Deploy Jev Hooks

Choose hosting and inference independently. A Next.js app on Vercel can call TypeSafe directly. A Cloudflare Worker can use Workers AI, TypeSafe, OpenRouter or Vercel AI Gateway.

Application	Browser entry	Server endpoint	Inference adapter
Next.js / Vercel	`jev-hooks/react`	App Router route, Node 24	`VercelJevAdapter` or another server adapter
React / Cloudflare	`jev-hooks/react`	Worker with `nodejs_compat`	`CloudflareJevAdapter` or another server adapter
Node backend	None required	Your handler / job	Any server adapter
Offline development	`jev-hooks/react` or root	Optional	`MockJudgmentAdapter`

The boundary#

Browser                            Server
useScore / useNoul                  authenticated POST /api/jev
        ↓                                     ↓
createJevClient({ endpoint })  →    createJevHandler({ adapter, authorize })
                                              ↓
                                      TypeSafe / provider

The browser owns its React subscription lifecycle and batching. The server validates and authorizes requests and calls the provider. Secrets never enter the browser bundle.

Next.js and Vercel: Node route and AI Gateway configuration.
Cloudflare Workers: Worker, AI binding and Wrangler configuration.
Production behavior: authorization, rate limits, caches, failures and usage accounting.

Other servers#

createJevHandler accepts a standard Request and returns a Promise<Response>. Adapt your framework's request/response boundary to it. The server package requires Node-compatible APIs; arbitrary edge runtimes are not claimed supported.

For React Server Components, keep interactive hooks in a "use client" component. Use run() on the server for a server computation. React hooks do not perform inference during server rendering.