Make it work

Author: mrubens

packages/types/src/provider-settings.ts

@@ -19,7 +19,6 @@ import {
 	moonshotModels,
 	openAiNativeModels,
 	qwenCodeModels,
-	rooModels,
 	sambaNovaModels,
 	vertexModels,
 	vscodeLlmModels,
@@ -678,7 +677,7 @@ export const MODELS_BY_PROVIDER: Record<
 		models: Object.keys(openAiNativeModels),
 	},
 	"qwen-code": { id: "qwen-code", label: "Qwen Code", models: Object.keys(qwenCodeModels) },
-	roo: { id: "roo", label: "Roo", models: Object.keys(rooModels) },
+	roo: { id: "roo", label: "Roo Code Cloud", models: [] },
 	sambanova: {
 		id: "sambanova",
 		label: "SambaNova",

packages/types/src/providers/roo.ts

@@ -1,53 +1,13 @@
 import type { ModelInfo } from "../model.js"
 
-export type RooModelId =
-	| "xai/grok-code-fast-1"
-	| "roo/code-supernova-1-million"
-	| "xai/grok-4-fast"
-	| "deepseek/deepseek-chat-v3.1"
+/**
+ * Roo Code Cloud is a dynamic provider - models are loaded from the /v1/models API endpoint.
+ * Default model ID used as fallback when no model is specified.
+ */
+export const rooDefaultModelId = "xai/grok-code-fast-1"
 
-export const rooDefaultModelId: RooModelId = "xai/grok-code-fast-1"
-
-export const rooModels = {
-	"xai/grok-code-fast-1": {
-		maxTokens: 16_384,
-		contextWindow: 262_144,
-		supportsImages: false,
-		supportsPromptCache: true,
-		inputPrice: 0,
-		outputPrice: 0,
-		description:
-			"A reasoning model that is blazing fast and excels at agentic coding, accessible for free through Roo Code Cloud for a limited time. (Note: the free prompts and completions are logged by xAI and used to improve the model.)",
-	},
-	"roo/code-supernova-1-million": {
-		maxTokens: 30_000,
-		contextWindow: 1_000_000,
-		supportsImages: true,
-		supportsPromptCache: true,
-		inputPrice: 0,
-		outputPrice: 0,
-		description:
-			"A versatile agentic coding stealth model with a 1M token context window that supports image inputs, accessible for free through Roo Code Cloud for a limited time. (Note: the free prompts and completions are logged by the model provider and used to improve the model.)",
-	},
-	"xai/grok-4-fast": {
-		maxTokens: 30_000,
-		contextWindow: 2_000_000,
-		supportsImages: false,
-		supportsPromptCache: false,
-		inputPrice: 0,
-		outputPrice: 0,
-		description:
-			"Grok 4 Fast is xAI's latest multimodal model with SOTA cost-efficiency and a 2M token context window. (Note: prompts and completions are logged by xAI and used to improve the model.)",
-		deprecated: true,
-	},
-	"deepseek/deepseek-chat-v3.1": {
-		maxTokens: 16_384,
-		contextWindow: 163_840,
-		supportsImages: false,
-		supportsPromptCache: false,
-		inputPrice: 0,
-		outputPrice: 0,
-		description:
-			"DeepSeek-V3.1 is a large hybrid reasoning model (671B parameters, 37B active). It extends the DeepSeek-V3 base with a two-phase long-context training process, reaching up to 128K tokens, and uses FP8 microscaling for efficient inference.",
-	},
-} as const satisfies Record<string, ModelInfo>
+/**
+ * Empty models object maintained for type compatibility.
+ * All model data comes dynamically from the API.
+ */
+export const rooModels = {} as const satisfies Record<string, ModelInfo>

src/api/providers/__tests__/roo.spec.ts

@@ -1,7 +1,7 @@
 // npx vitest run api/providers/__tests__/roo.spec.ts
 
 import { Anthropic } from "@anthropic-ai/sdk"
-import { rooDefaultModelId, rooModels } from "@roo-code/types"
+import { rooDefaultModelId } from "@roo-code/types"
 
 import { ApiHandlerOptions } from "../../../shared/api"
 
