Requesty dynamic model selection (RooCodeInc#1836)

* Extract reuseable ModelDescriptionMarkdown from OpenRouter model picker

* Requesty: Add model picker component

* Refactor readOpenRouterModels to allow any dynamic list filename

* Extract parsePrice to allow reuse by other providers

* Simplify model display name switch case

* Requesty: Add dynamic model list fetching from API

* Requesty: Add default model selection

* Requesty: Specify max_tokens when sending request

* Add changeset

---------

Co-authored-by: Dennis Bartlett <[email protected]>

Author: dtrugman

.changeset/red-cars-cry.md

@@ -0,0 +1,11 @@
+---
+"claude-dev": patch
+---
+
+Add dynamic model fetching for the Requesty provider.
+
+Instead of manually typing the model name, the extension dynamically fetches
+all the supported model names from Requesty's /v1/models API.
+
+This allows users to use a fuzzy search logic when choosing the models and
+also guarantees the information for each model is up to date.

src/api/providers/requesty.ts

@@ -2,7 +2,7 @@ import { Anthropic } from "@anthropic-ai/sdk"
 import OpenAI from "openai"
 import { withRetry } from "../retry"
 import { calculateApiCostOpenAI } from "../../utils/cost"
-import { ApiHandlerOptions, ModelInfo, openAiModelInfoSaneDefaults } from "../../shared/api"
+import { ApiHandlerOptions, ModelInfo, openAiModelInfoSaneDefaults, requestyDefaultModelId, requestyDefaultModelInfo  } from "../../shared/api"
 import { ApiHandler } from "../index"
 import { convertToOpenAiMessages } from "../transform/openai-format"
 import { ApiStream } from "../transform/stream"
@@ -25,7 +25,7 @@ export class RequestyHandler implements ApiHandler {
 
 	@withRetry()
 	async *createMessage(systemPrompt: string, messages: Anthropic.Messages.MessageParam[]): ApiStream {
-		const modelId = this.options.requestyModelId ?? ""
+		const model = this.getModel()
 
 		let openAiMessages: OpenAI.Chat.ChatCompletionMessageParam[] = [
 			{ role: "system", content: systemPrompt },
@@ -34,12 +34,13 @@ export class RequestyHandler implements ApiHandler {
 
 		// @ts-ignore-next-line
 		const stream = await this.client.chat.completions.create({
-			model: modelId,
+			model: model.id,
+			max_tokens: model.info.maxTokens || undefined,
 			messages: openAiMessages,
 			temperature: 0,
 			stream: true,
 			stream_options: { include_usage: true },
-			...(modelId === "openai/o3-mini" ? { reasoning_effort: this.options.o3MiniReasoningEffort || "medium" } : {}),
+			...(model.id === "openai/o3-mini" ? { reasoning_effort: this.options.o3MiniReasoningEffort || "medium" } : {}),
 		})
 
 		for await (const chunk of stream) {
@@ -89,9 +90,11 @@ export class RequestyHandler implements ApiHandler {
 	}
 
 	getModel(): { id: string; info: ModelInfo } {
-		return {
-			id: this.options.requestyModelId ?? "",
-			info: openAiModelInfoSaneDefaults,
+		const modelId = this.options.requestyModelId
+		const modelInfo = this.options.requestyModelInfo
+		if (modelId && modelInfo) {
+			return { id: modelId, info: modelInfo }
 		}
+		return { id: requestyDefaultModelId, info: requestyDefaultModelInfo }
 	}
 }

src/core/webview/ClineProvider.ts

@@ -95,6 +95,7 @@ type GlobalStateKey =
 	| "liteLlmModelId"
 	| "qwenApiLine"
 	| "requestyModelId"
+	| "requestyModelInfo"
 	| "togetherModelId"
 	| "mcpMarketplaceCatalog"
 	| "telemetrySetting"
@@ -103,6 +104,7 @@ export const GlobalFileNames = {
 	apiConversationHistory: "api_conversation_history.json",
 	uiMessages: "ui_messages.json",
 	openRouterModels: "openrouter_models.json",
+	requestyModels: "requesty_models.json",
 	mcpSettings: "cline_mcp_settings.json",
 	clineRules: ".clinerules",
 }
@@ -497,7 +499,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 							}),
 						)
 						// post last cached models in case the call to endpoint fails
-						this.readOpenRouterModels().then((openRouterModels) => {
+						this.readDynamicProviderModels(GlobalFileNames.openRouterModels).then((openRouterModels) => {
 							if (openRouterModels) {
 								this.postMessageToWebview({
 									type: "openRouterModels",
@@ -540,6 +542,34 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 							telemetryService.updateTelemetryState(isOptedIn)
 						})
 
