AWS Bedrock Embeddings

POST /v1/aws-bedrock/embeddings

Native passthrough to AWS Bedrock embedding models. Request and response shapes match Bedrock’s InvokeModel API byte-for-byte, with a usage block layered on top for credit tracking.

Headers

X-Quantized-Provider is not used for this endpoint — the provider is implicit (Bedrock).

Request body

The body shape varies by model. The router discriminates on the model field’s prefix:

• amazon.titan-* → Titan schema
• cohere.* → Cohere schema

Any other prefix is rejected with 422 before the catalog gate runs.

Titan (amazon.titan-embed-text-v2:0)

Cohere (cohere.embed-english-v3, cohere.embed-multilingual-v3)

Examples

curl -X POST https://api.quantized.us/v1/aws-bedrock/embeddings \
  -H "Authorization: Bearer sk-quantized-YOUR-KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "amazon.titan-embed-text-v2:0",
    "inputText": "The quick brown fox jumps over the lazy dog.",
    "dimensions": 512,
    "normalize": true,
    "embeddingTypes": ["float"]
  }'

curl -X POST https://api.quantized.us/v1/aws-bedrock/embeddings \
  -H "Authorization: Bearer sk-quantized-YOUR-KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cohere.embed-english-v3",
    "texts": ["first doc", "second doc", "third doc"],
    "input_type": "classification",
    "embedding_types": ["float"]
  }'

import httpx

# Titan
resp = httpx.post(
    "https://api.quantized.us/v1/aws-bedrock/embeddings",
    headers={"Authorization": "Bearer sk-quantized-YOUR-KEY"},
    json={
        "model": "amazon.titan-embed-text-v2:0",
        "inputText": "Hello world",
        "dimensions": 256,
    },
)
data = resp.json()
print(len(data["embedding"]))  # 256

# Cohere — multilingual, batch
resp = httpx.post(
    "https://api.quantized.us/v1/aws-bedrock/embeddings",
    headers={"Authorization": "Bearer sk-quantized-YOUR-KEY"},
    json={
        "model": "cohere.embed-multilingual-v3",
        "texts": ["Hello", "Hola", "Bonjour"],
        "input_type": "search_document",
    },
)
data = resp.json()
print(len(data["embeddings"]["float"]))  # 3

Response

Titan response

{
  "embedding": [0.0123, -0.0456, 0.0789, ...],
  "embeddingsByType": {
    "float": [0.0123, -0.0456, 0.0789, ...]
  },
  "inputTextTokenCount": 11,
  "usage": {
    "credits_used": 3,
    "credits_remaining": 999997
  }
}

Cohere response

Cohere v3 returns two distinct response shapes depending on whether the request included embedding_types. The router preserves both byte-for-byte.

With embedding_types (by-type dict):

{
  "id": "abc-123",
  "texts": ["first doc", "second doc"],
  "embeddings": {
    "float": [[0.0123, ...], [0.0456, ...]]
  },
  "response_type": "embeddings_by_type",
  "usage": {
    "credits_used": 2,
    "credits_remaining": 999998
  }
}

Without embedding_types (legacy list):

{
  "id": "abc-123",
  "texts": ["first doc"],
  "embeddings": [
    [0.0123, ...]
  ],
  "response_type": "embeddings_floats",
  "usage": {
    "credits_used": 1,
    "credits_remaining": 999999
  }
}

Models

All three are seeded with supported_features: ["bedrock_embeddings"] in GET /v1/models. Filter for Bedrock embedding models:

bedrock_embed_models = [
    m for m in models
    if "bedrock_embeddings" in m.get("supported_features", [])
]

Providers

This endpoint is Bedrock-only — X-Quantized-Provider is ignored. If you want to route OpenAI embedding models through Bedrock-style infrastructure, that’s not supported; use /v1/embeddings instead.

Errors