Embeddings

POST /<lane>/v1/embeddings — the OpenAI Embeddings shape. On /oss//hf this routes to the Hugging Face feature-extraction task under the hood (the HF OpenAI-compatible /v1 surface is chat-only), normalized back to the OpenAI response shape.

Request

curl "https://api.opengateway.one/oss/v1/embeddings" \
  -H "Authorization: Bearer $OPENGATEWAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Qwen/Qwen3-Embedding-0.6B",
    "input": ["hello world", "edge ai gateway"]
  }'

Response

{
  "object": "list",
  "data": [
    { "object": "embedding", "index": 0, "embedding": [0.0123, -0.0456, "..."] },
    { "object": "embedding", "index": 1, "embedding": [0.0789, -0.0011, "..."] }
  ],
  "model": "Qwen/Qwen3-Embedding-0.6B",
  "usage": { "prompt_tokens": 6, "total_tokens": 6 }
}

Dimensions

Models with Matryoshka representation (e.g. Qwen/Qwen3-Embedding-0.6B) accept a dimensions parameter to truncate vectors — smaller indexes, lower latency, little recall loss. A guaranteed-available fallback is thenlper/gte-large.

Capability gating

Requesting a non-embedding model here returns 422 model_task_mismatch. Pick an embedding-capable model (see GET /<lane>/v1/models).