OpenAI-compatible clients

The gateway is OpenAI Chat Completions-compatible. Any client that takes an OpenAI base URL + API key will work. The only things to be careful about are:

1. The base URL must include /v1.
2. The model field must be one of the four aliases.
3. Bearer token is the gateway key, not the OpenRouter key.

Python (openai SDK)

pip install openai

import os
from openai import OpenAI

client = OpenAI(
    base_url=os.environ["SMART_GATEWAY_BASE_URL"] + "/v1",
    api_key=os.environ["SMART_GATEWAY_API_KEY"],
)

resp = client.chat.completions.create(
    model="smart-router",
    messages=[{"role": "user", "content": "Reply with the word PONG."}],
    max_tokens=8,
)

print(resp.choices[0].message.content)

Set the environment variables first:

export SMART_GATEWAY_BASE_URL="https://smart-openrounter.bee1x.one"
export SMART_GATEWAY_API_KEY="$YOUR_GATEWAY_KEY"

Node / TypeScript (openai SDK)

npm install openai

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: `${process.env.SMART_GATEWAY_BASE_URL}/v1`,
  apiKey: process.env.SMART_GATEWAY_API_KEY,
});

const completion = await client.chat.completions.create({
  model: "smart-router",
  messages: [{ role: "user", content: "Reply with the word PONG." }],
  max_tokens: 8,
});

console.log(completion.choices[0].message.content);

Streaming in the Python SDK

stream = client.chat.completions.create(
    model="smart-router-flash",
    stream=True,
    messages=[{"role": "user", "content": "Count to 5 slowly."}],
    max_tokens=64,
)

for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)
print()

Tool calling

The gateway forwards tool definitions and tool messages to OpenRouter unchanged. The OpenAI SDK pattern works as-is:

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get current weather",
        "parameters": {
            "type": "object",
            "properties": {"city": {"type": "string"}},
            "required": ["city"],
        },
    },
}]

resp = client.chat.completions.create(
    model="smart-router",
    messages=[{"role": "user", "content": "Weather in Tokyo?"}],
    tools=tools,
    tool_choice="auto",
)

The response follows OpenAI's standard tool_calls shape.

Headers / metadata

To set sensitive, reasoning, session_id, goal_id, task_id, use the metadata field in the request body:

resp = client.chat.completions.create(
    model="smart-router",
    messages=[{"role": "user", "content": "..."}],
    extra_body={
        "metadata": {
            "sensitive": True,
            "reasoning": "high",
            "session_id": "proj-A-2026-08-23",
            "goal_id": "fix-flaky",
            "task_id": "task-17",
        }
    },
)

The OpenAI SDK supports extra_body for forwarding fields that are not part of the typed schema. The OpenAI Node SDK uses metadata directly in the request body.

To send control headers (X-Pro, X-Scout, etc.) you need to drop down to the raw HTTP layer (httpx in Python, fetch in Node) or use a custom HTTP client.

Responses API is not supported

The OpenAI Responses API (/v1/responses) is not implemented in v1.0.0. If you try it, you will get a 404. Use Chat Completions.

LiteLLM proxying

If you want to use the gateway as the upstream for a self-hosted LiteLLM proxy, point LiteLLM at the gateway exactly as you would point it at OpenAI:

model_list:
  - model_name: smart-router
    litellm_params:
      model: openai/smart-router
      api_base: https://smart-openrounter.bee1x.one/v1
      api_key: os.environ/SMART_GATEWAY_API_KEY

Note that the gateway itself already runs on OpenRouter direct (frozen-v3/15-transport-decision.md). Adding LiteLLM in front is optional and most useful if you need multi-provider governance or virtual keys; it does not make the routing decisions.

LangChain / LlamaIndex

Both frameworks have an "OpenAI-compatible" mode. Use the gateway base URL as the openai_api_base (LangChain) or api_base (LlamaIndex), pass the gateway key, and use model="smart-router" (or one of the other aliases).