😵‍💫 Integrating Alibaba DeepSeek AI in a Frontend Project — Why `netlify dev` Breaks Streaming and How to Fix It

🧾 Background

While integrating the Alibaba Cloud DeepSeek Assistant API into a frontend project, I wanted real-time AI responses via fetch + stream (ReadableStream). Backend used Express to call DashScope’s streaming endpoint with X-DashScope-SSE: enable, and the frontend expected to receive incremental AI tokens like a chat assistant.

This setup worked perfectly in production.

But in local development with netlify dev?

❌ Streaming completely failed, frontend received nothing, and eventually crashed with:

net::ERR_EMPTY_RESPONSE

🧨 Problem Summary

Even though the backend successfully received and printed streaming chunks like:

data: {"output":{"text":"Hello"}}

The frontend never got them. Instead:

• fetch().body.getReader().read() was never triggered
• The browser hung waiting for data
• Connection dropped with ERR_EMPTY_RESPONSE

🔍 Root Cause

The Netlify CLI (netlify dev) emulates serverless locally, but its internal proxy does not support chunked transfer encoding, which is required for streaming with res.write().

✅ Solution: Bypass Netlify CLI, Run Express Locally

✅ Step 1: Run your backend manually

ts-node src/server.ts

Make sure your Express server listens on a custom port (e.g. 3001):

app.listen(3001, '0.0.0.0', () => {
  console.log('Running at http://localhost:3001');
});

✅ Step 2: Configure frontend proxy (Vite)

// vite.config.ts
server: {
  proxy: {
    '/chat': {
      target: 'http://localhost:3001',
      changeOrigin: true,
    },
  },
}

Now your frontend code still calls /chat, but under the hood it hits your local Express server instead of the broken netlify dev proxy.

✅ Step 3: Make sure your backend properly streams

externalRes.on('data', (chunk) => {
  res.write(chunk); // Stream directly to frontend
});

externalRes.on('end', () => {
  res.end(); // Always close stream
});

💡 Use res.setHeader('Content-Type', 'text/event-stream') for best results if you’re mimicking SSE.

✅ Alternative: Deploy to Netlify

Deployed Netlify Functions do support streaming!

netlify deploy --prod

But don't rely on local dev for anything stream-based.

🧠 Summary

📝 Takeaway

If you're building real-time chat UIs, AI copilots, or stream-based assistants using Alibaba DeepSeek, OpenAI, or similar APIs:

⚠️ Don't use netlify dev during development.

✅ Use Express locally + Vite proxy for a reliable streaming experience.