Embedding

The embedding API allows you to get a vector representation of the input to be used from machine learning models or algorithms, leveraging models like gte-Qwen2.

API Call Parameters

• input: A string (or list of strings) to embed, such as "A white cat resting in Rome."
• model: The identifier for the embedding model, e.g., gte-Qwen2 or Qwen3-Embedding-8B.
• dimensions: The length of the vector to return, e.g. 1024. Regolo passes this straight to the model with no server-side truncation or renormalization. Whether a reduced dimension is supported — and how MRL truncation is applied — is up to the model.

Using Regolo ClientPythonCURL

import regolo

regolo.default_key = "<YOUR_REGOLO_KEY>"
regolo.default_embedder_model = "gte-Qwen2"


embeddings = regolo.static_embeddings(input_text=["A white cat resting in Rome", "A white cat resting in Paris"])

print(embeddings)

import requests
import json

url = 'https://api.regolo.ai/v1/embeddings'
headers = {
    'Authorization': 'Bearer YOUR_REGOLO_KEY',
    'Content-Type': 'application/json'
}

data = {
    "input": "A white cat resting in Rome",
    "model": "gte-Qwen2",
}

response = requests.post(url, headers=headers, data=json.dumps(data))

if response.status_code == 200:
    with open("./embedding.json", 'w') as _file:
        json.dump(response.json(), _file)
else:
    print("Failed embedding request:", response.status_code, response.text)

curl -X POST https://api.regolo.ai/v1/embeddings
-H "Content-Type: application/json"
-H "Authorization: Bearer YOUR_REGOLO_KEY"
-d '{
    "model": "gte-Qwen2",
    "input": "The quick brown fox jumps over the lazy dog"
}'

Reduced dimensions

Some embedding models can return shorter vectors directly, so you save storage and speed up similarity search without truncating on the client. Pass dimensions and the model does the work.

dimensions is forwarded to the model as-is. Regolo does not truncate or renormalize the vector — if the model applies Matryoshka Representation Learning (MRL) and renormalizes, that happens inside the model. If the model does not support a reduced dimension, it ignores the value or returns an error.

PythonCURL

import requests

url = 'https://api.regolo.ai/v1/embeddings'
headers = {
    'Authorization': 'Bearer YOUR_REGOLO_KEY',
    'Content-Type': 'application/json'
}

data = {
    "model": "Qwen3-Embedding-8B",
    "input": "A white cat resting in Rome",
    "dimensions": 1024
}

response = requests.post(url, headers=headers, json=data)
result = response.json()
print(len(result["data"][0]["embedding"]))  # 1024

curl -X POST https://api.regolo.ai/v1/embeddings
-H "Content-Type: application/json"
-H "Authorization: Bearer YOUR_REGOLO_KEY"
-d '{
    "model": "Qwen3-Embedding-8B",
    "input": "A white cat resting in Rome",
    "dimensions": 1024
}'

Support is model-dependent

Not every embedding model accepts dimensions. Qwen3-Embedding-8B does; gte-Qwen2 returns its fixed native dimension. Check the model in the catalog before relying on a specific length.

For the exhaustive API's endpoints documentation visit docs.api.regolo.ai.