feat: add auto discount (#607)

Co-authored-by: xjm <[email protected]>

Author: xiaojingming

packages/types/src/model.ts

@@ -84,6 +84,9 @@ export const modelInfoSchema = z.object({
 	deprecated: z.boolean().optional(),
 	// Flag to indicate if the model is free (no cost)
 	isFree: z.boolean().optional(),
+	// Credit consumption and discount for the model
+	creditConsumption: z.number().optional(),
+	creditDiscount: z.number().optional(),
 	/**
 	 * Service tiers with pricing information.
 	 * Each tier can have a name (for OpenAI service tiers) and pricing overrides.

src/api/providers/zgsm.ts

@@ -176,6 +176,8 @@ export class ZgsmAiHandler extends BaseProvider implements SingleCompletionHandl
 				})
 			}
 			let stream
+			let selectedLlm: string | undefined
+			let selectReason: string | undefined
 			try {
 				this.logger.info(`[RequestID]:`, requestId)
 				const { data, response } = await this.client.chat.completions
@@ -189,20 +191,31 @@ export class ZgsmAiHandler extends BaseProvider implements SingleCompletionHandl
 					.withResponse()
 				this.logger.info(`[ResponseID]:`, response.headers.get("x-request-id"))
 
-				stream = data
 				if (this.options.zgsmModelId === autoModeModelId) {
+					selectedLlm = response.headers.get("x-select-llm") || ""
+					selectReason = response.headers.get("x-select-reason") || ""
+
+					if (selectedLlm) {
+						this.logger.info(`[Selected LLM]:`, selectedLlm)
+					}
+					if (selectReason) {
+						this.logger.info(`[Select Reason]:`, selectReason)
+					}
+
 					const userInputHeader = response.headers.get("x-user-input")
 					if (userInputHeader) {
 						const decodedUserInput = Buffer.from(userInputHeader, "base64").toString("utf-8")
 						this.logger.info(`[x-user-input]: ${decodedUserInput}`)
 					}
 				}
+
+				stream = data
 			} catch (error) {
 				throw handleOpenAIError(error, this.providerName)
 			}
 
 			// 6. Optimize stream processing - use batch processing and buffer
-			yield* this.handleOptimizedStream(stream, modelInfo)
+			yield* this.handleOptimizedStream(stream, modelInfo, selectedLlm, selectReason)
 		} else {
 			// Non-streaming processing
 			const requestOptions = this.buildNonStreamingRequestOptions(
@@ -406,6 +419,8 @@ export class ZgsmAiHandler extends BaseProvider implements SingleCompletionHandl
 	private async *handleOptimizedStream(
 		stream: AsyncIterable<OpenAI.Chat.Completions.ChatCompletionChunk>,
 		modelInfo: ModelInfo,
+		selectedLlm?: string,
+		selectReason?: string,
 	): ApiStream {
 		const matcher = new XmlMatcher(
 			"think",
@@ -423,6 +438,14 @@ export class ZgsmAiHandler extends BaseProvider implements SingleCompletionHandl
 		let time = Date.now()
 		let isPrinted = false
 
+		// Yield selected LLM info if available (for Auto model mode)
+		if (selectedLlm && this.options.zgsmModelId === autoModeModelId) {
+			yield {
+				type: "text",
+				text: `[Selected LLM: ${selectedLlm}${selectReason ? ` (${selectReason})` : ""}]`,
+			}
+		}
+
 		// chunk
 		for await (const chunk of stream) {
 			const delta = chunk.choices[0]?.delta ?? {}

src/core/task/Task.ts

@@ -2038,6 +2038,36 @@ export class Task extends EventEmitter<TaskEvents> implements TaskLike {
 								}
 								break
 							case "text": {
+								// Check if it is Selected LLM information (only in Auto model mode).
+								if (
+									this.apiConfiguration.zgsmModelId === "Auto" &&
+									chunk.text?.startsWith("[Selected LLM:")
+								) {
+									// 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
+									}
+								}
+
 								assistantMessage += chunk.text
 
 								// Parse raw assistant message chunk into content blocks.

src/core/webview/__tests__/webviewMessageHandler.routerModels.spec.ts

@@ -8,13 +8,30 @@ vi.mock("vscode", () => ({
 		showErrorMessage: vi.fn(),
 		showWarningMessage: vi.fn(),
 		showInformationMessage: vi.fn(),
+		createTextEditorDecorationType: vi.fn(() => ({
+			dispose: vi.fn(),
+		})),
+		createOutputChannel: vi.fn(() => ({
+			appendLine: vi.fn(),
+			append: vi.fn(),
+			clear: vi.fn(),
+			show: vi.fn(),
+			hide: vi.fn(),
+			dispose: vi.fn(),
+		})),
 	},
 	workspace: {
 		workspaceFolders: undefined,
 		getConfiguration: vi.fn(() => ({
 			get: vi.fn(),
 			update: vi.fn(),
 		})),
+		createFileSystemWatcher: vi.fn(() => ({
+			onDidCreate: vi.fn(),
+			onDidChange: vi.fn(),
+			onDidDelete: vi.fn(),
+			dispose: vi.fn(),
+		})),
 	},
 	env: {
 		clipboard: { writeText: vi.fn() },
@@ -32,6 +49,7 @@ vi.mock("vscode", () => ({
 		Workspace: 2,
 		WorkspaceFolder: 3,
 	},
+	RelativePattern: vi.fn().mockImplementation((base, pattern) => ({ base, pattern })),
 }))
 
