feat(chat): enhance auto model selection display and information handling (#657)

Author: mini2s

src/api/providers/zgsm.ts

@@ -170,8 +170,9 @@ export class ZgsmAiHandler extends BaseProvider implements SingleCompletionHandl
 			if (fromWorkflow) {
 				requestOptions.extra_body.prompt_mode = "strict"
 			}
-			let stream
-			let selectedLlm: string | undefined
+			const isAuto = this.options.zgsmModelId === autoModeModelId
+			let stream: any
+			let selectedLLM: string | undefined = this.options.zgsmModelId
 			let selectReason: string | undefined
 			try {
 				this.logger.info(`[RequestID]:`, requestId)
@@ -190,9 +191,8 @@ export class ZgsmAiHandler extends BaseProvider implements SingleCompletionHandl
 					)
 					.withResponse()
 				this.logger.info(`[ResponseID]:`, response.headers.get("x-request-id"))
-
-				if (this.options.zgsmModelId === autoModeModelId) {
-					selectedLlm = response.headers.get("x-select-llm") || ""
+				if (isAuto) {
+					selectedLLM = response.headers.get("x-select-llm") || ""
 					selectReason = response.headers.get("x-select-reason") || ""
 					const isDev = process.env.NODE_ENV === "development"
 
@@ -209,7 +209,7 @@ export class ZgsmAiHandler extends BaseProvider implements SingleCompletionHandl
 			}
 
 			// 6. Optimize stream processing - use batch processing and buffer
-			yield* this.handleOptimizedStream(stream, modelInfo, selectedLlm, selectReason)
+			yield* this.handleOptimizedStream(stream, modelInfo, isAuto, selectedLLM, selectReason)
 		} else {
 			// Non-streaming processing
 			const requestOptions = this.buildNonStreamingRequestOptions(
@@ -416,7 +416,8 @@ export class ZgsmAiHandler extends BaseProvider implements SingleCompletionHandl
 	private async *handleOptimizedStream(
 		stream: AsyncIterable<OpenAI.Chat.Completions.ChatCompletionChunk>,
 		modelInfo: ModelInfo,
-		selectedLlm?: string,
+		isAuto?: boolean,
+		selectedLLM?: string,
 		selectReason?: string,
 	): ApiStream {
 		const matcher = new XmlMatcher(
@@ -437,10 +438,12 @@ export class ZgsmAiHandler extends BaseProvider implements SingleCompletionHandl
 		const isDev = process.env.NODE_ENV === "development"
 
 		// Yield selected LLM info if available (for Auto model mode)
-		if (selectedLlm && this.options.zgsmModelId === autoModeModelId) {
+		if (isAuto) {
 			yield {
 				type: "text",
-				text: `[Selected LLM: ${selectedLlm}${selectReason ? ` (${selectReason})` : ""}]`,
+				text: `["${selectedLLM}" ${selectReason ? `, "(${selectReason})"` : ""}]`,
+				isAuto,
+				originModelId: this.options.zgsmModelId,
 			}
 		}
 
@@ -455,7 +458,7 @@ export class ZgsmAiHandler extends BaseProvider implements SingleCompletionHandl
 			// Cache content for batch processing
 			if (delta.content) {
 				contentBuffer.push(delta.content)
-				if (isDev && !isPrinted && chunk.model && this.options.zgsmModelId === autoModeModelId) {
+				if (isDev && !isPrinted && chunk.model && isAuto) {
 					this.logger.info(`[Current Model]: ${chunk.model}`)
 					isPrinted = true
 				}

src/api/transform/stream.ts

@@ -17,6 +17,8 @@ export interface ApiStreamError {
 export interface ApiStreamTextChunk {
 	type: "text"
 	text: string
+	isAuto?: boolean
+	originModelId?: string
 }
 
