feat: Support OpenRouter presets

src/api/providers/fetchers/__tests__/modelCache.spec.ts

@@ -145,6 +145,24 @@ describe("getModels with new GetModelsOptions", () => {
 		expect(result).toEqual(mockModels)
 	})
 
+	it("forwards the OpenRouter API key to getOpenRouterModels", async () => {
+		const mockModels = {
+			"openrouter/model": {
+				maxTokens: 8192,
+				contextWindow: 128000,
+				supportsPromptCache: false,
+			},
+		}
+		mockGetOpenRouterModels.mockResolvedValue(mockModels)
+
+		await getModels({ provider: providerIdentifiers.openrouter, apiKey: "openrouter-key" })
+
+		expect(mockGetOpenRouterModels).toHaveBeenCalledWith({
+			openRouterApiKey: "openrouter-key",
+			openRouterBaseUrl: undefined,
+		})
+	})
+
 	it("calls getRequestyModels with optional API key", async () => {
 		const mockModels = {
 			"requesty/model": {
@@ -1125,9 +1143,10 @@ describe("NanoGPT key-scoped cache isolation", () => {
 
 describe("compound cache key derivation across scoping dimensions", () => {
 	// Exercises every branch of getCacheKey via the public getModels() entry point.
-	// litellm is url-scoped AND key-scoped; openrouter is neither, so it hits the bare
-	// provider fallback. The fetcher mocks let us observe the cache key the result is
-	// written under (first arg of the matching memoryCache.set call).
+	// litellm is url-scoped AND key-scoped; openrouter is key-scoped only, so it hits the
+	// key discriminator branch (or the bare provider fallback when no key is supplied). The
+	// fetcher mocks let us observe the cache key the result is written under (first arg of
+	// the matching memoryCache.set call).
 	const mockModels = {
 		"compound/model": {
 			maxTokens: 4096,
@@ -1193,14 +1212,30 @@ describe("compound cache key derivation across scoping dimensions", () => {
 		expect(cacheKey).toBe("litellm:http://host:4000")
 	})
 
-	it("falls back to the bare provider name for providers that are neither url- nor key-scoped", async () => {
-		await getModels({
-			provider: providerIdentifiers.openrouter,
-			apiKey: "ignored-key",
-			baseUrl: "http://ignored:4000",
-		})
+	it("falls back to the bare provider name for a key-scoped provider without an API key", async () => {
+		await getModels({ provider: providerIdentifiers.openrouter })
 		const cacheKey = writtenCacheKey()
 
 		expect(cacheKey).toBe("openrouter")
 	})
+
+	it("includes only the key discriminator for a key-scoped provider without a custom URL", async () => {
+		await getModels({ provider: providerIdentifiers.openrouter, apiKey: "openrouter-key" })
+		const cacheKey = writtenCacheKey()
+
+		expect(cacheKey).toMatch(/^openrouter:[0-9a-f]{8}$/)
+	})
+
+	it("writes different cache keys for two different OpenRouter API keys", async () => {
+		await getModels({ provider: providerIdentifiers.openrouter, apiKey: "key-one" })
+		const firstKey = writtenCacheKey()
+
+		mockSet.mockClear()
+		await getModels({ provider: providerIdentifiers.openrouter, apiKey: "key-two" })
+		const secondKey = writtenCacheKey()
+
+		expect(firstKey).toBeDefined()
+		expect(secondKey).toBeDefined()
+		expect(firstKey).not.toEqual(secondKey)
+	})
 })