 // Mock modelCache getModels/flushModels used by the handler

src/shared/ExtensionMessage.ts

@@ -500,6 +500,8 @@ export interface ClineApiReqInfo {
 	cancelReason?: ClineApiReqCancelReason
 	streamingFailedMessage?: string
 	apiProtocol?: "anthropic" | "openai"
+	selectedLlm?: string
+	selectReason?: string
 }
 
 export type ClineApiReqCancelReason = "streaming_failed" | "user_cancelled"

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

@@ -223,13 +223,13 @@ export const ChatRowContent = ({
 		vscode.postMessage({ type: "selectImages", context: "edit", messageTs: message.ts })
 	}, [message.ts])
 
-	const [cost, apiReqCancelReason, apiReqStreamingFailedMessage] = useMemo(() => {
+	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]
+			return [info?.cost, info?.cancelReason, info?.streamingFailedMessage, info?.selectedLlm, info?.selectReason]
 		}
 
-		return [undefined, undefined, undefined]
+		return [undefined, undefined, undefined, undefined, undefined]
 	}, [message.text, message.say])
 
 	// When resuming task, last wont be api_req_failed but a resume_task
@@ -1119,6 +1119,24 @@ 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>
+							)}
 							{(((cost === null || cost === undefined) && apiRequestFailedMessage) ||
 								apiReqStreamingFailedMessage) && (
 								<ErrorRow

webview-ui/src/components/settings/ModelPicker.tsx

@@ -304,26 +304,50 @@ export const ModelPicker = ({
 									)}
 								</CommandEmpty>
 								<CommandGroup>
-									{modelIds.map((model) => (
-										<CommandItem
-											key={model}
-											value={model}
-											onSelect={onSelect}
-											data-testid={`model-option-${model}`}
-											className={
-												model === "Auto" ? "border-b border-vscode-dropdown-border" : ""
-											}>
-											<span className="truncate" title={model}>
-												{model}
-											</span>
-											<Check
+									{modelIds.map((model) => {
+										const modelInfo = models?.[model]
+										const creditConsumption = modelInfo?.creditConsumption
+										const creditDiscount = modelInfo?.creditDiscount
+
+										return (
+											<CommandItem
+												key={model}
+												value={model}
+												onSelect={onSelect}
+												data-testid={`model-option-${model}`}
 												className={cn(
-													"size-4 p-0.5 ml-auto",
-													model === selectedModelId ? "opacity-100" : "opacity-0",
-												)}
-											/>
-										</CommandItem>
-									))}
+													model === "Auto" ? "border-b border-vscode-dropdown-border" : "",
+												)}>
+												<Check
+													className={cn(
+														"size-4 p-0.5",
+														model === selectedModelId ? "opacity-100" : "opacity-0",
+													)}
+												/>
+												<span className="truncate" title={model}>
+													{model}
+												</span>
+												{model === "Auto"
+													? creditDiscount && (
+															<span
+																className="ml-auto text-xs text-vscode-foreground bg-vscode-statusBarItem-prominentBackground px-1.5 py-0.5 rounded border border-vscode-button-border"
+																title={t("settings:autoMode.discountTitle")}>
+																{t("settings:autoMode.discount", {
+																	discount: `🎯 ${creditDiscount * 100}%`,
+																})}
+															</span>
+														)
+													: creditConsumption &&
+														creditConsumption !== -1 && (
+															<span
+																className="ml-auto text-sm text-vscode-descriptionForeground"
+																title={t("settings:autoMode.consumptionTitle")}>
+																{creditConsumption}x credit
+															</span>
+														)}
+											</CommandItem>
+										)
+									})}
 								</CommandGroup>
 							</CommandList>
 							{searchValue && !modelIds.includes(searchValue) && (

webview-ui/src/components/settings/ProviderRenderer.tsx

@@ -53,7 +53,15 @@ const ProviderRenderer: React.FC<ProviderRendererProps> = ({
 		switch (message.type) {
 			case "zgsmModels": {
 				const updatedModels = message.openAiModels ?? []
-				setOpenAiModels(Object.fromEntries(updatedModels.map((item) => [item, zgsmModels.default])))
+				const { fullResponseData = [] } = message
+				setOpenAiModels(
+					Object.fromEntries(
+						updatedModels.map((item) => [
+							item,
+							fullResponseData.find((itm) => itm.id === item) || zgsmModels.default,
+						]),
+					),
+				)
 				break
 			}
 			case "openAiModels": {

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

@@ -125,5 +125,9 @@
 			"description": "Generates executable test solutions based on analysis of project code.",
 			"initPrompt": "Generate test solution for the current project"
 		}
+	},
+	"autoMode": {
+		"selectedLlm": "Select model: {{selectedLlm}}",
+		"selectReason": "Select reason: {{selectReason}}"
 	}
 }

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

@@ -201,5 +201,10 @@
 			"failedToDisableCodebaseIndex": "Failed to disable codebase index:",
 			"failedToCopyToClipboard": "Failed to copy to clipboard:"
 		}
+	},
+	"autoMode": {
+		"discountTitle": "auto模式折扣",
+		"discount": "{{discount}} 折扣",
+		"consumptionTitle": "使用该模型消耗的credit"
 	}
 }