feat: Add provider-scoped router model fetches in frontend

Implements intelligent provider filtering in the frontend to request
only the models needed for the currently selected provider. This
dramatically reduces memory usage by avoiding unnecessary fetches
and smaller cache footprints.

- Add DYNAMIC_ROUTER_PROVIDERS set to identify providers needing models
- Disable useRouterModels for static providers (anthropic, openai-native, etc.)
- Pass provider-specific filters for dynamic providers
- Add observability logging to track request/response patterns
- Update tests for new behavior

Works with backend filtering (PR #8916) to achieve end-to-end
payload reduction.

webview-ui/src/components/ui/hooks/__tests__/useSelectedModel.spec.ts

@@ -291,7 +291,7 @@ describe("useSelectedModel", () => {
 	})
 
 	describe("loading and error states", () => {
-		it("should return loading state when router models are loading", () => {
+		it("should NOT set loading when router models are loading but provider is static (anthropic)", () => {
 			mockUseRouterModels.mockReturnValue({
 				data: undefined,
 				isLoading: true,
@@ -307,10 +307,11 @@ describe("useSelectedModel", () => {
 			const wrapper = createWrapper()
 			const { result } = renderHook(() => useSelectedModel(), { wrapper })
 
-			expect(result.current.isLoading).toBe(true)
+			// With static provider default (anthropic), useSelectedModel gates router fetches, so loading should be false
+			expect(result.current.isLoading).toBe(false)
 		})
 
-		it("should return loading state when open router model providers are loading", () => {
+		it("should NOT set loading when openrouter provider metadata is loading but provider is static (anthropic)", () => {
 			mockUseRouterModels.mockReturnValue({
 				data: { openrouter: {}, requesty: {}, glama: {}, unbound: {}, litellm: {}, "io-intelligence": {} },
 				isLoading: false,
@@ -326,10 +327,11 @@ describe("useSelectedModel", () => {
 			const wrapper = createWrapper()
 			const { result } = renderHook(() => useSelectedModel(), { wrapper })
 
-			expect(result.current.isLoading).toBe(true)
+			// With static provider default (anthropic), openrouter providers are irrelevant, so loading should be false
+			expect(result.current.isLoading).toBe(false)
 		})
 
-		it("should return error state when either hook has an error", () => {
+		it("should NOT set error when hooks error but provider is static (anthropic)", () => {
 			mockUseRouterModels.mockReturnValue({
 				data: undefined,
 				isLoading: false,
@@ -345,7 +347,8 @@ describe("useSelectedModel", () => {
 			const wrapper = createWrapper()
 			const { result } = renderHook(() => useSelectedModel(), { wrapper })
 
-			expect(result.current.isError).toBe(true)
+			// Error from gated routerModels should not bubble for static provider default
+			expect(result.current.isError).toBe(false)
 		})
 	})
 

webview-ui/src/components/ui/hooks/useRouterModels.ts

@@ -5,8 +5,16 @@ import { ExtensionMessage } from "@roo/ExtensionMessage"
 
 import { vscode } from "@src/utils/vscode"
 
-const getRouterModels = async () =>
+type UseRouterModelsOptions = {
+	providers?: string[] // subset filter (e.g. ["roo"])
+	enabled?: boolean // gate fetching entirely
+}
+
+let __routerModelsRequestCount = 0
+
+const getRouterModels = async (providers?: string[]) =>
 	new Promise<RouterModels>((resolve, reject) => {
+		const requestId = ++__routerModelsRequestCount
 		const cleanup = () => {
 			window.removeEventListener("message", handler)
 		}
@@ -24,6 +32,10 @@ const getRouterModels = async () =>
 				cleanup()
 
 				if (message.routerModels) {
+					const keys = Object.keys(message.routerModels || {})
+					console.debug(
+						`[useRouterModels] response #${requestId} providers=${JSON.stringify(providers || "all")} keys=${keys.join(",")}`,
+					)
 					resolve(message.routerModels)
 				} else {
 					reject(new Error("No router models in response"))
@@ -32,7 +44,21 @@ const getRouterModels = async () =>
 		}
 
 		window.addEventListener("message", handler)
-		vscode.postMessage({ type: "requestRouterModels" })
+		console.debug(
+			`[useRouterModels] request #${requestId} providers=${JSON.stringify(providers && providers.length ? providers : "all")}`,
+		)
+		if (providers && providers.length > 0) {
+			vscode.postMessage({ type: "requestRouterModels", values: { providers } })
+		} else {
+			vscode.postMessage({ type: "requestRouterModels" })
+		}
 	})
 
