Migrate existing model calls

Move an application that calls OpenAI or Anthropic directly onto Model Gateway. You change three things: the base URL, the API key, and the model name. Your messages, tools, and streaming code stay as they are.

Change	From	To
Base URL, OpenAI SDKs	The SDK default	`https://api.atlan.com/openai/v1`
Base URL, Anthropic SDK	The SDK default	`https://api.atlan.com/anthropic` (the SDK adds `/v1/messages`)
API key	Your provider key	Your Atlan agent key, `ATLAN_API_KEY`
Model name	The vendor’s model id	`<provider>/<model>`, copied from `GET /openai/v1/models`

Either dialect reaches every chat model your organization connects, so you don’t have to match the SDK to the provider.

Before you begin

You have an Atlan agent key in ATLAN_API_KEY. See Connect client to Model Gateway to create one.
You have the id of each model you’ll call, from GET /openai/v1/models.
Every operation your code calls is served. Files, batches, assistants, vector stores, fine-tuning, and Anthropic token counting aren’t; see Model API compatibility.

Change the client configuration

Pick your library and change the client setup.

OpenAI (Python)
OpenAI (TypeScript)
Anthropic (Python)
LangChain
Vercel AI SDK

Before

import os

from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
reply = client.chat.completions.create(model="<model>", messages=messages)

After

import os

from openai import OpenAI

client = OpenAI(
    base_url="https://api.atlan.com/openai/v1",
    api_key=os.environ["ATLAN_API_KEY"],
)
reply = client.chat.completions.create(model="<provider>/<model>", messages=messages)

Before

import OpenAI from "openai";

const client = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });
const reply = await client.chat.completions.create({ model: "<model>", messages });

After

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.atlan.com/openai/v1",
  apiKey: process.env.ATLAN_API_KEY,
});
const reply = await client.chat.completions.create({ model: "<provider>/<model>", messages });

The Anthropic SDK sends its api_key in the x-api-key header, which the gateway accepts on the Anthropic dialect.

Before

import os

from anthropic import Anthropic

client = Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
reply = client.messages.create(model="<model>", max_tokens=1024, messages=messages)

After

import os

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.atlan.com/anthropic",
    api_key=os.environ["ATLAN_API_KEY"],
)
reply = client.messages.create(model="<provider>/<model>", max_tokens=1024, messages=messages)

ChatOpenAI speaks the OpenAI dialect, so it reaches any chat model through the gateway.

Before

import os

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="<model>", api_key=os.environ["OPENAI_API_KEY"])

After

import os

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="<provider>/<model>",
    base_url="https://api.atlan.com/openai/v1",
    api_key=os.environ["ATLAN_API_KEY"],
)

Create an OpenAI provider pointed at the gateway, and call the chat completions API with .chat().

Before

import { openai } from "@ai-sdk/openai";
import { generateText } from "ai";

const { text } = await generateText({ model: openai("<model>"), prompt });

After

import { createOpenAI } from "@ai-sdk/openai";
import { generateText } from "ai";

const atlan = createOpenAI({
  baseURL: "https://api.atlan.com/openai/v1",
  apiKey: process.env.ATLAN_API_KEY,
});
const { text } = await generateText({ model: atlan.chat("<provider>/<model>"), prompt });

Replace each model name

Swap every model id for the gateway’s qualified name. A bare vendor id, without the provider part, returns 404 with model not found.

Check the fields you send

Fields that only tune or annotate a request, such as metadata or seed, are dropped when the serving provider has no place for them, and named in the x-gateway-model-dropped response header. Fields that change the reply, such as logprobs, are refused with 400 when the provider can’t honor them. If you use the Responses API, send the whole conversation each time: previous_response_id and stored conversations are refused.

Verify

Run your application. It answers as before.
Inspect one response’s headers: x-gateway-model-wire names the provider API that served it.
In Model Gateway studio, an organization admin finds the agent’s calls on Agents and in Usage. See Review model usage.

If a call fails, look up its status in Model Gateway errors.

Connect client to Model Gateway: create an agent key and set up coding clients such as Claude Code and Codex.
How Model Gateway works: what the gateway does with each request.
Model API compatibility: served endpoints and field handling.

Skill management

Agent management

AI Gateway

Metrics Studio

Ontology

Before you begin

Verify

​Before you begin

​Verify

​Related

Before you begin

Verify

Related