lib/hf/client.ts

import { InferenceClient } from "@huggingface/inference";

const JWT_SHAPE = /^eyJ[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+$/;

export function createInferenceClient(userAccessToken?: string) {
  // Prefer the signed-in user's HF OAuth access token (granted via the
  // `inference-api` scope on sign-in). Falls back to a server-side
  // HF_TOKEN env var if no user token is present (e.g. background jobs).
  if (userAccessToken) {
    return new InferenceClient(userAccessToken);
  }

  const serverToken = process.env.HF_TOKEN?.trim();
  if (!serverToken) {
    throw new Error(
      "No HuggingFace token available. Sign in to use your OAuth token, or set HF_TOKEN to a Personal Access Token (starts with `hf_`) from https://huggingface.co/settings/tokens.",
    );
  }
  if (JWT_SHAPE.test(serverToken)) {
    throw new Error(
      "HF_TOKEN looks like an OAuth access token (JWT). For server-side inference set HF_TOKEN to a long-lived Personal Access Token starting with `hf_` from https://huggingface.co/settings/tokens, or remove HF_TOKEN and rely on per-user OAuth.",
    );
  }
  return new InferenceClient(serverToken);
}