Quickstart

The KURE API serves Korean retrieval models. Embeddings (KURE-v3) are OpenAI-compatible: point the OpenAI SDK at our base URL with your key and you are ready to go. The same key also works for sparse embeddings (KURE-Sparse) and reranking (KURE-Reranker). This page starts with embeddings.

1. Get an API key

1. Click Sign in at the top right and sign in with GitHub or Google.
2. Accept the terms of service to sign up. Allowing us to store your inputs is optional; if you do, per-minute limits double and the daily token limit goes up 10× (Data policy).
3. Create a key in Dashboard → API keys. The full key is shown only once, so store it somewhere safe right away.

Pass the key through an environment variable rather than hard-coding it.

export KURE_API_KEY="kure-..."

2. Your first request

The base URL is https://kure.kullm.ai/api/v1.

# pip install openai
import os
 
from openai import OpenAI
 
client = OpenAI(base_url="https://kure.kullm.ai/api/v1", api_key=os.environ["KURE_API_KEY"])
 
res = client.embeddings.create(
    model="kure-v3",
    input="Where is Korea University?",
    extra_body={"input_type": "query"},
)
print(len(res.data[0].embedding))  # 2048

3. Tell queries and documents apart

KURE-v3 is trained to encode search queries and the documents being searched differently. Set input_type to "query" for queries and "document" for documents. Without it, the text is embedded as-is and retrieval quality may drop. See Models for details.

The OpenAI SDK has no input_type argument, so in Python pass it through extra_body.

4. A minimal search

Vectors are L2-normalized, so the dot product equals cosine similarity.

import os
 
import numpy as np
from openai import OpenAI
 
client = OpenAI(base_url="https://kure.kullm.ai/api/v1", api_key=os.environ["KURE_API_KEY"])
 
 
def embed(texts, input_type):
    res = client.embeddings.create(
        model="kure-v3",
        input=texts,
        dimensions=1024,
        extra_body={"input_type": input_type},
    )
    return np.array([d.embedding for d in res.data])
 
 
docs = [
    "고려대학교는 서울특별시 성북구 안암로 145에 있다.",
    "연세대학교는 서울특별시 서대문구 연세로 50에 있다.",
    "고려대학교 세종캠퍼스는 세종특별자치시 조치원읍에 있다.",
]
doc_vecs = embed(docs, "document")
query_vec = embed(["What is the address of Korea University's main campus?"], "query")[0]
 
scores = doc_vecs @ query_vec
for i in np.argsort(-scores):
    print(f"{scores[i]:.3f}  {docs[i]}")

Embed your documents once and keep them in a vector database; at search time, only the query needs embedding. You can also try it without signing in on the Playground.

Next steps

• Embeddings API: every parameter and the response format
• Sparse Embeddings API: sparse vectors, vector databases, hybrid search
• Rerank API: reorder first-stage results
• Models: choosing dimensions (MRL) and the maximum input length
• Limits & errors: rate limits and how to retry

For questions, bug reports, or higher limits, email kure.nlpai@gmail.com.