Embeddings & Reranking API

Embeddings

POST /v1/embeddings

Converts text into numerical vectors so semantic similarity becomes measurable. A typical use case is populating and searching a vector database.

Parameters

Parameter	Type	Description
model	string	Model ID, e.g. vllm/release/harrier-oss-v1-0.6b
input	string/array	Text or list of texts

curl

curl -X POST https://ai.noris.de/v1/embeddings \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "vllm/release/harrier-oss-v1-0.6b",
    "input": "AI infrastructure from Germany"
  }'

Response

The response contains an embedding array with the vector for each input element:

{
  "object": "list",
  "data": [
    { "object": "embedding", "index": 0, "embedding": [0.0123, -0.0456, "..."] }
  ],
  "model": "vllm/release/harrier-oss-v1-0.6b",
  "usage": { "prompt_tokens": 5, "total_tokens": 5 }
}

Use Case

Store the vectors alongside metadata in your vector database. At search time, the query is embedded as well and compared against the stored vectors via similarity search (e.g. cosine).

Reranking

POST /v1/rerank

A reranker scores query and documents pairwise, producing a more precise ranking than embedding search alone. A typical use case is the second stage of a RAG pipeline.

Parameters

Parameter	Type	Description
model	string	Model ID, e.g. vllm/release/bge-reranker-v2-m3
query	string	Search query
documents	array	List of candidate documents
top_n	integer	Number of top results to return
return_documents	boolean	Include document text in the response

curl

curl -X POST https://ai.noris.de/v1/rerank \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "vllm/release/bge-reranker-v2-m3",
    "query": "How does zero-data-retention work?",
    "documents": [
      "noris does not persist prompts.",
      "The weather in Nuremberg is sunny."
    ],
    "top_n": 2,
    "return_documents": true
  }'

Response

The response contains a results array, sorted by relevance:

{
  "id": "rerank-...",
  "results": [
    {
      "index": 0,
      "relevance_score": 0.98,
      "document": {"text": "noris does not persist prompts."}
    },
    {
      "index": 1,
      "relevance_score": 0.03,
      "document": {"text": "The weather in Nuremberg is sunny."}
    }
  ],
  "model": "vllm/release/bge-reranker-v2-m3"
}

Use Case

In a RAG pipeline, vector search returns a rough candidate set; the reranker refines this ordering before the top documents are passed to the LLM.