Proxy adapter

createProxyClient is the recommended production path. The browser POSTs { prompt, system, model, ... } to your backend; your backend forwards to the LLM provider holding the API key.

Use it

import { createProxyClient } from '@extedcoud/smart-components/adapters/proxy';

const client = createProxyClient({ url: '/api/smart' });

Backend contract

Your endpoint receives POST /api/smart with:

{ "prompt": "...", "system": "...", "model": "gpt-4o-mini", "maxTokens": 32, "temperature": 0.4, "stop": ["\n\n"] }

It must respond with:

{ "text": "the completion" }

For streaming, the endpoint should return text/event-stream with data: <chunk> lines. The adapter parses SSE natively.

Capabilities

Why proxy?

• Key safety. Browser code can't be trusted with provider keys.
• Rate limiting. Add your own quota / abuse limits at the proxy.
• Vendor swap. Change provider on the server without shipping a new client build.
• Auth. Attach your existing session cookies / JWT — no separate auth flow.