Choose a client

Every client needs the same two things: the gateway's base URL and the key in its API_KEY. Nothing is installed into the client.

Client	API	Base URL	Guide
Codex CLI	Responses	http://localhost:8080/api/v1	Codex CLI
OpenCode	Chat Completions	http://localhost:8080/api/v1	Editors and agents
Zed	Chat Completions, Completions	http://localhost:8080/api/v1	Editors and agents
OpenAI SDKs	Chat Completions, Responses	http://localhost:8080/api/v1	below
curl and scripts	Any route	http://localhost:8080/api/v1	API reference

Each client adds the route, such as /chat/completions, to the base URL, so the base URL ends in /api/v1, the default API_ROUTE_PREFIX. Replace localhost:8080 with the address at which your clients reach the gateway.

The key

Clients send the key as Authorization: Bearer <key>, which is what the OpenAI SDKs and most other clients send when they are given an API key. A request without the key, or with another one, gets HTTP 401 with the error code unauthorized. The gateway has one key, and every client that holds it has the same access, so treat it like a password and change it when someone should lose access.

Model names

GET /api/v1/models lists the model IDs your account can call in the gateway's region, and the gateway also accepts the short names in Models, such as gpt-5.5. A client sends the ID as it is listed; the gateway passes it to Bedrock unchanged.

Some clients know the models of their own vendor only and warn about, or refuse, a name they do not know. Their pages below show where to declare the model.

SDKs

The OpenAI SDK for Python, through Responses:

import os

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8080/api/v1", api_key=os.environ["API_KEY"])
response = client.responses.create(
    model="us.amazon.nova-2-lite-v1:0",
    input="Reply with exactly: BEDROCK_OK",
)
print(response.output_text)

The OpenAI SDK for JavaScript, through Chat Completions:

import OpenAI from "openai";

const openai = new OpenAI({
  baseURL: "http://localhost:8080/api/v1",
  apiKey: process.env.API_KEY,
});

const completion = await openai.chat.completions.create({
  model: "us.amazon.nova-2-lite-v1:0",
  messages: [{ role: "user", content: "Reply with exactly: BEDROCK_OK" }],
});
console.log(completion.choices[0].message.content);

Both print BEDROCK_OK. Options that exist only on Bedrock, such as the prompt cache controls, go in the SDK's extra_body option.

Other clients

Any client that speaks OpenAI Chat Completions or Responses can use the same base URL. Prefer Responses when the client offers both and the conversation uses tools with a reasoning model: Responses carries the model's reasoning from one turn to the next on its own, while Chat Completions needs the signing key for it. The protocol compatibility reference lists which request fields and tool types each route handles.