@@ -301,16 +301,19 @@ describe("RooHandler", () => {
 			const modelInfo = handler.getModel()
 			expect(modelInfo.id).toBe(mockOptions.apiModelId)
 			expect(modelInfo.info).toBeDefined()
-			// xai/grok-code-fast-1 is a valid model in rooModels
-			expect(modelInfo.info).toBe(rooModels["xai/grok-code-fast-1"])
+			// Models are loaded dynamically, so we just verify the structure
+			expect(modelInfo.info.maxTokens).toBeDefined()
+			expect(modelInfo.info.contextWindow).toBeDefined()
 		})
 
 		it("should return default model when no model specified", () => {
 			const handlerWithoutModel = new RooHandler({})
 			const modelInfo = handlerWithoutModel.getModel()
 			expect(modelInfo.id).toBe(rooDefaultModelId)
 			expect(modelInfo.info).toBeDefined()
-			expect(modelInfo.info).toBe(rooModels[rooDefaultModelId])
+			// Models are loaded dynamically
+			expect(modelInfo.info.maxTokens).toBeDefined()
+			expect(modelInfo.info.contextWindow).toBeDefined()
 		})
 
 		it("should handle unknown model ID with fallback info", () => {
@@ -320,24 +323,27 @@ describe("RooHandler", () => {
 			const modelInfo = handlerWithUnknownModel.getModel()
 			expect(modelInfo.id).toBe("unknown-model-id")
 			expect(modelInfo.info).toBeDefined()
-			// Should return fallback info for unknown models
-			expect(modelInfo.info.maxTokens).toBe(16_384)
-			expect(modelInfo.info.contextWindow).toBe(262_144)
-			expect(modelInfo.info.supportsImages).toBe(false)
-			expect(modelInfo.info.supportsPromptCache).toBe(true)
-			expect(modelInfo.info.inputPrice).toBe(0)
-			expect(modelInfo.info.outputPrice).toBe(0)
+			// Should return fallback info for unknown models (dynamic models will be merged in real usage)
+			expect(modelInfo.info.maxTokens).toBeDefined()
+			expect(modelInfo.info.contextWindow).toBeDefined()
+			expect(modelInfo.info.supportsImages).toBeDefined()
+			expect(modelInfo.info.supportsPromptCache).toBeDefined()
+			expect(modelInfo.info.inputPrice).toBeDefined()
+			expect(modelInfo.info.outputPrice).toBeDefined()
 		})
 
-		it("should return correct model info for all Roo models", () => {
-			// Test each model in rooModels
-			const modelIds = Object.keys(rooModels) as Array<keyof typeof rooModels>
+		it("should handle any model ID since models are loaded dynamically", () => {
+			// Test with various model IDs - they should all work since models are loaded dynamically
+			const testModelIds = ["xai/grok-code-fast-1", "roo/sonic", "deepseek/deepseek-chat-v3.1"]
 
-			for (const modelId of modelIds) {
+			for (const modelId of testModelIds) {
 				const handlerWithModel = new RooHandler({ apiModelId: modelId })
 				const modelInfo = handlerWithModel.getModel()
 				expect(modelInfo.id).toBe(modelId)
-				expect(modelInfo.info).toBe(rooModels[modelId])
+				expect(modelInfo.info).toBeDefined()
+				// Verify the structure has required fields
+				expect(modelInfo.info.maxTokens).toBeDefined()
+				expect(modelInfo.info.contextWindow).toBeDefined()
 			}
 		})
 	})

src/api/providers/fetchers/roo.ts

@@ -24,33 +24,54 @@ export async function getRooModels(baseUrl: string, apiKey?: string): Promise<Mo
 		}
 
 		// Normalize the URL to ensure proper /v1/models endpoint construction
+		// Remove any trailing /v1 to avoid duplication
 		const urlObj = new URL(baseUrl)
-		urlObj.pathname = urlObj.pathname.replace(/\/+$/, "").replace(/\/+/g, "/") + "/v1/models"
+		let pathname = urlObj.pathname.replace(/\/+$/, "").replace(/\/+/g, "/")
+		// Remove trailing /v1 if present to avoid /v1/v1/models
+		if (pathname.endsWith("/v1")) {
+			pathname = pathname.slice(0, -3)
+		}
+		urlObj.pathname = pathname + "/v1/models"
 		const url = urlObj.href
 
 		// Added timeout to prevent indefinite hanging
 		const response = await axios.get(url, { headers, timeout: 10000 })
 		const models: ModelRecord = {}
-
 		// Process the model info from the response
-		// Expected format: { object: "list", data: [{ id: string, object: "model", created: number, owned_by: string }] }
+		// Expected format: { object: "list", data: [{ id, name, description, context_window, max_tokens, tags, pricing }] }
 		if (response.data && response.data.data && Array.isArray(response.data.data)) {
 			for (const model of response.data.data) {
 				const modelId = model.id
 
