Remove ModelInfo objects from settings

Author: cte

e2e/src/suite/index.ts

@@ -24,15 +24,6 @@ export async function run() {
 		apiProvider: "openrouter" as const,
 		openRouterApiKey: process.env.OPENROUTER_API_KEY!,
 		openRouterModelId: "google/gemini-2.0-flash-001",
-		openRouterModelInfo: {
-			maxTokens: 8192,
-			contextWindow: 1000000,
-			supportsImages: true,
-			supportsPromptCache: false,
-			inputPrice: 0.1,
-			outputPrice: 0.4,
-			thinking: false,
-		},
 	})
 
 	await vscode.commands.executeCommand("roo-cline.SidebarProvider.focus")

src/activate/registerCommands.ts

@@ -2,6 +2,10 @@ import * as vscode from "vscode"
 import delay from "delay"
 
 import { ClineProvider } from "../core/webview/ClineProvider"
+import { ContextProxy } from "../core/config/ContextProxy"
+
+import { registerHumanRelayCallback, unregisterHumanRelayCallback, handleHumanRelayResponse } from "./humanRelay"
+import { handleNewTask } from "./handleTask"
 
 /**
  * Helper to get the visible ClineProvider instance or log if not found.
@@ -15,9 +19,6 @@ export function getVisibleProviderOrLog(outputChannel: vscode.OutputChannel): Cl
 	return visibleProvider
 }
 
-import { registerHumanRelayCallback, unregisterHumanRelayCallback, handleHumanRelayResponse } from "./humanRelay"
-import { handleNewTask } from "./handleTask"
-
 // Store panel references in both modes
 let sidebarPanel: vscode.WebviewView | undefined = undefined
 let tabPanel: vscode.WebviewPanel | undefined = undefined
@@ -142,7 +143,8 @@ export const openClineInNewTab = async ({ context, outputChannel }: Omit<Registe
 	// deserialize cached webview, but since we use retainContextWhenHidden, we
 	// don't need to use that event).
 	// https://github.com/microsoft/vscode-extension-samples/blob/main/webview-sample/src/extension.ts
-	const tabProvider = new ClineProvider(context, outputChannel, "editor")
+	const contextProxy = await ContextProxy.getInstance(context)
+	const tabProvider = new ClineProvider(context, outputChannel, "editor", contextProxy)
 	const lastCol = Math.max(...vscode.window.visibleTextEditors.map((editor) => editor.viewColumn || 0))
 
 	// Check if there are any visible text editors, otherwise open a new group

src/api/providers/__tests__/glama.test.ts

@@ -207,12 +207,6 @@ describe("GlamaHandler", () => {
 				apiModelId: "openai/gpt-4",
 				glamaModelId: "openai/gpt-4",
 				glamaApiKey: "test-key",
-				glamaModelInfo: {
-					maxTokens: 4096,
-					contextWindow: 8192,
-					supportsImages: true,
-					supportsPromptCache: false,
-				},
 			}
 			const nonAnthropicHandler = new GlamaHandler(nonAnthropicOptions)
 

src/api/providers/__tests__/openrouter.test.ts

@@ -24,7 +24,6 @@ describe("OpenRouterHandler", () => {
 	const mockOptions: ApiHandlerOptions = {
 		openRouterApiKey: "test-key",
 		openRouterModelId: "test-model",
-		openRouterModelInfo: mockOpenRouterModelInfo,
 	}
 
