pi-free 2.0.10 → 2.0.12

package/providers/deepinfra/deepinfra.ts

@@ -1,206 +1,208 @@
-/**
- * DeepInfra Provider Extension
- *
- * DeepInfra is an AI inference cloud with an OpenAI-compatible API for
- * 100+ open-source models (Llama, DeepSeek, Mistral, Qwen, Mixtral, etc.).
- *
- * NOTE: DeepInfra's /v1/openai/models buries real model data in a "metadata"
- * field (context_length, max_tokens, pricing, tags). We extract it here.
- * Pricing is per-MILLION tokens.
- *
- * Free tier:
- *   - $5 one-time credit on signup (no credit card)
- *   - ~5M tokens, expires after 90 days
- *   - 60 RPM (varies by model)
- *
- * Paid: pay-per-token after credits exhaust
- *
- * Endpoint:
- *   Chat: https://api.deepinfra.com/v1/openai/chat/completions
- *
- * Setup:
- *   1. Sign up at https://deepinfra.com/ (GitHub or email)
- *   2. Get API key from https://deepinfra.com/dash/api_keys
- *   3. Set DEEPINFRA_TOKEN env var (or add to ~/.pi/free.json)
- *
- * Usage:
- *   pi install git:github.com/apmantza/pi-free
- *   # Set DEEPINFRA_TOKEN env var
- *   # Models appear in /model selector as "deepinfra/meta-llama/..."
- */
-
-import type {
-	ExtensionAPI,
-	ProviderModelConfig,
-} from "@earendil-works/pi-coding-agent";
-import { getDeepinfraApiKey } from "../../config.ts";
-import {
-	BASE_URL_DEEPINFRA,
-	DEFAULT_FETCH_TIMEOUT_MS,
-	PROVIDER_DEEPINFRA,
-} from "../../constants.ts";
-import { createLogger } from "../../lib/logger.ts";
-import {
-	getProxyModelCompat,
-	isLikelyReasoningModel,
-} from "../../lib/provider-compat.ts";
-import { registerWithGlobalToggle } from "../../lib/registry.ts";
-import { fetchWithRetry } from "../../lib/util.ts";
-import { createReRegister, setupProvider } from "../../provider-helper.ts";
-
-const _logger = createLogger("deepinfra");
-
-// =============================================================================
-// Types
-// =============================================================================
-
-interface DeepInfraModel {
-	id: string;
-	metadata?: {
-		context_length?: number;
-		max_tokens?: number;
-		description?: string;
-		pricing?: {
-			input_tokens?: number;
-			output_tokens?: number;
-		};
-		tags?: string[];
-	};
-}
-
-// =============================================================================
-// Fetch
-// =============================================================================
-
-async function fetchDeepinfraModels(
-	apiKey: string,
-): Promise<ProviderModelConfig[]> {
-	const response = await fetchWithRetry(
-		`${BASE_URL_DEEPINFRA}/models`,
-		{
-			headers: {
-				Authorization: `Bearer ${apiKey}`,
-				"Content-Type": "application/json",
-			},
-		},
-		3,
-		1000,
-		DEFAULT_FETCH_TIMEOUT_MS,
-	);
-
-	if (!response.ok) {
-		throw new Error(
-			`DeepInfra API error: ${response.status} ${response.statusText}`,
-		);
-	}
-
-	const json = (await response.json()) as { data?: DeepInfraModel[] };
-	const models = json.data ?? [];
-
-	_logger.info(`[deepinfra] Fetched ${models.length} models`);
-
-	return models
-		.filter((m) => {
-			const id = m.id.toLowerCase();
-			// Filter out non-chat models
-			if (id.includes("embed")) return false;
-			if (id.includes("rerank")) return false;
-			if (id.includes("whisper")) return false;
-			if (id.includes("speech")) return false;
-			return true;
-		})
-		.map((m): ProviderModelConfig => {
-			const meta = m.metadata;
-			const name = m.id.split("/").pop() || m.id;
-
-			// Reasoning: check tags first, fall back to name heuristic
-			const reasoning =
-				meta?.tags?.includes("reasoning") ??
-				isLikelyReasoningModel({ id: m.id, name });
-
-			// Pricing is per-MILLION tokens. Divide to get per-token (Pi convention).
-			const inputCost = (meta?.pricing?.input_tokens ?? 0.3) / 1_000_000;
-			const outputCost = (meta?.pricing?.output_tokens ?? 0.9) / 1_000_000;
-
-			return {
-				id: m.id,
-				name,
-				reasoning,
-				input: ["text"],
-				cost: {
-					input: inputCost,
-					output: outputCost,
-					cacheRead: 0,
-					cacheWrite: 0,
-				},
-				contextWindow: meta?.context_length ?? 128_000,
