SSE 流式输出教程：用 Go 实现大模型打字机效果（Server-Sent Events 原理 + openai-go 实战）

stream := client.Responses.NewStreaming(ctx, responses.ResponseNewParams{
	Model: openai.ChatModelGPT5_2,
	Input: responses.ResponseNewParamsInputUnion{
		OfString: openai.String("写一首关于编程的俳句"),
	},
})
defer func() { _ = stream.Close() }()

for stream.Next() {
	event := stream.Current()
	print(event.Delta) // 增量文本，逐 token 到达
}
if stream.Err() != nil {
	// 处理错误
}

package main

import (
	"context"
	"fmt"
	"net/http"
	"os"
	"strconv"
	"time"

	"github.com/openai/openai-go/v3"
	"github.com/openai/openai-go/v3/option"
	"github.com/openai/openai-go/v3/responses"
)

var client *openai.Client

func send(w http.ResponseWriter, flusher http.Flusher, payload string) {
	fmt.Fprintf(w, "data: %s\n\n", strconv.Quote(payload)) // JSON 字符串保证单行
	flusher.Flush()                                         // 立即推给浏览器
}

func chatHandler(w http.ResponseWriter, r *http.Request) {
	flusher, ok := w.(http.Flusher)
	if !ok {
		http.Error(w, "streaming unsupported", http.StatusInternalServerError)
		return
	}
	w.Header().Set("Content-Type", "text/event-stream")
	w.Header().Set("Cache-Control", "no-cache")
	w.Header().Set("X-Accel-Buffering", "no") // 关键：禁用 Nginx 缓冲，见下文

	// 客户端断开连接时，ctx 自动取消，模型调用随之停止（省 token）
	stream := client.Responses.NewStreaming(r.Context(), responses.ResponseNewParams{
		Model: openai.ChatModelGPT5_2,
		Input: responses.ResponseNewParamsInputUnion{
			OfString: openai.String(r.URL.Query().Get("q")),
		},
	})
	defer func() { _ = stream.Close() }()

	for stream.Next() {
		event := stream.Current()
		if event.Delta == "" {
			continue
		}
		send(w, flusher, event.Delta)
	}
	if err := stream.Err(); err != nil && err != context.Canceled {
		send(w, flusher, "[error] "+err.Error())
	}
	send(w, flusher, "[done]") // 结束标记，前端收到后主动关闭连接
}

func main() {
	client = openai.NewClient(option.WithAPIKey(os.Getenv("OPENAI_API_KEY")))
	http.HandleFunc("/chat", chatHandler)
	srv := &http.Server{Addr: ":8080", ReadHeaderTimeout: 10 * time.Second}
	_ = srv.ListenAndServe()
}

<script>
const es = new EventSource('/chat?q=' + encodeURIComponent('用三句话介绍 Go 语言'));
es.onmessage = (e) => {
	const chunk = JSON.parse(e.data); // 服务端发的是 JSON 字符串
	if (chunk === '[done]') { es.close(); return; }
	document.body.innerText += chunk;
};
es.onerror = () => es.close();
</script>