Migrate from plain AI SDK

If you already call the Vercel AI SDK directly, routing through ai-lcr is a one-line change. Your business logic doesn't move.

If your code already calls the Vercel AI SDK, you don't rewrite anything. You move the model definition into an ai-lcr config once, then swap the model argument. Every generateText / streamText / generateObject call stays exactly as it is.

Before — single provider, price hard-wired, no fallback

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

const { text } = await generateText({
  model: openai("gpt-4o-mini"), // one provider; if it 429s or dies, so do you
  prompt,
});

After — cheapest-first routing, automatic fallback, zero logic changes

import { createLCR } from "ai-lcr";
import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
import { openai } from "@ai-sdk/openai";
import { generateText } from "ai";

// 1. Define providers + a cheapest-first list — once, at startup.
const openrouter = createOpenAICompatible({
  name: "openrouter",
  baseURL: "https://openrouter.ai/api/v1",
  apiKey: process.env.OPENROUTER_API_KEY,
});

const lcr = createLCR({
  autoSort: true,
  models: {
    smart: [
      { model: openrouter("openai/gpt-4o-mini"), label: "openrouter", cost: { input: 0.15, output: 0.6 } },
      { model: openai("gpt-4o-mini"), label: "openai-direct", cost: { input: 0.15, output: 0.6 } },
    ],
  },
  onCost: ({ provider, costUsd }) => console.log(`${provider}: $${costUsd.toFixed(6)}`),
});

// 2. Swap the model argument. Nothing else moves.
const { text } = await generateText({
  model: lcr("smart"), // ← the only change in your call sites
  prompt,
});

The diff at every call site is literally openai("gpt-4o-mini") → lcr("smart"). Streaming, tools, structured output, and agents all keep working because lcr("smart") is an AI SDK model.

What you get for that one line

Fallback — if the cheapest provider errors, rate-limits, or times out, the call retries on the next provider in the list instead of failing.
Cost tracking — onCost fires per call with the real cost on whichever provider actually served it.
One place to change prices/providers — add or reorder providers in the config; call sites never change again.

Migration checklist

Move each provider("model") you call today into an ai-lcr models entry.
Add at least one second provider for the same logical model (that's the point).
Replace provider("model") with lcr("yourKey") at the call sites.
Wire onCost to your logging so you can see the savings.

Already on Vercel AI Gateway or OpenRouter?

You don't have to remove them — ai-lcr routes across providers, including a gateway as one entry in the list:

models: {
  smart: [
    { model: kunavo("gemini-3-flash"), label: "kunavo" },          // cheapest niche provider
    { model: gateway("google/gemini-3-flash"), label: "ai-gateway" }, // hosted gateway as fallback
  ],
}

A gateway picks within its own catalog; ai-lcr picks between gateways and direct providers — and runs in-process, so it adds no network hop of its own.