Hot Reload

What it is

Hot Reload pushes prompt changes to running applications in real time. When an editor saves a prompt, every connected SDK refreshes its cache automatically — no restart, no redeploy.

Why it matters

Without it, updating a prompt means restarting your AI service. With it, the prompt is a runtime resource: change the wording, hit save, and production picks it up within milliseconds.

How it works

1. The SDK calls watch(), which opens a WebSocket to the server at /ws.
2. When a prompt is created, updated or deleted, the server broadcasts a small event — { type, key, env, version } — to every connected client.
3. Each SDK checks the event: if it has that prompt key cached, it re-fetches the latest content and fires an update event for your code to react to.

The web UI connects the same way — the Prompts page shows a "Live updates connected" badge when its WebSocket is open.

Using it from the SDK

import { PromptOpsClient } from '@promptops/client'

const client = new PromptOpsClient({
  server: 'http://localhost:8080',
  namespace: 'prod',
  appName: 'checkout-service', // shown in the Connected clients panel
})

// prime the cache
await client.render('sql.generator', { question: 'list all users' })

// react to live changes
client.on('update', (e) => console.log('prompt hot-reloaded:', e.key))
client.on('connect', () => console.log('watching'))
client.watch()

After watch(), any edit to sql.generator in the UI triggers the update callback and refreshes the cached content automatically.

Seeing who is connected

Every watch() connection is tracked. The Observability page lists them under Connected clients — see Observability.

Next steps

• Prompts — the resource that gets hot-reloaded.
• SDK — full client reference.