src/api/providers/fetchers/__tests__/openrouter.spec.ts

@@ -540,4 +540,205 @@ describe("OpenRouter API", () => {
 			expect(resultWithoutTools.supportedParameters).toContain("max_tokens")
 		})
 	})
+
+	describe("getOpenRouterModels auth and private/preset models", () => {
+		it("omits the Authorization header and skips user/preset endpoints when no API key is provided", async () => {
+			const axios = await import("axios")
+			const getSpy = vi.spyOn(axios.default, "get").mockResolvedValue({ data: { data: [] } })
+
+			await getOpenRouterModels()
+
+			expect(getSpy).toHaveBeenCalledWith(expect.stringContaining("/models"), { headers: undefined })
+			expect(getSpy).not.toHaveBeenCalledWith(expect.stringContaining("/models/user"), expect.anything())
+			expect(getSpy).not.toHaveBeenCalledWith(expect.stringContaining("/presets"), expect.anything())
+
+			getSpy.mockRestore()
+		})
+
+		it("sends the Authorization header to models, user models, and presets when a key is provided", async () => {
+			const axios = await import("axios")
+			const getSpy = vi.spyOn(axios.default, "get").mockResolvedValue({ data: { data: [] } })
+
+			await getOpenRouterModels({ openRouterApiKey: "test-key" })
+
+			const authHeader = expect.objectContaining({ headers: { Authorization: "Bearer test-key" } })
+
+			expect(getSpy).toHaveBeenCalledWith(expect.stringContaining("/models"), authHeader)
+			expect(getSpy).toHaveBeenCalledWith(expect.stringContaining("/models/user"), authHeader)
+			expect(getSpy).toHaveBeenCalledWith(expect.stringContaining("/presets"), authHeader)
+
+			getSpy.mockRestore()
+		})
+
+		it("merges public, user, and preset models into the returned record", async () => {
+			const publicModel = {
+				id: "openai/gpt-4o",
+				name: "GPT-4o",
+				context_length: 128000,
+				pricing: { prompt: "0.000005", completion: "0.000015" },
+			}
+			const userModel = {
+				id: "private/account-model",
+				name: "Account model",
+				context_length: 65536,
+				pricing: { prompt: "0", completion: "0" },
+			}
+
+			const axios = await import("axios")
+			const getSpy = vi
+				.spyOn(axios.default, "get")
+				.mockResolvedValueOnce({ data: { data: [publicModel] } })
+				.mockResolvedValueOnce({ data: { data: [userModel] } })
+				.mockResolvedValueOnce({
+					data: {
+						data: [
+							{
+								id: "preset-1",
+								name: "Flash",
+								slug: "flash",
+								description: null,
+								models: ["openai/gpt-4o"],
+							},
+						],
+					},
+				})
+
+			const models = await getOpenRouterModels({ openRouterApiKey: "test-key" })
+
+			expect(models["openai/gpt-4o"]).toBeDefined()
+			expect(models["private/account-model"]).toBeDefined()
+			const preset = models["@preset/flash"]
+			expect(preset).toBeDefined()
+
+			getSpy.mockRestore()
+		})
+
+		it("derives a single-model preset context window from its underlying model", async () => {
+			const publicModel = {
+				id: "openai/gpt-4o",
+				name: "GPT-4o",
+				context_length: 128000,
+				pricing: { prompt: "0.000005", completion: "0.000015" },
+			}
+
+			const axios = await import("axios")
+			const getSpy = vi
+				.spyOn(axios.default, "get")
+				.mockResolvedValueOnce({ data: { data: [publicModel] } })
+				.mockResolvedValueOnce({ data: { data: [] } })
+				.mockResolvedValueOnce({
+					data: {
+						data: [
+							{
+								id: "preset-1",
+								name: "Flash",
+								slug: "flash",
+								description: null,
+								models: ["openai/gpt-4o"],
+							},
+						],
+					},
+				})
+
+			const models = await getOpenRouterModels({ openRouterApiKey: "test-key" })
+
+			const preset = models["@preset/flash"]
+			expect(preset).toBeDefined()
+			expect(preset).not.toHaveProperty("contextWindow")
+			expect(preset).not.toHaveProperty("description")
+			expect(preset?.supportsPromptCache).toBe(false)
+