 export interface ApiStreamReasoningChunk {

src/core/task/Task.ts

@@ -1918,6 +1918,7 @@ export class Task extends EventEmitter<TaskEvents> implements TaskLike {
 				maxReadFileLine = -1,
 				maxReadCharacterLimit = 20000,
 				apiRequestBlockHide = true,
+				apiConfiguration,
 			} = (await this.providerRef.deref()?.getState()) ?? {}
 
 			await this.say(
@@ -1928,6 +1929,7 @@ export class Task extends EventEmitter<TaskEvents> implements TaskLike {
 						: currentUserContent.map((block) => formatContentBlockToMarkdown(block)).join("\n\n") +
 							"\n\nLoading...",
 					apiProtocol,
+					originModelId: apiConfiguration?.zgsmModelId,
 				}),
 			)
 
@@ -1970,6 +1972,7 @@ export class Task extends EventEmitter<TaskEvents> implements TaskLike {
 					? undefined
 					: finalUserContent.map((block) => formatContentBlockToMarkdown(block)).join("\n\n"),
 				apiProtocol,
+				originModelId: apiConfiguration?.zgsmModelId,
 			} satisfies ClineApiReqInfo)
 
 			await this.saveClineMessages()
@@ -2154,32 +2157,25 @@ export class Task extends EventEmitter<TaskEvents> implements TaskLike {
 							case "text": {
 								// Check if it is Selected LLM information (only in Auto model mode).
 								if (
-									this.apiConfiguration.zgsmModelId === "Auto" &&
-									chunk.text?.startsWith("[Selected LLM:")
+									chunk.isAuto &&
+									lastApiReqIndex >= 0 &&
+									this.clineMessages[lastApiReqIndex] &&
+									this.apiConfiguration.apiProvider === "zgsm"
 								) {
 									// Extract Selected LLM and Reason information and update the api_req_started message.
-									const match = chunk.text.match(/\[Selected LLM:\s*([^\]]+)\]/)
-									if (match && lastApiReqIndex >= 0 && this.clineMessages[lastApiReqIndex]) {
-										const existingData = JSON.parse(
-											this.clineMessages[lastApiReqIndex].text || "{}",
-										)
-										// Parse the model name and reason
-										const fullInfo = match[1]
-										const reasonMatch = fullInfo.match(/^(.+?)\s*\((.+?)\)$/)
-										const selectedLlm = reasonMatch ? reasonMatch[1].trim() : fullInfo.trim()
-										const selectReason = reasonMatch ? reasonMatch[2].trim() : undefined
-
-										this.clineMessages[lastApiReqIndex].text = JSON.stringify({
-											...existingData,
-											selectedLlm,
-											selectReason,
-										} satisfies ClineApiReqInfo)
-										// Save the selection information but do not add it to the assistant message to avoid it being processed by the parser.
-										console.log(
-											`[Auto Model] Selected: ${selectedLlm}${selectReason ? ` (${selectReason})` : ""}`,
-										)
-										break
-									}
+									const existingData = JSON.parse(this.clineMessages[lastApiReqIndex].text || "{}")
+									const [selectedLLM, selectReason] = JSON.parse(chunk.text)
+
+									this.clineMessages[lastApiReqIndex].text = JSON.stringify({
+										...existingData,
+										selectedLLM,
+										selectReason,
+										isAuto: chunk.isAuto,
+										originModelId: chunk.originModelId,
+									} satisfies ClineApiReqInfo)
+									// Save the selection information but do not add it to the assistant message to avoid it being processed by the parser.
+									console.log(`[Backend Model Route Detail] ${selectedLLM}${selectReason}`)
+									break
 								}
 
 								assistantMessage += chunk.text

src/shared/ExtensionMessage.ts

@@ -496,8 +496,10 @@ export interface ClineApiReqInfo {
 	cancelReason?: ClineApiReqCancelReason
 	streamingFailedMessage?: string
 	apiProtocol?: "anthropic" | "openai"
-	selectedLlm?: string
+	selectedLLM?: string
 	selectReason?: string
+	isAuto?: boolean
+	originModelId?: string
 }
 
 export type ClineApiReqCancelReason = "streaming_failed" | "user_cancelled"

webview-ui/src/components/chat/ChatRow.tsx

@@ -265,14 +265,23 @@ export const ChatRowContent = ({
 		vscode.postMessage({ type: "selectImages", context: "edit", messageTs: message.ts })
 	}, [message.ts])
 
