upstream actually used. May include a version suffix (text-embedding-3-small-2024).
Embeddings
One vector per input. len(Embeddings) == len(req.Inputs) always — the library reorders if the wire response is out-of-order tagged (Cohere v1 was, v3 is not).
Usage
Token usage. Embedding endpoints populate PromptTokens (sometimes also TotalTokens). May be nil.
Raw
Original wire bytes for callers that want to forward the response without re-marshaling.

See also