Anthropic SDK

The Anthropic SDKs speak the Messages format, which the RFS Router serves for every model. Their base URL has no /v1: they add it themselves.

npm i @anthropic-ai/sdk

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

const ai = new Anthropic({
  baseURL: "https://relayfor.si/api",
  apiKey: process.env.RELAYFOR_ROUTER_KEY,
});

const message = await ai.messages.create({
  model: "anthropic/claude-sonnet-5.5",
  max_tokens: 1024,
  system: "You run the $NOVA community. Be brief.",
  messages: [{ role: "user", content: "Welcome the 1,000th holder in one line." }],
});

console.log(message.content);

Model namesCopy Anchor Link

Any id from Models works, such as openai/gpt-6.1-sol or google/gemini-3.8-flash. Claude's own names work too: claude-sonnet-5-5 and dated names like claude-haiku-4-5-20251001 map to the listed model.

StreamingCopy Anchor Link

const stream = ai.messages.stream({
  model: "anthropic/claude-sonnet-5.5",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Draft today's update for holders." }],
});

stream.on("text", (text) => process.stdout.write(text));
const final = await stream.finalMessage();

Extended thinkingCopy Anchor Link

const message = await ai.messages.create({
  model: "anthropic/claude-sonnet-5.5",
  max_tokens: 4096,
  thinking: { type: "enabled", budget_tokens: 2048 },
  messages: [{ role: "user", content: "Which of our three launch ideas is safest, and why?" }],
});

Thinking is billed as output. See Reasoning.

What it costCopy Anchor Link

Messages answers carry no cost: each call's charge is in the account's ledger within a minute, under call: and the answer's x-relayfor-call-id header. A busy model answers 529 overloaded_error, which the SDK retries.