Skip to main content

​
Base URL

https://api.caprioletech.com

​
Get started in two steps

​
1. Create an API key

Create your key from the Capriole AI API page after signing in.

​
2. Call POST /v1/chat

Send a simple text request. 
curl https://api.caprioletech.com/v1/chat \
  -X POST \
  -H "Authorization: Bearer YOUR_CAPRIOLE_AI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai-latest",
    "input": "Hello World!"
  }'

import requests

API_KEY = "YOUR_CAPRIOLE_AI_API_KEY"

response = requests.post(
    "https://api.caprioletech.com/v1/chat",
    headers={
        "Authorization": f"Bearer {API_KEY}",
        "Content-Type": "application/json",
    },
    json={
        "model": "openai-latest",
        "input": "Hello World!"
    },
)

response.raise_for_status()
print(response.json())

​
Protocol-compatible endpoints

Use POST /v1/responses for OpenAI Responses clients, POST /v1/chat/completions for OpenAI Chat Completions clients, and POST /v1/messages for Anthropic Messages clients. Use openai-latest, claude-latest, or google-latest when you want Capriole AI to choose our recommended flagship model for that provider. Use a concrete model ID when you need one exact model version. Latest aliases are input shortcuts. Capriole AI records usage against the resolved model, while protocol-compatible response bodies keep the model fields returned by the upstream-compatible endpoint.

​
Quick answers

​
Should the base URL include /v1?

Use https://api.caprioletech.com for direct API calls. Some coding agents ask for a provider base URL; follow the integration page for that agent because OpenAI-compatible and Anthropic-compatible clients expect different base URLs.

​
Which latest aliases work on each endpoint?

POST /v1/chat and POST /v1/chat/completions support all three latest aliases. POST /v1/responses supports openai-latest. POST /v1/messages supports claude-latest plus concrete Claude Messages IDs such as claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, and claude-sonnet-4-6.

​
Why does the response model not show the alias I sent?

That is expected. The alias is only used to pick the model. Responses keep the model naming behavior of the selected endpoint.

​
Why did I get an unsupported model error?

Use a latest alias only on an endpoint that supports it, or switch to a concrete model ID listed by GET /v1/models.

​
Does /v1/messages/count_tokens always work?

It accepts claude-latest and concrete Claude Messages IDs. Token counting depends on the selected upstream supporting Anthropic token counting.