 				if (!modelId) continue
 
-				// For Roo Code Cloud, we provide basic model info
-				// The actual detailed model info is stored in the static rooModels definition
-				// This just confirms which models are available
+				// Extract model data from the API response
+				const contextWindow = model.context_window || 262_144
+				const maxTokens = model.max_tokens || 16_384
+				const tags = model.tags || []
+				const pricing = model.pricing || {}
+
+				// Determine if the model supports images based on tags
+				const supportsImages = tags.includes("vision") || tags.includes("image")
+
+				// Parse pricing (API returns strings, convert to numbers)
+				// Handle both direct pricing and cache pricing if available
+				const inputPrice = pricing.input ? parseFloat(pricing.input) : 0
+				const outputPrice = pricing.output ? parseFloat(pricing.output) : 0
+				const cacheReadPrice = pricing.input_cache_read ? parseFloat(pricing.input_cache_read) : undefined
+				const cacheWritePrice = pricing.input_cache_write ? parseFloat(pricing.input_cache_write) : undefined
+
 				models[modelId] = {
-					maxTokens: 16_384, // Default fallback
-					contextWindow: 262_144, // Default fallback
-					supportsImages: false,
-					supportsPromptCache: true,
-					inputPrice: 0,
-					outputPrice: 0,
-					description: `Model available through Roo Code Cloud`,
+					maxTokens,
+					contextWindow,
+					supportsImages,
+					supportsPromptCache: Boolean(cacheReadPrice !== undefined || cacheWritePrice !== undefined),
+					inputPrice,
+					outputPrice,
+					cacheWritesPrice: cacheWritePrice,
+					cacheReadsPrice: cacheReadPrice,
+					description: model.description || model.name || `Model available through Roo Code Cloud`,
+					deprecated: model.deprecated || false,
 				}
 			}
 		} else {

src/api/providers/roo.ts

@@ -1,7 +1,7 @@
 import { Anthropic } from "@anthropic-ai/sdk"
 import OpenAI from "openai"
 
-import { AuthState, rooDefaultModelId, rooModels, type RooModelId, type ModelInfo } from "@roo-code/types"
+import { AuthState, rooDefaultModelId, type ModelInfo } from "@roo-code/types"
 import { CloudService } from "@roo-code/cloud"
 
 import type { ApiHandlerOptions, ModelRecord } from "../../shared/api"
@@ -14,7 +14,7 @@ import { getModels } from "../providers/fetchers/modelCache"
 
 export class RooHandler extends BaseOpenAiCompatibleProvider<string> {
 	private authStateListener?: (state: { state: AuthState }) => void
-	private mergedModels: Record<string, ModelInfo> = rooModels as Record<string, ModelInfo>
+	private mergedModels: Record<string, ModelInfo> = {}
 	private modelsLoaded = false
 
 	constructor(options: ApiHandlerOptions) {
@@ -31,14 +31,14 @@ export class RooHandler extends BaseOpenAiCompatibleProvider<string> {
 		super({
 			...options,
 			providerName: "Roo Code Cloud",
-			baseURL,
+			baseURL: `${baseURL}/v1`, // OpenAI client needs /v1 suffix
 			apiKey: sessionToken || "unauthenticated", // Use a placeholder if no token.
 			defaultProviderModelId: rooDefaultModelId,
-			providerModels: rooModels as Record<string, ModelInfo>,
+			providerModels: {},
 			defaultTemperature: 0.7,
 		})
 
-		// Load dynamic models asynchronously
+		// Load dynamic models asynchronously - pass base URL without /v1
 		this.loadDynamicModels(baseURL, sessionToken).catch((error) => {
 			console.error("[RooHandler] Failed to load dynamic models:", error)
 		})
@@ -122,8 +122,8 @@ export class RooHandler extends BaseOpenAiCompatibleProvider<string> {
 			})
 			this.modelsLoaded = true
 
-			// Merge dynamic models with static models, preferring static model info
-			this.mergedModels = { ...dynamicModels, ...rooModels } as Record<string, ModelInfo>
+			// Use dynamic models directly - no static fallbacks needed
+			this.mergedModels = dynamicModels as Record<string, ModelInfo>
 		} catch (error) {
 			console.error("[RooHandler] Error loading dynamic models:", error)
 			// Keep using static models as fallback

src/core/config/ProviderSettingsManager.ts

@@ -11,7 +11,6 @@ import {
 	DEFAULT_CONSECUTIVE_MISTAKE_LIMIT,
 	getModelId,
 	type ProviderName,
-	type RooModelId,
 } from "@roo-code/types"
 import { TelemetryService } from "@roo-code/telemetry"
 
@@ -24,7 +23,7 @@ type ModelMigrations = {
 
 const MODEL_MIGRATIONS: ModelMigrations = {
 	roo: {
-		"roo/code-supernova": "roo/code-supernova-1-million" as RooModelId,
+		"roo/code-supernova": "roo/code-supernova-1-million",
 	},
 } as const satisfies ModelMigrations
 

