Embeddings

Convert text into vectors for semantic search, RAG, and content recommendations.

Overview

Embeddings convert text into vectors so applications can retrieve content by semantic similarity. Common uses include semantic search, retrieval-augmented generation (RAG), and related-content recommendations.

Embeddings retrieve candidates from a large collection. Rerank receives a query and candidate documents and refines their order. You can use either capability independently or combine them into a two-stage retrieval pipeline.

API access

Base URL

https://api.agtcloud.cn/v1

Create embeddings

curl 'https://api.agtcloud.cn/v1/embeddings' \
  -H 'Authorization: Bearer your-api-key' \
  -H 'Content-Type: application/json' \
  --data '{
    "model": "MODEL_ID",
    "input": [
      "AGTCloud provides a unified model API.",
      "Embeddings can power semantic retrieval."
    ]
  }'

Parameter	Required	Description
`model`	Yes	Embedding model ID
`input`	Yes	A text value, text array, or token array
`encoding_format`	No	Encoding format for returned vectors
`dimensions`	No	Requested output dimensions; supported by selected models only

Read the result

data[].embedding: generated vector.
data[].index: input index associated with the vector.
usage: token usage for the request.

Keep indexing and queries consistent

Setting	Possible result when inconsistent
Model and version	Old and new vectors use different spaces and cannot be compared reliably
Vector dimensions	Vector-store writes or similarity calculations fail
Text preprocessing	Query and document representations drift apart
Distance metric	Score meaning and ranking change