Streaming & Realtime with Surf

Name: Surf
Author: Surf

Build commands that stream progressive output via SSE and push real-time events to AI agents via WebSocket with session-scoped delivery.

When to Use Streaming

Not every command needs streaming. Use it for operations that take more than a few hundred milliseconds and where the agent benefits from progressive results: AI text generation, large data exports, complex computations, or real-time monitoring.

Request Flow

Agent

→

POST /surf/execute

→

SSE Stream

→

chunk 1

chunk 2

done

Streaming Commands

Mark a command as streaming with stream: true and use ctx.emit() to send progressive chunks:

streaming.ts

import { createSurf } from '@surfjs/core'
 
const surf = createSurf({
  name: 'AI Writer',
  commands: {
    // Streaming text generation
    generate: {
      description: 'Generate text with streaming output',
      stream: true,
      params: {
        prompt: { type: 'string', required: true },
        maxTokens: { type: 'number', default: 500 },
        model: { type: 'string', default: 'gpt-4' },
      },
      run: async ({ prompt, maxTokens, model }, ctx) => {
        const stream = await openai.chat.completions.create({
          model,
          messages: [{ role: 'user', content: prompt }],
          max_tokens: maxTokens,
          stream: true,
        })
 
        for await (const chunk of stream) {
          const text = chunk.choices[0]?.delta?.content
          if (text) {
            // ctx.emit() sends an SSE chunk to the client
            ctx.emit!({ text, done: false })
          }
        }
 
        // Return value is sent as the final "done" event
        return { done: true, model }
      },
    },
 
    // Streaming data export
    'export': {
      description: 'Export large datasets with progress',
      stream: true,
      auth: 'required',
      params: {
        table: { type: 'string', required: true },
        format: { type: 'string', enum: ['json', 'csv'], default: 'json' },
      },
      run: async ({ table, format }, ctx) => {
        const total = await db.count(table)
        let processed = 0
 
        for await (const batch of db.streamBatches(table, 100)) {
          processed += batch.length
          ctx.emit!({
            rows: format === 'csv' ? toCsv(batch) : batch,
            progress: Math.round((processed / total) * 100),
          })
        }
 
        return { total: processed, format }
      },
    },
  },
})

💡 Tip: Both the command definition must have stream: true and the client request must include stream: true for SSE to activate.

Client-Side Streaming

The client SDK handles SSE streams automatically:

agent-stream.ts

import { SurfClient } from '@surfjs/client'
 
const client = await SurfClient.discover('https://ai-writer.com')
 
// Execute returns the final result after streaming completes
const result = await client.execute('generate', {
  prompt: 'Explain quantum computing in 3 sentences',
})
// result => { done: true, model: 'gpt-4' }
 
// For raw HTTP SSE access:
const response = await fetch('https://ai-writer.com/surf/execute', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    command: 'generate',
    params: { prompt: 'Hello world' },
    stream: true,
  }),
})
 
// Parse SSE events
const reader = response.body.getReader()
const decoder = new TextDecoder()
while (true) {
  const { done, value } = await reader.read()
  if (done) break
  const text = decoder.decode(value)
  // data: { "type": "chunk", "data": { "text": "Hello" } }
  // data: { "type": "done", "result": { "done": true } }
}

WebSocket & Events

For real-time bidirectional communication, Surf supports WebSocket connections. Define events in your config and emit them from anywhere in your code:

realtime-server.ts

import { createSurf } from '@surfjs/core'
import http from 'node:http'
import express from 'express'
 