 	beforeEach(() => {
@@ -52,7 +51,6 @@ describe("OpenRouterHandler", () => {
 
 			expect(result).toEqual({
 				id: mockOptions.openRouterModelId,
-				info: mockOptions.openRouterModelInfo,
 				maxTokens: 1000,
 				thinking: undefined,
 				temperature: 0,
@@ -77,11 +75,6 @@ describe("OpenRouterHandler", () => {
 			const handler = new OpenRouterHandler({
 				openRouterApiKey: "test-key",
 				openRouterModelId: "test-model",
-				openRouterModelInfo: {
-					...mockOpenRouterModelInfo,
-					maxTokens: 128_000,
-					thinking: true,
-				},
 				modelMaxTokens: 32_768,
 				modelMaxThinkingTokens: 16_384,
 			})
@@ -188,10 +181,6 @@ describe("OpenRouterHandler", () => {
 		it("adds cache control for supported models", async () => {
 			const handler = new OpenRouterHandler({
 				...mockOptions,
-				openRouterModelInfo: {
-					...mockOpenRouterModelInfo,
-					supportsPromptCache: true,
-				},
 				openRouterModelId: "anthropic/claude-3.5-sonnet",
 			})
 

src/api/providers/__tests__/requesty.test.ts

@@ -14,22 +14,23 @@ describe("RequestyHandler", () => {
 	let handler: RequestyHandler
 	let mockCreate: jest.Mock
 
+	const modelInfo: ModelInfo = {
+		maxTokens: 8192,
+		contextWindow: 200_000,
+		supportsImages: true,
+		supportsComputerUse: true,
+		supportsPromptCache: true,
+		inputPrice: 3.0,
+		outputPrice: 15.0,
+		cacheWritesPrice: 3.75,
+		cacheReadsPrice: 0.3,
+		description:
+			"Claude 3.7 Sonnet is an advanced large language model with improved reasoning, coding, and problem-solving capabilities. It introduces a hybrid reasoning approach, allowing users to choose between rapid responses and extended, step-by-step processing for complex tasks. The model demonstrates notable improvements in coding, particularly in front-end development and full-stack updates, and excels in agentic workflows, where it can autonomously navigate multi-step processes. Claude 3.7 Sonnet maintains performance parity with its predecessor in standard mode while offering an extended reasoning mode for enhanced accuracy in math, coding, and instruction-following tasks. Read more at the [blog post here](https://www.anthropic.com/news/claude-3-7-sonnet)",
+	}
+
 	const defaultOptions: ApiHandlerOptions = {
 		requestyApiKey: "test-key",
 		requestyModelId: "test-model",
-		requestyModelInfo: {
-			maxTokens: 8192,
-			contextWindow: 200_000,
-			supportsImages: true,
-			supportsComputerUse: true,
-			supportsPromptCache: true,
-			inputPrice: 3.0,
-			outputPrice: 15.0,
-			cacheWritesPrice: 3.75,
-			cacheReadsPrice: 0.3,
-			description:
-				"Claude 3.7 Sonnet is an advanced large language model with improved reasoning, coding, and problem-solving capabilities. It introduces a hybrid reasoning approach, allowing users to choose between rapid responses and extended, step-by-step processing for complex tasks. The model demonstrates notable improvements in coding, particularly in front-end development and full-stack updates, and excels in agentic workflows, where it can autonomously navigate multi-step processes. Claude 3.7 Sonnet maintains performance parity with its predecessor in standard mode while offering an extended reasoning mode for enhanced accuracy in math, coding, and instruction-following tasks. Read more at the [blog post here](https://www.anthropic.com/news/claude-3-7-sonnet)",
-		},
 		openAiStreamingEnabled: true,
 		includeMaxTokens: true, // Add this to match the implementation
 	}
@@ -185,7 +186,7 @@ describe("RequestyHandler", () => {
 					],
 					stream: true,
 					stream_options: { include_usage: true },
-					max_tokens: defaultOptions.requestyModelInfo?.maxTokens,
+					max_tokens: modelInfo.maxTokens,
 				})
 			})
 
