Switch from another API

Most moves are a new base URL, a new key and new model ids. Model ids here are our own: take them from Models or GET /v1/models, never from another provider's list.

From the OpenAI SDK

Change two lines. Everything else (messages, tools, streaming, the Responses API) stays.

from openai import OpenAI

client = OpenAI(
    base_url="https://pinerouter.ai/v1",  # was https://api.openai.com/v1
    api_key="sk-...",                         # a key from this site
)
r = client.chat.completions.create(model="gpt-5-5", messages=[{"role": "user", "content": "Hi"}])

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.example.com/v1", // was https://api.openai.com/v1
  apiKey: "sk-...",                      // a key from this site
});

Or set the environment variables the SDK already reads: OPENAI_BASE_URL=https://pinerouter.ai/v1 and OPENAI_API_KEY=sk-....

From the Anthropic SDK

Point the SDK at our address without /v1. It calls POST /v1/messages, which any text model here can answer.

import anthropic

client = anthropic.Anthropic(base_url="https://pinerouter.ai", api_key="sk-...")
msg = client.messages.create(
    model="gpt-5-2",
    max_tokens=500,
    messages=[{"role": "user", "content": "Hi"}],
)
print(msg.content[0].text)

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({ baseURL: "https://pinerouter.ai", apiKey: "sk-..." });
const msg = await client.messages.create({
  model: "gpt-5-2",
  max_tokens: 500,
  messages: [{ role: "user", content: "Hi" }],
});

Environment variables work too: ANTHROPIC_BASE_URL=https://pinerouter.ai and ANTHROPIC_API_KEY=sk-... (Claude Code reads ANTHROPIC_AUTH_TOKEN; see Claude Code).

• Claude models are not on the Models list today. Pick another text model, such as gpt-5-2, gemini-3-8-flash or kimi-k3.
• Tools, system prompts and streaming are passed to the model in Messages format. Whether a feature such as extended thinking works depends on the model you pick.
• The Message Batches, Files and token counting endpoints are not available.

From OpenRouter

The OpenAI and Anthropic SDK setups above apply unchanged. Streaming works the same way, and usage comes in the last chunk; see Streaming.

From fal, Replicate and other queue-style media APIs

The flow is the same idea: submit, then poll for the result.

curl https://pinerouter.ai/v1/jobs \
  -H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" \
  -d '{"model": "nano-banana-pro", "input": {"prompt": "A ceramic mug on a beige backdrop"}}'

What is different:

• Webhooks or polling. Pass webhook_url to be called when the job finishes (Standard Webhooks signatures), or poll every 5 to 10 seconds; see Jobs. A copy-paste helper is on SDKs.
• Files go in as https:// links. Pass any public link, or upload with POST /v1/files and use the returned url (kept 1 day), in the model's field such as image_url or image_urls.
• Results last 7 days. Download what you want to keep. See Data and logging.
• No cancel. A started job runs to the end. A failed job costs nothing.
• Cost is known up front. Each model's page lists its price per setting. A job reserves its expected cost plus 20% and settles at the real cost.
• Field names differ by model. Do not copy inputs from another provider's schema; read the model's page.

You can also use OpenAI-style POST /v1/images/generations and POST /v1/videos for image and video models if your code already speaks that format.