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

@@ -218,12 +218,18 @@ describe("webviewMessageHandler - requestRouterModels", () => {
 		})
 
 		// Verify getModels was called for each provider
-		expect(mockGetModels).toHaveBeenCalledWith({ provider: "deepinfra" })
 		expect(mockGetModels).toHaveBeenCalledWith({ provider: "openrouter" })
 		expect(mockGetModels).toHaveBeenCalledWith({ provider: "requesty", apiKey: "requesty-key" })
 		expect(mockGetModels).toHaveBeenCalledWith({ provider: "glama" })
 		expect(mockGetModels).toHaveBeenCalledWith({ provider: "unbound", apiKey: "unbound-key" })
 		expect(mockGetModels).toHaveBeenCalledWith({ provider: "vercel-ai-gateway" })
+		expect(mockGetModels).toHaveBeenCalledWith({ provider: "deepinfra" })
+		expect(mockGetModels).toHaveBeenCalledWith(
+			expect.objectContaining({
+				provider: "roo",
+				baseUrl: expect.any(String),
+			}),
+		)
 		expect(mockGetModels).toHaveBeenCalledWith({
 			provider: "litellm",
 			apiKey: "litellm-key",
@@ -242,6 +248,7 @@ describe("webviewMessageHandler - requestRouterModels", () => {
 				glama: mockModels,
 				unbound: mockModels,
 				litellm: mockModels,
+				roo: mockModels,
 				ollama: {},
 				lmstudio: {},
 				"vercel-ai-gateway": mockModels,
@@ -332,6 +339,7 @@ describe("webviewMessageHandler - requestRouterModels", () => {
 				requesty: mockModels,
 				glama: mockModels,
 				unbound: mockModels,
+				roo: mockModels,
 				litellm: {},
 				ollama: {},
 				lmstudio: {},
@@ -360,6 +368,7 @@ describe("webviewMessageHandler - requestRouterModels", () => {
 			.mockRejectedValueOnce(new Error("Unbound API error")) // unbound
 			.mockResolvedValueOnce(mockModels) // vercel-ai-gateway
 			.mockResolvedValueOnce(mockModels) // deepinfra
+			.mockResolvedValueOnce(mockModels) // roo
 			.mockRejectedValueOnce(new Error("LiteLLM connection failed")) // litellm
 
 		await webviewMessageHandler(mockClineProvider, {
@@ -375,6 +384,7 @@ describe("webviewMessageHandler - requestRouterModels", () => {
 				requesty: {},
 				glama: mockModels,
 				unbound: {},
+				roo: mockModels,
 				litellm: {},
 				ollama: {},
 				lmstudio: {},
@@ -416,6 +426,7 @@ describe("webviewMessageHandler - requestRouterModels", () => {
 			.mockRejectedValueOnce(new Error("Unbound API error")) // unbound
 			.mockRejectedValueOnce(new Error("Vercel AI Gateway error")) // vercel-ai-gateway
 			.mockRejectedValueOnce(new Error("DeepInfra API error")) // deepinfra
+			.mockRejectedValueOnce(new Error("Roo API error")) // roo
 			.mockRejectedValueOnce(new Error("LiteLLM connection failed")) // litellm
 
 		await webviewMessageHandler(mockClineProvider, {
@@ -458,6 +469,20 @@ describe("webviewMessageHandler - requestRouterModels", () => {
 			values: { provider: "deepinfra" },
 		})
 
+		expect(mockClineProvider.postMessageToWebview).toHaveBeenCalledWith({
+			type: "singleRouterModelFetchResponse",
+			success: false,
+			error: "Vercel AI Gateway error",
+			values: { provider: "vercel-ai-gateway" },
+		})
+
+		expect(mockClineProvider.postMessageToWebview).toHaveBeenCalledWith({
+			type: "singleRouterModelFetchResponse",
+			success: false,
+			error: "Roo API error",
+			values: { provider: "roo" },
+		})
+
 		expect(mockClineProvider.postMessageToWebview).toHaveBeenCalledWith({
 			type: "singleRouterModelFetchResponse",
 			success: false,

src/core/webview/webviewMessageHandler.ts

@@ -805,6 +805,16 @@ export const webviewMessageHandler = async (
 						baseUrl: apiConfiguration.deepInfraBaseUrl,
 					},
 				},
+				{
+					key: "roo",
+					options: {
+						provider: "roo",
+						baseUrl: process.env.ROO_CODE_PROVIDER_URL ?? "https://api.roocode.com/proxy",
+						apiKey: CloudService.hasInstance()
+							? CloudService.instance.authService?.getSessionToken()
+							: undefined,
+					},
+				},
 			]
 
 			// Add IO Intelligence if API key is provided.