-				maxTokens: meta?.max_tokens ?? 16_384,
-				compat: getProxyModelCompat({ id: m.id, name }),
-			};
-		});
-}
-
-// =============================================================================
-// Extension Entry Point
-// =============================================================================
-
-export default async function deepinfraProvider(pi: ExtensionAPI) {
-	const apiKey = getDeepinfraApiKey();
-
-	if (!apiKey) {
-		_logger.info(
-			"[deepinfra] Skipping — DEEPINFRA_TOKEN not set. Sign up at https://deepinfra.com/",
-		);
-		return;
-	}
-
-	// Fetch models
-	const allModels = await fetchDeepinfraModels(apiKey);
-
-	if (allModels.length === 0) {
-		_logger.warn("[deepinfra] No chat models available");
-		return;
-	}
-
-	// DeepInfra is a trial credit provider — $5 one-time credit, no truly free models.
-	// All models are marked as paid. When free-only mode is ON, no models are shown.
-	// Toggle free-only OFF to see all models.
-	const freeModels: ProviderModelConfig[] = [];
-	const stored = { free: freeModels, all: allModels };
-
-	_logger.info(
-		`[deepinfra] Registered ${allModels.length} chat models (trial credit, 0 free)`,
-	);
-
-	// Create re-register function
-	const reRegister = createReRegister(pi, {
-		providerId: PROVIDER_DEEPINFRA,
-		baseUrl: BASE_URL_DEEPINFRA,
-		apiKey,
-	});
-
-	// Register with global toggle
-	registerWithGlobalToggle(PROVIDER_DEEPINFRA, stored, reRegister, true);
-
-	// Setup provider with toggle command
-	setupProvider(
-		pi,
-		{
-			providerId: PROVIDER_DEEPINFRA,
-			initialShowPaid: true, // trial credit: default to showing all models
-			tosUrl: "https://deepinfra.com/pricing",
-			reRegister: (models, _stored) => {
-				if (_stored) {
-					stored.free = _stored.free;
-					stored.all = _stored.all;
-				}
-				reRegister(models);
-			},
-		},
-		stored,
-	);
-
-	// Initial registration — DeepInfra is a trial-credit provider,
-	// so always show all models. Users see them immediately on setup.
-	reRegister(allModels);
-}
+/**
+ * DeepInfra Provider Extension
+ *
+ * DeepInfra is an AI inference cloud with an OpenAI-compatible API for
+ * 100+ open-source models (Llama, DeepSeek, Mistral, Qwen, Mixtral, etc.).
+ *
+ * NOTE: DeepInfra's /v1/openai/models buries real model data in a "metadata"
+ * field (context_length, max_tokens, pricing, tags). We extract it here.
+ * Pricing is per-MILLION tokens.
+ *
+ * Free tier:
+ *   - $5 one-time credit on signup (no credit card)
+ *   - ~5M tokens, expires after 90 days
+ *   - 60 RPM (varies by model)
+ *
+ * Paid: pay-per-token after credits exhaust
+ *
+ * Endpoint:
+ *   Chat: https://api.deepinfra.com/v1/openai/chat/completions
+ *
+ * Setup:
+ *   1. Sign up at https://deepinfra.com/ (GitHub or email)
+ *   2. Get API key from https://deepinfra.com/dash/api_keys
+ *   3. Set DEEPINFRA_TOKEN env var (or add to ~/.pi/free.json)
+ *
+ * Usage:
+ *   pi install git:github.com/apmantza/pi-free
+ *   # Set DEEPINFRA_TOKEN env var
+ *   # Models appear in /model selector as "deepinfra/meta-llama/..."
+ */
+
+import type {
+	ExtensionAPI,
+	ProviderModelConfig,
+} from "@earendil-works/pi-coding-agent";
+import { getDeepinfraApiKey } from "../../config.ts";
+import {
+	BASE_URL_DEEPINFRA,
+	DEFAULT_FETCH_TIMEOUT_MS,
+	PROVIDER_DEEPINFRA,
+} from "../../constants.ts";
+import { createLogger } from "../../lib/logger.ts";
+import {
+	getProxyModelCompat,
+	isLikelyReasoningModel,
+} from "../../lib/provider-compat.ts";
+import { isFreeModel, registerWithGlobalToggle } from "../../lib/registry.ts";
+import { fetchWithRetry } from "../../lib/util.ts";
+import { createReRegister, setupProvider } from "../../provider-helper.ts";
+
+const _logger = createLogger("deepinfra");
+
+// =============================================================================
+// Types
+// =============================================================================
+
+interface DeepInfraModel {
+	id: string;
+	metadata?: {
+		context_length?: number;
+		max_tokens?: number;
+		description?: string;
+		pricing?: {
+			input_tokens?: number;