-	const [cost, apiReqCancelReason, apiReqStreamingFailedMessage, selectedLlm, selectReason] = useMemo(() => {
-		if (message.text !== null && message.text !== undefined && message.say === "api_req_started") {
-			const info = safeJsonParse<ClineApiReqInfo>(message.text)
-			return [info?.cost, info?.cancelReason, info?.streamingFailedMessage, info?.selectedLlm, info?.selectReason]
-		}
+	const [cost, apiReqCancelReason, apiReqStreamingFailedMessage, selectedLLM, selectReason, isAuto, originModelId] =
+		useMemo(() => {
+			if (message.text !== null && message.text !== undefined && message.say === "api_req_started") {
+				const info = safeJsonParse<ClineApiReqInfo>(message.text)
+				return [
+					info?.cost,
+					info?.cancelReason,
+					info?.streamingFailedMessage,
+					info?.selectedLLM,
+					info?.selectReason,
+					info?.isAuto,
+					info?.originModelId,
+				]
+			}
 
-		return [undefined, undefined, undefined, undefined, undefined]
-	}, [message.text, message.say])
+			return [undefined, undefined, undefined, undefined, undefined]
+		}, [message.text, message.say])
 
 	// When resuming task, last wont be api_req_failed but a resume_task
 	// message, so api_req_started will show loading spinner. That's why we just
@@ -1196,24 +1205,22 @@ export const ChatRowContent = ({
 									${Number(cost || 0)?.toFixed(4)}
 								</div>
 							</div>
-							{(selectedLlm || selectReason) && (
-								<div className="mt-2 flex items-center flex-wrap gap-2">
-									{selectedLlm && (
-										<div
-											className="text-xs text-vscode-descriptionForeground border-vscode-dropdown-border/50 border px-1.5 py-0.5 rounded-lg"
-											title="Selected Model">
-											{t("chat:autoMode.selectedLlm", { selectedLlm })}
-										</div>
-									)}
-									{selectReason && (
-										<div
-											className="text-xs text-vscode-descriptionForeground border-vscode-dropdown-border/50 border px-1.5 py-0.5 rounded-lg"
-											title="Selection Reason">
-											{t("chat:autoMode.selectReason", { selectReason })}
-										</div>
-									)}
-								</div>
-							)}
+							<div className="mt-2 flex items-center flex-wrap gap-2">
+								{(selectedLLM || originModelId) && (
+									<div
+										className="text-xs text-vscode-descriptionForeground border-vscode-dropdown-border/50 border px-1.5 py-0.5 rounded-lg"
+										title="Selected Model">
+										{isAuto ? t("chat:autoMode.selectedLLM", { selectedLLM }) : originModelId}
+									</div>
+								)}
+								{selectReason && (
+									<div
+										className="text-xs text-vscode-descriptionForeground border-vscode-dropdown-border/50 border px-1.5 py-0.5 rounded-lg"
+										title="Selection Reason">
+										{t("chat:autoMode.selectReason", { selectReason })}
+									</div>
+								)}
+							</div>
 							{(((cost === null || cost === undefined) && apiRequestFailedMessage) ||
 								apiReqStreamingFailedMessage) && (
 								<ErrorRow

webview-ui/src/i18n/costrict-i18n/locales/en/chat.json

@@ -127,7 +127,7 @@
 		}
 	},
 	"autoMode": {
-		"selectedLlm": "Auto model: {{selectedLlm}}",
+		"selectedLLM": "Auto model: {{selectedLLM}}",
 		"selectReason": "Select reason: {{selectReason}}"
 	}
 }