Create text embeddings

AI Gateway provides synchronous text embeddings at POST /v1/embeddings. Gateway returns vectors but does not store them or provide chunking, vector search, retrieval, or RAG orchestration. Your application handles those responsibilities.

Before you start

Before you send a request, confirm all of the following:

1. Your organization has the Embeddings feature enabled.
2. Your effective policy allows the Embeddings operation.
3. Your credential can access an embedding-capable public model.
4. The selected route is operational and priced for embeddings.

The feature and model grant are separate. Enabling the operation does not grant a model, and granting a model does not enable the operation.

Send an embedding request

Use the OpenAI-compatible base URL from Developer Portal → Quickstart. Choose model: "auto" when your policy provides an automatic embedding model, or provide an accessible public embedding model ID:

curl --fail-with-body --show-error --silent \
  -X POST "${CONSCIA_BASE_URL:?Set the OpenAI-compatible /v1 base URL}/embeddings" \
  -H "Authorization: Bearer ${CONSCIA_API_KEY:?Set CONSCIA_API_KEY first}" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "your-public-embedding-model-id",
    "input": ["Text to embed"]
  }'

The v1 provider baseline is Amazon Bedrock Titan Text Embeddings V2. Confirm the available public model and dimensions in the Developer Portal because organization access and route availability are scoped to the credential.

Input can be a string or string array. The current contract returns float vectors and supports dimensions 256, 512, or 1024 when the selected model and route support them. Vector storage and retrieval are outside the Gateway contract.

Usage and errors

Gateway records embedding usage separately from chat or response inference. A request can be denied when the operation is disabled, the model is not allowed, the route is not priced, or an allowance or rate limit is exceeded.

Keep x-request-id and x-correlation-id for diagnosis. See Troubleshoot AI Gateway requests before escalating. Do not send API keys, input text, or vectors to support.