+			output_tokens?: number;
+		};
+		tags?: string[];
+	};
+}
+
+// =============================================================================
+// Fetch
+// =============================================================================
+
+async function fetchDeepinfraModels(
+	apiKey: string,
+): Promise<ProviderModelConfig[]> {
+	const response = await fetchWithRetry(
+		`${BASE_URL_DEEPINFRA}/models`,
+		{
+			headers: {
+				Authorization: `Bearer ${apiKey}`,
+				"Content-Type": "application/json",
+			},
+		},
+		3,
+		1000,
+		DEFAULT_FETCH_TIMEOUT_MS,
+	);
+
+	if (!response.ok) {
+		throw new Error(
+			`DeepInfra API error: ${response.status} ${response.statusText}`,
+		);
+	}
+
+	const json = (await response.json()) as { data?: DeepInfraModel[] };
+	const models = json.data ?? [];
+
+	_logger.info(`[deepinfra] Fetched ${models.length} models`);
+
+	return models
+		.filter((m) => {
+			const id = m.id.toLowerCase();
+			// Filter out non-chat models
+			if (id.includes("embed")) return false;
+			if (id.includes("rerank")) return false;
+			if (id.includes("whisper")) return false;
+			if (id.includes("speech")) return false;
+			return true;
+		})
+		.map((m): ProviderModelConfig => {
+			const meta = m.metadata;
+			const name = m.id.split("/").pop() || m.id;
+
+			// Reasoning: check tags first, fall back to name heuristic
+			const reasoning =
+				meta?.tags?.includes("reasoning") ??
+				isLikelyReasoningModel({ id: m.id, name });
+
+			// Pricing is per-MILLION tokens. Divide to get per-token (Pi convention).
+			const inputCost = (meta?.pricing?.input_tokens ?? 0.3) / 1_000_000;
+			const outputCost = (meta?.pricing?.output_tokens ?? 0.9) / 1_000_000;
+
+			return {
+				id: m.id,
+				name,
+				reasoning,
+				input: ["text"],
+				cost: {
+					input: inputCost,
+					output: outputCost,
+					cacheRead: 0,
+					cacheWrite: 0,
+				},
+				contextWindow: meta?.context_length ?? 128_000,
+				maxTokens: meta?.max_tokens ?? 16_384,
+				compat: getProxyModelCompat({ id: m.id, name }),
+				_pricingKnown: meta?.pricing !== undefined,
+			} as ProviderModelConfig & { _pricingKnown?: boolean };
+		});
+}
+
+// =============================================================================
+// Extension Entry Point
+// =============================================================================
+
+export default async function deepinfraProvider(pi: ExtensionAPI) {
+	const apiKey = getDeepinfraApiKey();
+
+	if (!apiKey) {
+		_logger.info(
+			"[deepinfra] Skipping — DEEPINFRA_TOKEN not set. Sign up at https://deepinfra.com/",
+		);
+		return;
+	}
+
+	// Fetch models
+	const allModels = await fetchDeepinfraModels(apiKey);
+
+	if (allModels.length === 0) {
+		_logger.warn("[deepinfra] No chat models available");
+		return;
+	}
+
+	// DeepInfra is a trial credit provider — $5 one-time credit, no truly free models.
+	// Use isFreeModel for consistent detection across all providers.
+	const freeModels = allModels.filter((m) =>
+		isFreeModel({ ...m, provider: PROVIDER_DEEPINFRA }, allModels),
+	);
+	const stored = { free: freeModels, all: allModels };
+
+	_logger.info(
+		`[deepinfra] Registered ${allModels.length} chat models (trial credit, 0 free)`,
+	);
+
+	// Create re-register function
+	const reRegister = createReRegister(pi, {
+		providerId: PROVIDER_DEEPINFRA,
+		baseUrl: BASE_URL_DEEPINFRA,
+		apiKey,
+	});
+
+	// Register with global toggle
+	registerWithGlobalToggle(PROVIDER_DEEPINFRA, stored, reRegister, true);
+
+	// Setup provider with toggle command
+	setupProvider(
+		pi,
+		{
+			providerId: PROVIDER_DEEPINFRA,
+			initialShowPaid: true, // trial credit: default to showing all models
+			tosUrl: "https://deepinfra.com/pricing",
+			reRegister: (models, _stored) => {
+				if (_stored) {
+					stored.free = _stored.free;
+					stored.all = _stored.all;
+				}
+				reRegister(models);
+			},
+		},
+		stored,
+	);
+
+	// Initial registration — DeepInfra is a trial-credit provider,
+	// so always show all models. Users see them immediately on setup.
+	reRegister(allModels);
+}