@@ -279,20 +280,17 @@ describe("RequestyHandler", () => {
 			const result = handler.getModel()
 			expect(result).toEqual({
 				id: defaultOptions.requestyModelId,
-				info: defaultOptions.requestyModelInfo,
+				info: modelInfo,
 			})
 		})
 
 		it("should use sane defaults when no model info provided", () => {
-			handler = new RequestyHandler({
-				...defaultOptions,
-				requestyModelInfo: undefined,
-			})
-
+			handler = new RequestyHandler(defaultOptions)
 			const result = handler.getModel()
+
 			expect(result).toEqual({
 				id: defaultOptions.requestyModelId,
-				info: defaultOptions.requestyModelInfo,
+				info: modelInfo,
 			})
 		})
 	})

src/api/providers/__tests__/unbound.test.ts

@@ -74,14 +74,6 @@ describe("UnboundHandler", () => {
 			apiModelId: "anthropic/claude-3-5-sonnet-20241022",
 			unboundApiKey: "test-api-key",
 			unboundModelId: "anthropic/claude-3-5-sonnet-20241022",
-			unboundModelInfo: {
-				description: "Anthropic's Claude 3 Sonnet model",
-				maxTokens: 8192,
-				contextWindow: 200000,
-				supportsPromptCache: true,
-				inputPrice: 0.01,
-				outputPrice: 0.02,
-			},
 		}
 		handler = new UnboundHandler(mockOptions)
 		mockCreate.mockClear()
@@ -220,14 +212,6 @@ describe("UnboundHandler", () => {
 				apiModelId: "openai/gpt-4o",
 				unboundApiKey: "test-key",
 				unboundModelId: "openai/gpt-4o",
-				unboundModelInfo: {
-					description: "OpenAI's GPT-4",
-					maxTokens: undefined,
-					contextWindow: 128000,
-					supportsPromptCache: true,
-					inputPrice: 0.01,
-					outputPrice: 0.03,
-				},
 			}
 			const nonAnthropicHandler = new UnboundHandler(nonAnthropicOptions)
 
@@ -254,13 +238,6 @@ describe("UnboundHandler", () => {
 				apiModelId: "openai/o3-mini",
 				unboundApiKey: "test-key",
 				unboundModelId: "openai/o3-mini",
-				unboundModelInfo: {
-					maxTokens: undefined,
-					contextWindow: 128000,
-					supportsPromptCache: true,
-					inputPrice: 0.01,
-					outputPrice: 0.03,
-				},
 			}
 			const openaiHandler = new UnboundHandler(openaiOptions)
 
@@ -291,7 +268,6 @@ describe("UnboundHandler", () => {
 			const handlerWithInvalidModel = new UnboundHandler({
 				...mockOptions,
 				unboundModelId: "invalid/model",
-				unboundModelInfo: undefined,
 			})
 			const modelInfo = handlerWithInvalidModel.getModel()
 			expect(modelInfo.id).toBe("anthropic/claude-3-5-sonnet-20241022") // Default model

src/api/providers/fetchers/cache.ts

@@ -0,0 +1,84 @@
+import * as path from "path"
+import fs from "fs/promises"
+
+import NodeCache from "node-cache"
+
+import { ContextProxy } from "../../../core/config/ContextProxy"
+import { getCacheDirectoryPath } from "../../../shared/storagePathManager"
+import { fileExistsAtPath } from "../../../utils/fs"
+import type { ModelInfo } from "../../../schemas"
+import { getOpenRouterModels } from "./openrouter"
+import { getRequestyModels } from "./requesty"
+import { getGlamaModels } from "./glama"
+import { getUnboundModels } from "./unbound"
+
+export type RouterName = "openrouter" | "requesty" | "glama" | "unbound"
+
+export type ModelRecord = Record<string, ModelInfo>
+
+const memoryCache = new NodeCache({
+	stdTTL: 5 * 60,
+	checkperiod: 5 * 60,
+})
+
+async function writeModels(router: RouterName, data: ModelRecord) {
+	const filename = `${router}_models.json`
+	const cacheDir = await getCacheDirectoryPath(ContextProxy.instance.globalStorageUri.fsPath)
+	await fs.writeFile(path.join(cacheDir, filename), JSON.stringify(data))
+}
+
+async function readModels(router: RouterName): Promise<ModelRecord | undefined> {
+	const filename = `${router}_models.json`
+	const cacheDir = await getCacheDirectoryPath(ContextProxy.instance.globalStorageUri.fsPath)
+	const filePath = path.join(cacheDir, filename)
+	const exists = await fileExistsAtPath(filePath)
+	return exists ? JSON.parse(await fs.readFile(filePath, "utf8")) : undefined
+}
+
+/**
+ * Get models from the cache or fetch them from the provider and cache them.
+ * There are two caches:
+ * 1. Memory cache - This is a simple in-memory cache that is used to store models for a short period of time.
+ * 2. File cache - This is a file-based cache that is used to store models for a longer period of time.
+ *
+ * @param router - The router to fetch models from.
+ * @returns The models from the cache or the fetched models.
+ */
+export const getModels = async (router: RouterName): Promise<ModelRecord> => {
+	let models = memoryCache.get<ModelRecord>(router)
+
+	if (models) {
+		return models
+	}
+
+	switch (router) {
+		case "openrouter":
+			models = await getOpenRouterModels()
+			break
+		case "requesty":
+			models = await getRequestyModels()
+			break
+		case "glama":
+			models = await getGlamaModels()
+			break
+		case "unbound":
+			models = await getUnboundModels()
+			break
+	}
+
+	if (Object.keys(models).length > 0) {
+		memoryCache.set(router, models)
+
+		try {
+			await writeModels(router, models)
+		} catch (error) {}
+
+		return models
+	}
+
+	try {
+		models = await readModels(router)
+	} catch (error) {}
+
+	return models ?? {}
+}

