Monorch

LiteLLM proxy

Point the OpenAI-compatible client at LiteLLM (or OpenRouter / vLLM) with baseUrl. Same openai() helper as direct OpenAI.

Pattern

litellm.ts
import { agent } from "@monorch/ai";
import { openai } from "@monorch/ai/openai";

const model = openai(process.env.LIVE_MODEL ?? "gpt-4.1-mini", {
  apiKey: process.env.LITELLM_API_KEY ?? process.env.OPENAI_API_KEY,
  baseUrl: process.env.LITELLM_URL ?? "http://127.0.0.1:4000/v1",
  timeoutMs: 60_000,
});

const bot = agent({
  name: "proxy",
  model,
  instructions: "Be concise.",
});

await bot.run("ping");

Notes

Auth missing → OPENAI_AUTH. Non-OK HTTP → OPENAI_HTTP. Agent opts.signal forwards into provider fetch. See Errors and openai reference.

FAQ