Automatic fallbacks across OpenAI, Anthropic, Groq, and local models — with attempts, cost, and cache on every response.
—
“Quiet model switches get a reputation fast. We don't do quiet.”
res.provider · res.attempts · res.costfast / smart / cheap — change models in one config object. Typed route names so route("smrt") fails at compile time.
const llm = createRouter({
routes: {
fast: { primary: "groq/llama-3.3-70b" },
smart: { primary: "anthropic/claude-4-5" },
cheap: { primary: "openai/gpt-4o-mini" },
},
default: "smart",
})
await llm.route("fast").complete(ticket) Runs in your process, with your API keys, over fetch. No hop. No signup. Nobody else sees your prompts.
Deterministic key from model plus messages. Set cache: { ttl: "24h" } and repeats come back as cached: true at zero cost.
The mock provider answers from a queue you control — including failures, so you can assert the fallback path instead of hoping for it.
const llm = createRouter({
primary: "mock/fail",
fallbacks: ["mock/ok"],
})
const res = await llm.complete("hi")
expect(res.provider).toBe("mock")
expect(res.attempts).toHaveLength(2) Attempts walk the chain in order and stop at the first success. If every provider is down you get AllProvidersFailed with each error attached.
usage, cost, latencyMs, cached, and the full attempts trace — no wrapper, no vendor dashboard.
Providers
Runtime deps
Public api
Attempts
Cache
Cost
Start with one provider, add fallbacks when you need them. Any OpenAI-compatible endpoint works with a baseUrl, and local models never leave your machine.
createRouter is the entire entry point. Model strings, not classes. Same option keys at global, route, and call.
import { createRouter } from "llm-sdk"
const llm = createRouter({
primary: "anthropic/claude-sonnet-4-5",
fallbacks: ["openai/gpt-4o"],
})
const res = await llm.complete("Summarise this…")
console.log(res.provider, res.attempts)Router · fallbacks · routes · cache · cost — without an agent framework.