1

server-node/src/services/llmClient.ts

@@ -309,6 +309,92 @@ export class UpstreamLlmClient {
     return content;
   }
 
+  async streamMessageContent(params: {
+    systemPrompt: string;
+    userPrompt: string;
+    model?: string;
+    signal?: AbortSignal;
+    timeoutMs?: number;
+    debugLabel?: string;
+    onUpdate?: (text: string) => void;
+  }) {
+    const response = await this.requestCompletion(
+      {
+        model: params.model,
+        stream: true,
+        messages: [
+          { role: 'system', content: params.systemPrompt },
+          { role: 'user', content: params.userPrompt },
+        ],
+      },
+      {
+        signal: params.signal,
+        timeoutMs: params.timeoutMs,
+        debugLabel: params.debugLabel,
+      },
+    );
+
+    if (!response.body) {
+      throw upstreamError('LLM 流式响应体不可用');
+    }
+
+    const reader = response.body.getReader();
+    const decoder = new TextDecoder('utf-8');
+    let buffer = '';
+    let accumulatedText = '';
+
+    for (;;) {
+      const { done, value } = await reader.read();
+      if (done) {
+        break;
+      }
+
+      buffer += decoder.decode(value, { stream: true });
+
+      while (buffer.includes('\n\n')) {
+        const boundary = buffer.indexOf('\n\n');
+        const eventBlock = buffer.slice(0, boundary);
+        buffer = buffer.slice(boundary + 2);
+
+        for (const rawLine of eventBlock.split(/\r?\n/u)) {
+          const line = rawLine.trim();
+          if (!line.startsWith('data:')) {
+            continue;
+          }
+
+          const data = line.slice(5).trim();
+          if (!data || data === '[DONE]') {
+            continue;
+          }
+
+          try {
+            const parsed = JSON.parse(data) as {
+              choices?: Array<{
+                delta?: {
+                  content?: string;
+                };
+              }>;
+            };
+            const delta = parsed.choices?.[0]?.delta?.content;
+            if (typeof delta === 'string' && delta.length > 0) {
+              accumulatedText += delta;
+              params.onUpdate?.(accumulatedText);
+            }
+          } catch {
+            // Ignore malformed SSE frames from the upstream model.
+          }
+        }
+      }
+    }
+
+    const content = accumulatedText.trim();
+    if (!content) {
+      throw upstreamError('LLM 返回内容为空');
+    }
+
+    return content;
+  }
+
   async forwardCompletion(
     request: ExpressRequest,
     body: Record<string, unknown>,