Migrate from OpenAI, Anthropic or another API

OceanToken speaks the OpenAI API, and every chat model also speaks the Anthropic Messages API. Moving an app is two settings, the base URL and the key, plus the model names.

From the OpenAI SDK

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.oceantoken.ai/v1",  # was https://api.openai.com/v1
    api_key=os.environ["OCEANTOKEN_API_KEY"],  # an OceanToken key, sk-...
)

In TypeScript the options are baseURL and apiKey. Then add the publisher in front of each model name: gpt-5-mini becomes openai/gpt-5-mini. The Models page lists every ID.

No code change at all: the OpenAI SDKs for Python and Node.js read two environment variables, so an app that creates its client without arguments moves with

export OPENAI_BASE_URL=https://api.oceantoken.ai/v1
export OPENAI_API_KEY=$OCEANTOKEN_API_KEY

and the new model names. Many tools built on the OpenAI SDK read the same variables.

From the Anthropic SDK

import os
import anthropic

client = anthropic.Anthropic(
    base_url="https://api.oceantoken.ai",  # no /v1: the SDK adds it
    api_key=os.environ["OCEANTOKEN_API_KEY"],
)

Or set ANTHROPIC_BASE_URL=https://api.oceantoken.ai and ANTHROPIC_API_KEY to your OceanToken key. Claude models are named anthropic/claude-sonnet-4.6, anthropic/claude-haiku-4.5 and so on, and the same code calls any other chat model on OceanToken by its ID.

From another multi-model API

If you use another service that offers many models behind one OpenAI-compatible API:

• Base URL and key: https://api.oceantoken.ai/v1 and your OceanToken key.
• Model IDs have the same publisher/model form, so most carry over unchanged. Check each against GET /v1/models or the Models page; a name that does not exist is refused with 400 model_not_found, never swapped for another model.
• Backup models: put them in fallbacks, a list of model IDs tried in order if your model fails (Retries and failover). Other ways of listing backup models are not supported.
• Routing preferences that pick a hosting provider, and attribution headers, are not needed and not supported. Each model is served from official or authorized sources, failover between them is automatic, and the price does not depend on which one answers (Retries and failover).
• :free models exist here too.
• Video uses POST /v1/videos and a job you poll (Video).

What works the same

• Request and response bodies of each shape, including streaming, tool calling, image input and structured output.
• The error envelope, {"error": {"message", "type", "param", "code"}}.
• Every OpenAI-compatible SDK and framework: set the base URL and key. Guides: Vercel AI SDK, coding agents.

What is different

Topic	On OceanToken
Parameters a model does not support	Dropped, and the request runs. A success does not prove a parameter took effect (Which API shape).
Errors	A few statuses differ: an empty balance is 400 insufficient_credits, not a 429; an unknown model is 400. Branch on code (Migrating from OpenAI).
Not offered	Batches, fine-tuning, and listing or reading uploaded files answer 501 not_supported (Errors). Send each request to its regular endpoint.
Billing	Prepaid: requests run while your balance is above $0. Each key can have its own credit limit (Create an API key).
What a request cost	The x-oceantoken-cost header, GET /v1/usage/{request_id} for streams, and Logs in the console (What a request cost).
Generated media	Images and video come back as links valid for 24 hours; download what you keep (Limits → Retention).
Request ID	Log the x-oceantoken-request-id header and quote it when you write to us (Errors).

Checklist

1. Create a key for the app, with a credit limit, and add credits.
2. Change the base URL and key, or set the environment variables above.
3. Rename the models; check each with GET /v1/models.
4. If you branch on error statuses, read Errors; retry only what is retryable.
5. Run your tests, including one with each parameter you depend on, and check the requests under Logs in the console.