API
An OpenAI-compatible endpoint for the stealth model. Most SDKs work by swapping the base URL and setting the model to ox-alpha. Mint a key in the Playground and you are live in two lines.
Base URL
https://oxalpha.run/api/v1
Model id
ox-alpha
Auth
Bearer <your key>
Sign in, open the Playground, and create a key under API keys. You will see the secret once.
Quickstart
curl https://oxalpha.run/api/v1/chat/completions \
-H "Authorization: Bearer $OX_ALPHA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "ox-alpha",
"messages": [{"role": "user", "content": "Explain your 1M context in one line."}],
"stream": false
}'import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.OX_ALPHA_API_KEY,
baseURL: "https://oxalpha.run/api/v1",
});
const res = await client.chat.completions.create({
model: "ox-alpha",
messages: [{ role: "user", content: "Explain your 1M context in one line." }],
});
console.log(res.choices[0].message.content);from openai import OpenAI
client = OpenAI(
api_key="OX_ALPHA_API_KEY",
base_url="https://oxalpha.run/api/v1",
)
res = client.chat.completions.create(
model="ox-alpha",
messages=[{"role": "user", "content": "Explain your 1M context in one line."}],
)
print(res.choices[0].message.content)Reference
OpenAI Chat Completions compatible. Send messages and optionally temperature, max_tokens, top_p, and stream. Set stream: true for server-sent events. The model is always ox-alpha; any model value you pass is normalized to it. The response is the standard OpenAI chat.completion shape (or an SSE stream of chat.completion.chunk objects).
| Parameter | Type | Default | Notes |
|---|---|---|---|
| messages | array | required | Chat messages, each { role, content }. role is system, user, or assistant. |
| stream | boolean | false | Stream server-sent events (chat.completion.chunk) instead of one response. |
| temperature | number | 0.7 | Sampling temperature, 0 to 2. Lower is more focused. |
| max_tokens | number | 8192 | Max tokens generated (includes reasoning tokens). Capped at 16,384. |
| top_p | number | 1 | Nucleus sampling, 0.01 to 1. |
| model | string | ox-alpha | Ignored; every request is normalized to ox-alpha. |
Works with the official OpenAI SDKs and anything that speaks their format. Just change the base URL.
Every request is attributed to your key. Manage and revoke keys any time from the Playground.
Generous limits while the model is free. Pricing and availability follow the underlying stealth preview.