+            
+						// post last cached models in case the call to endpoint fails
+						this.readDynamicProviderModels(GlobalFileNames.requestyModels).then((requestyModels) => {
+							if (requestyModels) {
+								this.postMessageToWebview({
+									type: "requestyModels",
+									requestyModels,
+								})
+							}
+						})
+            
+						// gui relies on model info to be up-to-date to provide the most accurate pricing, so we need to fetch the latest details on launch.
+						// we do this for all users since many users switch between api providers and if they were to switch back to openrouter it would be showing outdated model info if we hadn't retrieved the latest at this point
+						// (see normalizeApiConfiguration > openrouter)
+						this.refreshRequestyModels().then(async (requestyModels) => {
+							if (requestyModels) {
+								// update model info in state (this needs to be done here since we don't want to update state while settings is open, and we may refresh models there)
+								const { apiConfiguration } = await this.getState()
+								if (apiConfiguration.requestyModelId) {
+									await this.updateGlobalState(
+										"requestyModelInfo",
+										requestyModels[apiConfiguration.requestyModelId],
+									)
+									await this.postStateToWebview()
+								}
+							}
+						})
+            
 						break
 					case "newTask":
 						// Code that should run in response to the hello message command
@@ -582,6 +612,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 								deepSeekApiKey,
 								requestyApiKey,
 								requestyModelId,
+								requestyModelInfo,
 								togetherApiKey,
 								togetherModelId,
 								qwenApiKey,
@@ -635,6 +666,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 							await this.updateGlobalState("liteLlmModelId", liteLlmModelId)
 							await this.updateGlobalState("qwenApiLine", qwenApiLine)
 							await this.updateGlobalState("requestyModelId", requestyModelId)
+							await this.updateGlobalState("requestyModelInfo", requestyModelInfo)
 							await this.updateGlobalState("togetherModelId", togetherModelId)
 							if (this.cline) {
 								this.cline.api = buildApiHandler(message.apiConfiguration)
@@ -731,6 +763,9 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 					case "refreshOpenRouterModels":
 						await this.refreshOpenRouterModels()
 						break
+					case "refreshRequestyModels":
+						await this.refreshRequestyModels()
+						break
 					case "refreshOpenAiModels":
 						const { apiConfiguration } = await this.getState()
 						const openAiModels = await this.getOpenAiModels(
@@ -996,6 +1031,10 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 				await this.updateGlobalState("previousModeModelId", apiConfiguration.openRouterModelId)
 				await this.updateGlobalState("previousModeModelInfo", apiConfiguration.openRouterModelInfo)
 				break
+			case "requesty":
+				await this.updateGlobalState("previousModeModelId", apiConfiguration.requestyModelId)
+				await this.updateGlobalState("previousModeModelInfo", apiConfiguration.requestyModelInfo)
+				break
 			case "vscode-lm":
 				await this.updateGlobalState("previousModeModelId", apiConfiguration.vsCodeLmModelSelector)
 				break
@@ -1028,6 +1067,10 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 					await this.updateGlobalState("openRouterModelId", newModelId)
 					await this.updateGlobalState("openRouterModelInfo", newModelInfo)
 					break
+				case "requesty":
+					await this.updateGlobalState("requestyModelId", newModelId)
+					await this.updateGlobalState("requestyModelInfo", newModelInfo)
+					break
 				case "vscode-lm":
 					await this.updateGlobalState("vsCodeLmModelSelector", newModelId)
 					break
@@ -1500,16 +1543,61 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 		return cacheDir
 	}
 
-	async readOpenRouterModels(): Promise<Record<string, ModelInfo> | undefined> {
-		const openRouterModelsFilePath = path.join(await this.ensureCacheDirectoryExists(), GlobalFileNames.openRouterModels)
-		const fileExists = await fileExistsAtPath(openRouterModelsFilePath)
+	async readDynamicProviderModels(filename: string): Promise<Record<string, ModelInfo> | undefined> {
+		const filePath = path.join(await this.ensureCacheDirectoryExists(), filename)
+		const fileExists = await fileExistsAtPath(filePath)
 		if (fileExists) {
-			const fileContents = await fs.readFile(openRouterModelsFilePath, "utf8")
+			const fileContents = await fs.readFile(filePath, "utf8")
 			return JSON.parse(fileContents)
 		}
 		return undefined
 	}
 
