Skip to main content
Embeddings convert text into vectors. Reranking scores candidate documents against a query. Together they are the core building blocks for semantic search, knowledge bases, and retrieval-augmented generation.

​
Embeddings

An embedding is a list of numbers that represents the meaning of text. Text with similar meaning tends to have similar vectors, which lets you search by meaning instead of exact keywords. Use /v1/embeddings to create vectors: 
{
  "model": "text-embedding-3-small",
  "input": "Moxus AI is a unified LLM API gateway."
}
ParameterDescription
modelEmbedding model name
inputA string or an array of strings
encoding_formatOptional output format, usually float or base64
Example response: 
{
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "index": 0,
      "embedding": [0.023, -0.015, 0.042, -0.008]
    }
  ],
  "model": "text-embedding-3-small",
  "usage": {
    "prompt_tokens": 12,
    "total_tokens": 12
  }
}

​
Python example

from openai import OpenAI

client = OpenAI(api_key="sk-your-key", base_url="https://moxus.ai/v1")

response = client.embeddings.create(
    model="text-embedding-3-small",
    input="Moxus AI is a unified LLM API gateway.",
)

vector = response.data[0].embedding
print(f"Vector dimensions: {len(vector)}")
print(vector[:5])

​
Batch embeddings

texts = [
    "Moxus AI supports multiple providers.",
    "OpenAI provides GPT models.",
    "Claude is developed by Anthropic.",
]

response = client.embeddings.create(
    model="text-embedding-3-small",
    input=texts,
)

for index, item in enumerate(response.data):
    print(index, len(item.embedding))
After you create vectors, store them in a vector database such as Milvus, Qdrant, Chroma, Pinecone, or pgvector. At query time, embed the user question and search for nearest vectors. 
import numpy as np

def cosine_similarity(a, b):
    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))

​
Reranking

Reranking takes a query and candidate documents, then returns a relevance score for each candidate. It is commonly used after vector search to improve result quality. 
{
  "model": "rerank-model",
  "query": "How does Moxus AI billing work?",
  "documents": [
    "Moxus AI bills by token usage.",
    "Moxus AI supports multiple model providers.",
    "Python is a programming language."
  ],
  "top_n": 2
}
Example response: 
{
  "results": [
    {
      "index": 0,
      "relevance_score": 0.92,
      "document": "Moxus AI bills by token usage."
    },
    {
      "index": 1,
      "relevance_score": 0.65,
      "document": "Moxus AI supports multiple model providers."
    }
  ]
}
The exact endpoint and response shape can vary by provider. Use the model detail page as the source of truth for supported parameters.

​
Typical RAG flow

  1. Split source documents into chunks.
  2. Create embeddings for each chunk.
  3. Store vectors and metadata in a vector database.
  4. Embed the user query.
  5. Retrieve the most similar chunks.
  6. Optionally rerank the retrieved chunks.
  7. Send the best context to a chat model for the final answer.

​
Practical guidance

  • Choose an embedding model that matches your language and domain.
  • Keep chunk sizes consistent and include useful metadata.
  • Rerank only the top candidates to control cost and latency.
  • Monitor token usage in dashboard and usage.

​
Next steps