Docs

Overview

ONYX is a single HTTP endpoint in front of the Claude and GPT model families. You send one request shape regardless of which model answers it — routing happens on the model field, not on the URL or the client library.

• One base URL for every model on the platform.
• Request and response bodies match the OpenAI chat-completions schema.
• Streaming, tool calls, and JSON mode all pass through unchanged.

Authentication

Every request needs a bearer key in the Authorization header. Keys are scoped to your account and can carry an optional monthly spend cap, set when you create them.

Authorization: Bearer onx_live_…

Keys aren't tied to a single model or region — a key that works for Sonnet works for Opus, Haiku, and GPT without changes.

Base URL

https://onyxtech.cloud/v1

All endpoints are relative to this. There is no per-model or per-region subdomain — traffic is routed internally after the request lands.

Chat completions

POST /v1/chat/completions — the only endpoint you need for text generation, across every model on the platform.

$ curl https://onyxtech.cloud/v1/chat/completions \
  -H "Authorization: Bearer $ONYX_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-5",
    "messages": [
      {"role": "user", "content": "Draft the migration plan."}
    ]
  }'

Required fields:

Streaming

Set "stream": true to receive server-sent events instead of a single JSON body. Each event carries one incremental token; the stream ends with data: [DONE].

from openai import OpenAI

client = OpenAI(base_url="https://onyxtech.cloud/v1", api_key=ONYX_KEY)

for chunk in client.chat.completions.create(
    model="claude-opus-4.8",
    messages=[{"role": "user", "content": "Stream this."}],
    stream=True,
):
    print(chunk.choices[0].delta.content or "", end="")

Idle connections send a comment-only heartbeat every 15 seconds so intermediate proxies don't time out the socket mid-generation.

Models

Current per-token pricing for each model lives on the pricing table — it's the same table the billing meter reads from, so it never drifts out of sync with what you're charged.

Tool calls

Pass a tools array using the standard OpenAI function-calling schema. The model returns a tool_calls array instead of plain text when it decides to call one — send the tool's result back as a role: "tool" message to continue the conversation.

{
  "tools": [{
    "type": "function",
    "function": {
      "name": "get_weather",
      "parameters": { "type": "object", "properties": { "city": {"type": "string"} } }
    }
  }]
}

Errors

Errors are JSON with an HTTP status that matches the failure. The error.message field says what happened and, where relevant, what to change before retrying.

Rate limits

Limits are per key, measured in requests per minute, and vary by plan — see Pricing for the current numbers. Every response includes the remaining budget:

X-RateLimit-Limit: 2000
X-RateLimit-Remaining: 1958
X-RateLimit-Reset: 1737490260

Billing & usage

Usage is metered per token, per model, and billed monthly at the rates on the pricing table. Per-key spend caps stop a key from generating further charges once its limit is hit for the period — the key starts returning 402 instead of silently continuing.