src/api/providers/fetchers/glama.ts

@@ -0,0 +1,42 @@
+import axios from "axios"
+
+import { ModelInfo } from "../../../shared/api"
+import { parseApiPrice } from "../../../utils/cost"
+
+export async function getGlamaModels(): Promise<Record<string, ModelInfo>> {
+	const models: Record<string, ModelInfo> = {}
+
+	try {
+		const response = await axios.get("https://glama.ai/api/gateway/v1/models")
+		const rawModels = response.data
+
+		for (const rawModel of rawModels) {
+			const modelInfo: ModelInfo = {
+				maxTokens: rawModel.maxTokensOutput,
+				contextWindow: rawModel.maxTokensInput,
+				supportsImages: rawModel.capabilities?.includes("input:image"),
+				supportsComputerUse: rawModel.capabilities?.includes("computer_use"),
+				supportsPromptCache: rawModel.capabilities?.includes("caching"),
+				inputPrice: parseApiPrice(rawModel.pricePerToken?.input),
+				outputPrice: parseApiPrice(rawModel.pricePerToken?.output),
+				description: undefined,
+				cacheWritesPrice: parseApiPrice(rawModel.pricePerToken?.cacheWrite),
+				cacheReadsPrice: parseApiPrice(rawModel.pricePerToken?.cacheRead),
+			}
+
+			switch (rawModel.id) {
+				case rawModel.id.startsWith("anthropic/"):
+					modelInfo.maxTokens = 8192
+					break
+				default:
+					break
+			}
+
+			models[rawModel.id] = modelInfo
+		}
+	} catch (error) {
+		console.error(`Error fetching Glama models: ${JSON.stringify(error, Object.getOwnPropertyNames(error), 2)}`)
+	}
+
+	return models
+}

src/api/providers/fetchers/index.ts

@@ -0,0 +1,4 @@
+export { getOpenRouterModels } from "./openrouter"
+export { getRequestyModels } from "./requesty"
+export { getGlamaModels } from "./glama"
+export { getUnboundModels } from "./unbound"

src/api/providers/fetchers/openrouter.ts

@@ -46,7 +46,7 @@ const openRouterModelsResponseSchema = z.object({
 
 type OpenRouterModelsResponse = z.infer<typeof openRouterModelsResponseSchema>
 
-export async function getOpenRouterModels(options?: ApiHandlerOptions) {
+export async function getOpenRouterModels(options?: ApiHandlerOptions): Promise<Record<string, ModelInfo>> {
 	const models: Record<string, ModelInfo> = {}
 	const baseURL = options?.openRouterBaseUrl || "https://openrouter.ai/api/v1"