+			getSpy.mockRestore()
+		})
+
+		it("derives a multi-model preset context window as the max across its models", async () => {
+			const smallModel = {
+				id: "openai/gpt-4o-mini",
+				name: "GPT-4o mini",
+				context_length: 128000,
+				pricing: { prompt: "0.00000015", completion: "0.0000006" },
+			}
+			const largeModel = {
+				id: "anthropic/claude-3.7-sonnet",
+				name: "Claude 3.7 Sonnet",
+				context_length: 200000,
+				pricing: { prompt: "0.000003", completion: "0.000015" },
+			}
+
+			const axios = await import("axios")
+			const getSpy = vi
+				.spyOn(axios.default, "get")
+				.mockResolvedValueOnce({ data: { data: [smallModel, largeModel] } })
+				.mockResolvedValueOnce({ data: { data: [] } })
+				.mockResolvedValueOnce({
+					data: {
+						data: [
+							{
+								id: "preset-1",
+								name: "Mixed",
+								slug: "mixed",
+								description: null,
+								models: ["openai/gpt-4o-mini", "anthropic/claude-3.7-sonnet"],
+							},
+						],
+					},
+				})
+
+			const models = await getOpenRouterModels({ openRouterApiKey: "test-key" })
+
+			const preset = models["@preset/mixed"]
+			expect(preset).toBeDefined()
+			expect(preset).not.toHaveProperty("contextWindow")
+			expect(preset).not.toHaveProperty("description")
+			expect(preset?.supportsPromptCache).toBe(false)
+
+			getSpy.mockRestore()
+		})
+
+		it("falls back to a conservative context window when a preset's models cannot be resolved", async () => {
+			const axios = await import("axios")
+			const getSpy = vi
+				.spyOn(axios.default, "get")
+				.mockResolvedValueOnce({ data: { data: [] } })
+				.mockResolvedValueOnce({ data: { data: [] } })
+				.mockResolvedValueOnce({
+					data: {
+						data: [
+							{
+								id: "preset-1",
+								name: "Orphan",
+								slug: "orphan",
+								description: null,
+								models: ["unknown/model-not-in-list"],
+							},
+						],
+					},
+				})
+
+			const models = await getOpenRouterModels({ openRouterApiKey: "test-key" })
+
+			const preset = models["@preset/orphan"]
+			expect(preset).toBeDefined()
+			expect(preset).not.toHaveProperty("contextWindow")
+			expect(preset).not.toHaveProperty("description")
+			expect(preset?.supportsPromptCache).toBe(false)
+
+			getSpy.mockRestore()
+		})
+	})
+
+	describe("parseOpenRouterModel accepts preset ids", () => {
+		it("parses an @preset/flash id without rejecting the @ prefix", () => {
+			const result = parseOpenRouterModel({
+				id: "@preset/flash",
+				model: { name: "Flash preset", context_length: 200000 },
+				inputModality: ["text"],
+				outputModality: ["text"],
+				maxTokens: undefined,
+			})
+
+			expect(result.contextWindow).toBe(200000)
+			expect(result.supportsPromptCache).toBe(false)
+		})
+	})
 })

src/api/providers/fetchers/modelCache.ts

@@ -90,6 +90,7 @@ const URL_SCOPED_PROVIDERS: ReadonlySet<RouterName> = new Set([
 // identity -- see the URL_SCOPED_PROVIDERS comment above for why this matters despite caching
 // being skipped for both.
 const KEY_SCOPED_PROVIDERS: ReadonlySet<RouterName> = new Set([
+	providerIdentifiers.openrouter, // Per-key private/preset models (e.g. @preset/*)
 	providerIdentifiers.litellm, // Per-key model allowlists are a first-class LiteLLM proxy feature
 	providerIdentifiers.poe, // Per-account model availability
 	providerIdentifiers.requesty, // Per-account custom model policies
@@ -228,7 +229,10 @@ async function fetchModelsFromProvider(options: GetModelsOptions): Promise<Model
 
 	switch (provider) {
 		case providerIdentifiers.openrouter:
-			models = await getOpenRouterModels()
+			models = await getOpenRouterModels({
+				openRouterApiKey: options.apiKey,
+				openRouterBaseUrl: options.baseUrl,
+			})
 			break
 		case providerIdentifiers.requesty:
 			// Requesty models endpoint requires an API key for per-user custom policies.