+	adjustPriceToMillionTokens(price: any) {
+		if (price) {
+			return parseFloat(price) * 1_000_000
+		}
+		return undefined
+	}
+
+	async refreshRequestyModels() {
+		const requestyModelsFilePath = path.join(await this.ensureCacheDirectoryExists(), GlobalFileNames.requestyModels)
+
+		let models: Record<string, ModelInfo> = {}
+		try {
+			const response = await axios.get("https://router.requesty.ai/v1/models")
+			if (response.data?.data) {
+				for (const model of response.data.data) {
+					const modelInfo: ModelInfo = {
+						maxTokens: model.max_output_tokens,
+						contextWindow: model.context_window,
+						supportsImages: model.supports_images || undefined,
+						supportsComputerUse: model.supports_computer_use || undefined,
+						supportsPromptCache: model.supports_caching || undefined,
+						inputPrice: this.adjustPriceToMillionTokens(model.input_price),
+						outputPrice: this.adjustPriceToMillionTokens(model.output_price),
+						cacheWritesPrice: this.adjustPriceToMillionTokens(model.caching_price),
+						cacheReadsPrice: this.adjustPriceToMillionTokens(model.cached_price),
+						description: model.description,
+					}
+					models[model.id] = modelInfo
+				}
+				await fs.writeFile(requestyModelsFilePath, JSON.stringify(models))
+				console.log("Requesty models fetched and saved", models)
+			} else {
+				console.error("Invalid response from Requesty API")
+			}
+		} catch (error) {
+			console.error("Error fetching Requesty models:", error)
+		}
+
+		await this.postMessageToWebview({
+			type: "requestyModels",
+			requestyModels: models,
+		})
+		return models
+	}
+
 	async refreshOpenRouterModels() {
 		const openRouterModelsFilePath = path.join(await this.ensureCacheDirectoryExists(), GlobalFileNames.openRouterModels)
 
@@ -1544,20 +1632,14 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 			*/
 			if (response.data?.data) {
 				const rawModels = response.data.data
-				const parsePrice = (price: any) => {
-					if (price) {
-						return parseFloat(price) * 1_000_000
-					}
-					return undefined
-				}
 				for (const rawModel of rawModels) {
 					const modelInfo: ModelInfo = {
 						maxTokens: rawModel.top_provider?.max_completion_tokens,
 						contextWindow: rawModel.context_length,
 						supportsImages: rawModel.architecture?.modality?.includes("image"),
 						supportsPromptCache: false,
-						inputPrice: parsePrice(rawModel.pricing?.prompt),
-						outputPrice: parsePrice(rawModel.pricing?.completion),
+						inputPrice: this.adjustPriceToMillionTokens(rawModel.pricing?.prompt),
+						outputPrice: this.adjustPriceToMillionTokens(rawModel.pricing?.completion),
 						description: rawModel.description,
 					}
 
@@ -1858,6 +1940,7 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 			deepSeekApiKey,
 			requestyApiKey,
 			requestyModelId,
+			requestyModelInfo,
 			togetherApiKey,
 			togetherModelId,
 			qwenApiKey,
@@ -1911,6 +1994,7 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 			this.getSecret("deepSeekApiKey") as Promise<string | undefined>,
 			this.getSecret("requestyApiKey") as Promise<string | undefined>,
 			this.getGlobalState("requestyModelId") as Promise<string | undefined>,
+			this.getGlobalState("requestyModelInfo") as Promise<ModelInfo | undefined>,
 			this.getSecret("togetherApiKey") as Promise<string | undefined>,
 			this.getGlobalState("togetherModelId") as Promise<string | undefined>,
 			this.getSecret("qwenApiKey") as Promise<string | undefined>,
@@ -1987,6 +2071,7 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 				deepSeekApiKey,
 				requestyApiKey,
 				requestyModelId,
+				requestyModelInfo,
 				togetherApiKey,
 				togetherModelId,
 				qwenApiKey,

src/shared/ExtensionMessage.ts

@@ -22,6 +22,7 @@ export interface ExtensionMessage {
 		| "invoke"
 		| "partialMessage"
 		| "openRouterModels"
+		| "requestyModels"
 		| "openAiModels"
 		| "mcpServers"
 		| "relinquishControl"
@@ -51,6 +52,7 @@ export interface ExtensionMessage {
 	filePaths?: string[]
 	partialMessage?: ClineMessage
 	openRouterModels?: Record<string, ModelInfo>
+	requestyModels?: Record<string, ModelInfo>
 	openAiModels?: string[]
 	mcpServers?: McpServer[]
 	mcpMarketplaceCatalog?: McpMarketplaceCatalog

