feat: added performance statistics info when using embedded LLM

src/agent/core/AgentCore.ts

@@ -110,6 +110,7 @@ export class AgentCore {
         let assistantTextContent = '';
         const accumulatedToolCalls: ToolCall[] = [];
         let finishReason = 'stop';
+        let performanceInfo: StreamChunk['performanceInfo'];
 
         for await (const chunk of this.llmProvider.generateStream(conversationHistory, systemPrompt, tools)) {
           if (chunk.type === 'text') {
@@ -120,6 +121,7 @@ export class AgentCore {
             accumulatedToolCalls.push(chunk.toolCall);
           } else if (chunk.type === 'done') {
             finishReason = chunk.finishReason ?? 'stop';
+            performanceInfo = chunk.performanceInfo;
           }
         }
 
@@ -163,10 +165,9 @@ export class AgentCore {
           // LLM finished with text response (stop reason)
           this.agentState.updateMessage(this.currentAssistantMessageId, assistantTextContent);
           continueLoop = false;
+          yield { type: 'done', content: '', finishReason, performanceInfo };
         }
       }
-
-      yield { type: 'done', content: '', finishReason: 'stop' };
     } finally {
       this.currentUserMessageId = null;
       this.currentAssistantMessageId = null;

src/agent/llm/LocalBrowserLLMProvider.ts

@@ -296,6 +296,14 @@ export class LocalBrowserLLMProvider implements LLMProvider {
       yield { type: 'tool_call', content: '', toolCall };
     }
 
-    yield { type: 'done', content: '', finishReason };
+    yield {
+      type: 'done',
+      content: '',
+      finishReason,
+      performanceInfo: {
+        prefillTps,
+        generationTps,
+      },
+    };
   }
 }

src/agent/llm/StreamingTypes.ts

@@ -4,10 +4,16 @@
 
 import type { ToolCall } from '../core/AgentState';
 
+export interface PerformanceInfo {
+  prefillTps?: number;
+  generationTps?: number;
+}
+
 export interface StreamChunk {
   type: 'text' | 'tool_call' | 'tool_result' | 'done';
   content: string;
   toolCall?: ToolCall;
   toolResult?: { name: string; success: boolean; result: string };
-  finishReason?: string; // 'stop' | 'tool_calls' — present on 'done' chunks
+  performanceInfo?: PerformanceInfo;
+  finishReason?: string;
 }

src/components/ChatBox.css

@@ -226,6 +226,12 @@
   font-weight: bold;
 }
 
+.message-performance-info {
+  margin-top: 8px;
+  font-size: 10px;
+  color: #909090;
+}
+
 /* Abort link styling */
 .abort-link {
   background: none !important;

src/components/ChatBox.tsx

@@ -434,6 +434,7 @@ const ChatBox: React.FC<ChatBoxProps> = ({ isVisible }) => {
               key={message.id}
               content={message.content}
               isStreaming={message.isStreaming}
+              performanceInfo={message.performanceInfo}
               onAbort={message.isStreaming ? handleAbort : undefined}
             />
           )

src/components/chat/AssistantMessage.tsx

@@ -3,11 +3,13 @@ import ReactMarkdown from 'react-markdown';
 import remarkGfm from 'remark-gfm';
 import { Prism as SyntaxHighlighter } from 'react-syntax-highlighter';
 import { vscDarkPlus } from 'react-syntax-highlighter/dist/esm/styles/prism';
+import type { PerformanceInfo } from '../../agent/llm/StreamingTypes';
 
 interface AssistantMessageProps {
   content: string;
   isStreaming?: boolean;
   onAbort?: () => void;
+  performanceInfo?: PerformanceInfo;
 }
 
 // Memoized code component to prevent SyntaxHighlighter re-renders
@@ -30,7 +32,19 @@ const CodeComponent = memo(({ inline, className, children, ...props }: any) => {
   );
 });
 
-const AssistantMessage: React.FC<AssistantMessageProps> = ({ content, isStreaming, onAbort }) => {
+const formatTps = (value?: number): string | null => {
+  if (typeof value !== 'number' || !Number.isFinite(value) || value <= 0) {
+    return null;
+  }
+
+  return value.toFixed(1);
+};
+
+const AssistantMessage: React.FC<AssistantMessageProps> = ({ content, isStreaming, onAbort, performanceInfo }) => {
+  const prefillTps = formatTps(performanceInfo?.prefillTps);
+  const generationTps = formatTps(performanceInfo?.generationTps);
+  const hasPerformanceInfo = Boolean(prefillTps || generationTps);
+
   const renderContent = () => {
     // Handle special abort link for streaming messages
     if (isStreaming && onAbort && content.includes('click here to abort')) {
@@ -83,6 +97,13 @@ const AssistantMessage: React.FC<AssistantMessageProps> = ({ content, isStreamin
     <div className="message-container message-assistant">
       <div className="message-content">
         {renderContent()}
+        {hasPerformanceInfo && (
+          <div className="message-performance-info">
+            {prefillTps ? `Prefill: ${prefillTps} t/s` : 'Prefill: -'}
+            {' · '}
+            {generationTps ? `Generation: ${generationTps} t/s` : 'Generation: -'}
+          </div>
+        )}
       </div>
     </div>
   );

src/hooks/useStreamProcessor.ts

@@ -38,6 +38,7 @@ export const useStreamProcessor = (options: StreamProcessorOptions): StreamProce
       let assistantResponse = '';
       let tokenCount = 0;
       let hasTextContent = false;
+      let performanceInfo: ChatMessage['performanceInfo'];
 
       console.log(`------------ ${logPrefix} ------------`);
       console.log(input);
@@ -65,7 +66,8 @@ export const useStreamProcessor = (options: StreamProcessorOptions): StreamProce
               ...msg,
               content: assistantResponse,
               isStreaming: false,
-              tokenCount: undefined
+              tokenCount: undefined,
+              performanceInfo
             }));
 
             console.log('------------ ASSISTANT ------------');
@@ -98,19 +100,22 @@ export const useStreamProcessor = (options: StreamProcessorOptions): StreamProce
           assistantResponse = '';
           tokenCount = 0;
           hasTextContent = false;
+          performanceInfo = undefined;
 
           // Create a fresh streaming placeholder for the next LLM response
           const nextMsg = createStreamingMessage();
           currentStreamingId = nextMsg.id;
           onMessageAdd(nextMsg);
         } else if (chunk.type === 'done') {
+          performanceInfo = chunk.performanceInfo;
           // Finalize the streaming message
           if (hasTextContent) {
             onMessageUpdate(currentStreamingId, (msg) => ({
               ...msg,
               content: assistantResponse,
               isStreaming: false,
-              tokenCount: undefined
+              tokenCount: undefined,
+              performanceInfo
             }));
           } else {
             // No text in final response — remove empty placeholder

src/types/projectTypes.ts

@@ -1,4 +1,5 @@
 import { Transform, type TransformFnParams } from 'class-transformer';
+import type { PerformanceInfo } from '../agent/llm/StreamingTypes';
 
 export interface TimeSignature {
     numerator: number;
@@ -11,6 +12,7 @@ export interface ChatMessage {
     content: string;
     isStreaming?: boolean;
     tokenCount?: number;
+    performanceInfo?: PerformanceInfo;
 }
 
 /**