package/providers/dynamic-built-in/index.ts

@@ -5,15 +5,18 @@
  * standard /models endpoints when the user has configured an API key.
  *
  * Uses a single generic fetch function instead of per-provider boilerplate.
- * Discovery runs concurrently with 1s timeout per provider, fire-and-forget
- * so extension init never blocks. Pi's built-in defaults serve until
- * discovery completes and replaces them.
+ * Discovery runs concurrently and is awaited by the extension entry point.
+ * Pi only flushes provider registrations after async extension startup, so
+ * dynamic providers must register before setup returns.
  *
  * Providers handled:
  * - mistral (MISTRAL_API_KEY)
  * - groq (GROQ_API_KEY)
  * - cerebras (CEREBRAS_API_KEY)
  * - xai (XAI_API_KEY)
+ * - opencode (OPENCODE_API_KEY from auth.json)
+ * - openrouter (OPENROUTER_API_KEY from auth.json)
+ * - fastrouter (always discovered, FASTROUTER_API_KEY)
  * - huggingface (HF_TOKEN - optional, special-cased API shape)
  *
  * OpenAI is intentionally skipped per user request.
@@ -25,14 +28,22 @@ import type {
 } from "@earendil-works/pi-coding-agent";
 import {
 	getCerebrasApiKey,
+	getFastrouterApiKey,
+	getFastrouterShowPaid,
 	getGroqApiKey,
 	getHfToken,
 	getMistralApiKey,
+	getOpencodeApiKey,
+	getOpencodeShowPaid,
+	getOpenrouterApiKey,
+	getOpenrouterShowPaid,
 	getXaiApiKey,
 } from "../../config.ts";
+import { DEFAULT_FETCH_TIMEOUT_MS } from "../../constants.ts";
 import { createLogger } from "../../lib/logger.ts";
 import { getProxyModelCompat } from "../../lib/provider-compat.ts";
 import { isFreeModel, registerWithGlobalToggle } from "../../lib/registry.ts";
+import { fetchOpenRouterCompatibleModels } from "../model-fetcher.ts";
 import { createToggleState } from "../../lib/toggle-state.ts";
 import { enhanceWithCI } from "../../provider-helper.ts";
 
@@ -68,7 +79,7 @@ async function fetchModelsFromEndpoint(
 
 	const response = await fetch(url, {
 		headers,
-		signal: AbortSignal.timeout(opts.timeoutMs ?? 1_000),
+		signal: AbortSignal.timeout(opts.timeoutMs ?? DEFAULT_FETCH_TIMEOUT_MS),
 	});
 
 	if (!response.ok) {
@@ -101,9 +112,10 @@ async function fetchModelsFromEndpoint(
 				((m.max_tokens ?? m.max_completion_tokens) as number) ??
 				opts.modelDefaults?.maxTokens ??
 				16_384,
+			_pricingKnown: false as boolean | undefined,
 			...opts.modelDefaults,
 			...(opts.compat ? { compat: opts.compat } : {}),
-		} satisfies ProviderModelConfig;
+		} satisfies ProviderModelConfig & { _pricingKnown?: boolean };
 	});
 }
 
@@ -123,7 +135,7 @@ async function fetchHuggingFaceModels(
 
 	const response = await fetch(
 		"https://api-inference.huggingface.co/models?pipeline_tag=text-generation&limit=50",
-		{ headers, signal: AbortSignal.timeout(1_000) },
+		{ headers, signal: AbortSignal.timeout(DEFAULT_FETCH_TIMEOUT_MS) },
 	);
 
 	if (!response.ok) {
@@ -159,11 +171,16 @@ interface DynamicProviderDef {
 	getApiKey: () => string | undefined;
 	baseUrl: string;
 	api: "openai-completions" | "mistral-conversations" | "anthropic-messages";
-	defaultShowPaid: boolean;
+	defaultShowPaid: boolean | (() => boolean);
 	/** Optional per-provider compat overrides (e.g., DeepSeek proxy). */
 	compat?: ProviderModelConfig["compat"];
 	/** Per-model field defaults when the API doesn't expose them. */
 	modelDefaults?: Partial<ProviderModelConfig>;
+	/**
+	 * Custom model fetcher (e.g., OpenRouter uses its own pricing-aware fetcher).
+	 * When not provided, fetchModelsFromEndpoint is used (no pricing, _pricingKnown=false).
+	 */
+	fetchModels?: (apiKey: string) => Promise<ProviderModelConfig[]>;
 }
 