const surf = createSurf({
  name: 'Ops Dashboard',
  // Define events with scoping
  events: {
    'metrics.update': {
      description: 'Real-time metrics push',
      scope: 'session', // Only the subscribing session receives this
      data: {
        cpu: { type: 'number' },
        memory: { type: 'number' },
        timestamp: { type: 'number' },
      },
    },
    'alert.triggered': {
      description: 'System alert',
      scope: 'broadcast', // All connected clients receive this
      data: {
        severity: { type: 'string' },
        message: { type: 'string' },
      },
    },
  },
  commands: {
    status: {
      description: 'Get current system status',
      hints: { idempotent: true, sideEffects: false },
      run: async () => ({
        cpu: await metrics.cpu(),
        memory: await metrics.memory(),
        requests: await metrics.requestsPerMinute(),
        uptime: process.uptime(),
      }),
    },
    'metrics.stream': {
      description: 'Stream live metrics via SSE',
      stream: true,
      params: {
        interval: { type: 'number', default: 1000, description: 'Interval in ms' },
      },
      run: async ({ interval }, ctx) => {
        for (let i = 0; i < 60; i++) {
          ctx.emit!({
            cpu: await metrics.cpu(),
            memory: await metrics.memory(),
            timestamp: Date.now(),
          })
          await sleep(interval)
        }
        return { streamed: 60 }
      },
    },
  },
})
 
const app = express()
app.use(express.json())
app.use(surf.middleware())
 
const server = http.createServer(app)
surf.wsHandler(server) // Attach WebSocket transport (requires 'ws' package)
 
server.listen(3000)
 
// Emit events from your application logic
setInterval(async () => {
  surf.emit('metrics.update', {
    cpu: await metrics.cpu(),
    memory: await metrics.memory(),
    timestamp: Date.now(),
  })
}, 5000)
 
// Broadcast alerts to all connected clients
alerting.on('alert', (alert) => {
  surf.emit('alert.triggered', {
    severity: alert.severity,
    message: alert.message,
  })
})

Event Scoping — Security Feature

Event scoping is a key security feature that controls who receives each event. Surf supports three scopes:

Scope	Delivery	Example
session	Only the triggering session	Cart updates, user metrics
global	All subscribers (inc. server-side)	System announcements
broadcast	All connected sessions	Stock updates, live alerts

⚠️ Security: Events default to session scope, preventing one user from receiving another user's updates. Only use broadcast for non-sensitive data.

Agent WebSocket Client

Agents connect via WebSocket to execute commands in real-time and receive event push notifications:

agent-realtime.ts

import { SurfClient } from '@surfjs/client'
 
const client = await SurfClient.discover('https://ops-dashboard.com')
 
// Connect via WebSocket
const ws = await client.connect()
 
// Execute commands over WebSocket
const status = await ws.execute('status', {})
console.log('CPU:', status.cpu)
 
// Listen for events
ws.on('alert.triggered', (data) => {
  console.log(`[ALERT] ${data.severity}: ${data.message}`)
})
 
ws.on('metrics.update', (data) => {
  console.log(`CPU: ${data.cpu}%, Memory: ${data.memory}%`)
})
 
// Clean up
client.disconnect()

Summary

Surf's streaming and event system lets you build real-time AI agent integrations with minimal code. Use SSE streaming with ctx.emit() for progressive output, WebSocket for bidirectional communication, and event scoping for secure, session-aware delivery. The client SDK handles both transparently.

Ask your agent

Copy these prompts into Claude, OpenClaw, or any AI agent

OpenClaw / Claude

Add Surf streaming to my Node.js app. Install @surfjs/core and use createSurf() to define a "generate" command with stream: true that uses ctx.emit() to send chunks. Also define events with session scoping for real-time updates. Mount on Express with surf.middleware() and attach WebSocket with surf.wsHandler(server).

Claude Code / Cursor

My Next.js app has a live feed. Add Surf with createSurf() including: a streaming "generate" command using ctx.emit(), event definitions with session/broadcast scopes, and WebSocket support via surf.wsHandler(). Use @surfjs/core.

Test streaming with @surfjs/client

Using @surfjs/client, connect to http://localhost:3000 and call the "generate" command with { prompt: "explain Surf.js in 3 sentences" }. Then connect via WebSocket and listen for events.

💡 Works with OpenClaw, Claude Code, Cursor, Codex, and any agent that can make HTTP requests.

← Information Sites Back to Docs →