-export const useRouterModels = () => useQuery({ queryKey: ["routerModels"], queryFn: getRouterModels })
+export const useRouterModels = (opts: UseRouterModelsOptions = {}) => {
+	const providers = opts.providers && opts.providers.length ? [...opts.providers] : undefined
+	return useQuery({
+		queryKey: ["routerModels", providers?.slice().sort().join(",") || "all"],
+		queryFn: () => getRouterModels(providers),
+		enabled: opts.enabled !== false,
+	})
+}

webview-ui/src/components/ui/hooks/useSelectedModel.ts

@@ -67,30 +67,56 @@ import { useOpenRouterModelProviders } from "./useOpenRouterModelProviders"
 import { useLmStudioModels } from "./useLmStudioModels"
 import { useOllamaModels } from "./useOllamaModels"
 
+const DYNAMIC_ROUTER_PROVIDERS = new Set<ProviderName>([
+	"openrouter",
+	"vercel-ai-gateway",
+	"litellm",
+	"deepinfra",
+	"io-intelligence",
+	"requesty",
+	"unbound",
+	"glama",
+	"roo",
+])
+
 export const useSelectedModel = (apiConfiguration?: ProviderSettings) => {
 	const provider = apiConfiguration?.apiProvider || "anthropic"
 	const openRouterModelId = provider === "openrouter" ? apiConfiguration?.openRouterModelId : undefined
 	const lmStudioModelId = provider === "lmstudio" ? apiConfiguration?.lmStudioModelId : undefined
 	const ollamaModelId = provider === "ollama" ? apiConfiguration?.ollamaModelId : undefined
 
-	const routerModels = useRouterModels()
+	// Only fetch router models for dynamic router providers we actually need
+	const shouldFetchRouterModels = DYNAMIC_ROUTER_PROVIDERS.has(provider as ProviderName)
+	const routerModels = useRouterModels({
+		providers: shouldFetchRouterModels ? [provider] : undefined,
+		enabled: shouldFetchRouterModels, // disable entirely for static providers
+	})
+
 	const openRouterModelProviders = useOpenRouterModelProviders(openRouterModelId)
 	const lmStudioModels = useLmStudioModels(lmStudioModelId)
 	const ollamaModels = useOllamaModels(ollamaModelId)
 
+	// Compute readiness only for the data actually needed for the selected provider
+	const needRouterModels = shouldFetchRouterModels
+	const needOpenRouterProviders = provider === "openrouter"
+	const needLmStudio = typeof lmStudioModelId !== "undefined"
+	const needOllama = typeof ollamaModelId !== "undefined"
+
+	const isReady =
+		(!needLmStudio || typeof lmStudioModels.data !== "undefined") &&
+		(!needOllama || typeof ollamaModels.data !== "undefined") &&
+		(!needRouterModels || typeof routerModels.data !== "undefined") &&
+		(!needOpenRouterProviders || typeof openRouterModelProviders.data !== "undefined")
+
 	const { id, info } =
-		apiConfiguration &&
-		(typeof lmStudioModelId === "undefined" || typeof lmStudioModels.data !== "undefined") &&
-		(typeof ollamaModelId === "undefined" || typeof ollamaModels.data !== "undefined") &&
-		typeof routerModels.data !== "undefined" &&
-		typeof openRouterModelProviders.data !== "undefined"
+		apiConfiguration && isReady
 			? getSelectedModel({
 					provider,
 					apiConfiguration,
-					routerModels: routerModels.data,
-					openRouterModelProviders: openRouterModelProviders.data,
-					lmStudioModels: lmStudioModels.data,
-					ollamaModels: ollamaModels.data,
+					routerModels: (routerModels.data || ({} as RouterModels)) as RouterModels,
+					openRouterModelProviders: (openRouterModelProviders.data || {}) as Record<string, ModelInfo>,
+					lmStudioModels: (lmStudioModels.data || undefined) as ModelRecord | undefined,
+					ollamaModels: (ollamaModels.data || undefined) as ModelRecord | undefined,
 				})
 			: { id: anthropicDefaultModelId, info: undefined }
 
@@ -99,13 +125,15 @@ export const useSelectedModel = (apiConfiguration?: ProviderSettings) => {
 		id,
 		info,
 		isLoading:
-			routerModels.isLoading ||
-			openRouterModelProviders.isLoading ||
-			(apiConfiguration?.lmStudioModelId && lmStudioModels!.isLoading),
+			(needRouterModels && routerModels.isLoading) ||
+			(needOpenRouterProviders && openRouterModelProviders.isLoading) ||
+			(needLmStudio && lmStudioModels!.isLoading) ||
+			(needOllama && ollamaModels!.isLoading),
 		isError:
-			routerModels.isError ||
-			openRouterModelProviders.isError ||
-			(apiConfiguration?.lmStudioModelId && lmStudioModels!.isError),
+			(needRouterModels && routerModels.isError) ||
+			(needOpenRouterProviders && openRouterModelProviders.isError) ||
+			(needLmStudio && lmStudioModels!.isError) ||
+			(needOllama && ollamaModels!.isError),
 	}
 }