src/shared/WebviewMessage.ts

@@ -27,6 +27,7 @@ export interface WebviewMessage {
 		| "openMention"
 		| "cancelTask"
 		| "refreshOpenRouterModels"
+		| "refreshRequestyModels"
 		| "refreshOpenAiModels"
 		| "openMcpSettings"
 		| "restartMcpServer"

src/shared/api.ts

@@ -49,6 +49,7 @@ export interface ApiHandlerOptions {
 	deepSeekApiKey?: string
 	requestyApiKey?: string
 	requestyModelId?: string
+	requestyModelInfo?: ModelInfo
 	togetherApiKey?: string
 	togetherModelId?: string
 	qwenApiKey?: string
@@ -802,6 +803,22 @@ export const liteLlmModelInfoSaneDefaults: ModelInfo = {
 	outputPrice: 0,
 }
 
+// Requesty
+// https://requesty.ai/models
+export const requestyDefaultModelId = "anthropic/claude-3-5-sonnet-latest"
+export const requestyDefaultModelInfo: ModelInfo = {
+	maxTokens: 8192,
+	contextWindow: 200_000,
+	supportsImages: true,
+	supportsComputerUse: false,
+	supportsPromptCache: true,
+	inputPrice: 3.0,
+	outputPrice: 15.0,
+	cacheWritesPrice: 3.75,
+	cacheReadsPrice: 0.3,
+	description: "Anthropic's most intelligent model. Highest level of intelligence and capability.",
+}
+
 // X AI
 // https://docs.x.ai/docs/api-reference
 export type XAIModelId = keyof typeof xaiModels
@@ -880,3 +897,4 @@ export const xaiModels = {
 		description: "X AI's Grok Beta model (legacy) with 131K context window",
 	},
 } as const satisfies Record<string, ModelInfo>
+

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

@@ -214,7 +214,7 @@ const ChatTextArea = forwardRef<HTMLTextAreaElement, ChatTextAreaProps>(
 		},
 		ref,
 	) => {
-		const { filePaths, chatSettings, apiConfiguration, openRouterModels, platform } = useExtensionState()
+		const { filePaths, chatSettings, apiConfiguration, openRouterModels, requestyModels, platform } = useExtensionState()
 		const [isTextAreaFocused, setIsTextAreaFocused] = useState(false)
 		const [gitCommits, setGitCommits] = useState<any[]>([])
 
@@ -635,14 +635,14 @@ const ChatTextArea = forwardRef<HTMLTextAreaElement, ChatTextAreaProps>(
 		// Separate the API config submission logic
 		const submitApiConfig = useCallback(() => {
 			const apiValidationResult = validateApiConfiguration(apiConfiguration)
-			const modelIdValidationResult = validateModelId(apiConfiguration, openRouterModels)
+			const modelIdValidationResult = validateModelId(apiConfiguration, openRouterModels, requestyModels)
 
 			if (!apiValidationResult && !modelIdValidationResult) {
 				vscode.postMessage({ type: "apiConfiguration", apiConfiguration })
 			} else {
 				vscode.postMessage({ type: "getLatestState" })
 			}
-		}, [apiConfiguration, openRouterModels])
+		}, [apiConfiguration, openRouterModels, requestyModels])
 
 		const onModeToggle = useCallback(() => {
 			// if (textAreaDisabled) return
@@ -742,9 +742,6 @@ const ChatTextArea = forwardRef<HTMLTextAreaElement, ChatTextAreaProps>(
 			const unknownModel = "unknown"
 			if (!apiConfiguration) return unknownModel
 			switch (selectedProvider) {
-				case "anthropic":
-				case "openrouter":
-					return `${selectedProvider}:${selectedModelId}`
 				case "openai":
 					return `openai-compat:${selectedModelId}`
 				case "vscode-lm":
@@ -758,7 +755,8 @@ const ChatTextArea = forwardRef<HTMLTextAreaElement, ChatTextAreaProps>(
 				case "litellm":
 					return `${selectedProvider}:${apiConfiguration.liteLlmModelId}`
 				case "requesty":
-					return `${selectedProvider}:${apiConfiguration.requestyModelId}`
+				case "anthropic":
+				case "openrouter":
 				default:
 					return `${selectedProvider}:${selectedModelId}`
 			}