 const DYNAMIC_PROVIDERS: DynamicProviderDef[] = [
@@ -196,6 +213,28 @@ const DYNAMIC_PROVIDERS: DynamicProviderDef[] = [
 		api: "openai-completions",
 		defaultShowPaid: false,
 	},
+	{
+		providerId: "opencode",
+		getApiKey: getOpencodeApiKey,
+		baseUrl: "https://opencode.ai/zen/v1",
+		api: "openai-completions",
+		defaultShowPaid: getOpencodeShowPaid,
+		// OpenCode API returns no pricing — _pricingKnown=false, name-based detection
+	},
+	{
+		providerId: "openrouter",
+		getApiKey: getOpenrouterApiKey,
+		baseUrl: "https://openrouter.ai/api/v1",
+		api: "openai-completions",
+		defaultShowPaid: getOpenrouterShowPaid,
+		// OpenRouter returns full pricing — use its dedicated fetcher
+		fetchModels: (apiKey) =>
+			fetchOpenRouterCompatibleModels({
+				baseUrl: "https://openrouter.ai/api/v1",
+				apiKey,
+				freeOnly: false,
+			}),
+	},
 ];
 
 // =============================================================================
@@ -210,22 +249,27 @@ async function discoverAndRegister(
 	let allModels: ProviderModelConfig[];
 
 	try {
-		allModels = await fetchModelsFromEndpoint({
-			baseUrl: config.baseUrl,
-			apiKey,
-			compat: config.compat,
-			modelDefaults: config.modelDefaults,
-			timeoutMs: 1_000,
-		});
+		if (config.fetchModels) {
+			allModels = await config.fetchModels(apiKey);
+		} else {
+			allModels = await fetchModelsFromEndpoint({
+				baseUrl: config.baseUrl,
+				apiKey,
+				compat: config.compat,
+				modelDefaults: config.modelDefaults,
+				timeoutMs: DEFAULT_FETCH_TIMEOUT_MS,
+			});
+		}
 
 		// Apply DeepSeek proxy compat to matching models
 		allModels = allModels.map((m) => ({
 			...m,
 			compat: getProxyModelCompat(m) ?? m.compat,
 		}));
-	} catch {
+	} catch (error) {
 		_logger.info(
 			`[dynamic] ${config.providerId}: discovery failed, Pi keeps its defaults`,
+			{ error: error instanceof Error ? error.message : String(error) },
 		);
 		return;
 	}
@@ -248,9 +292,10 @@ async function discoverAndRegisterHF(
 	let allModels: ProviderModelConfig[];
 	try {
 		allModels = await fetchHuggingFaceModels(apiKey);
-	} catch {
+	} catch (error) {
 		_logger.info(
 			"[dynamic] huggingface: discovery failed, Pi keeps its defaults",
+			{ error: error instanceof Error ? error.message : String(error) },
 		);
 		return;
 	}
@@ -289,7 +334,10 @@ async function registerProvider(
 	// Toggle state
 	const toggleState = createToggleState({
 		providerId: config.providerId,
-		initialShowPaid: config.defaultShowPaid,
+		initialShowPaid:
+			typeof config.defaultShowPaid === "function"
+				? config.defaultShowPaid()
+				: config.defaultShowPaid,
 		initialModels: { free: freeModels, all: allModels },
 	});
 
@@ -341,16 +389,18 @@ async function registerProvider(
 }
 
 // =============================================================================
-// Main Entry — Fire-and-Forget
+// Main Entry
 // =============================================================================
 
 /**
  * Kick off model discovery for all configured providers.
- * Runs each fetch concurrently with a 1s timeout so the worst-case
- * wall time is ~1s, not `n * 1s`. Extension init never blocks.
+ * Runs each fetch concurrently so startup waits for the slowest provider,
+ * not `n * provider latency`.
  *
- * Pi's built-in defaults serve until discovery completes and this
- * function replaces them via pi.registerProvider().
+ * Pi flushes provider registrations after async extension startup completes,
+ * so this function must await discovery before returning. Otherwise late
+ * pi.registerProvider() calls may not be visible to startup flows such as
+ * `pi --list-models` or the initial model picker.
  */
 export async function setupDynamicBuiltInProviders(
 	pi: ExtensionAPI,
@@ -368,18 +418,41 @@ export async function setupDynamicBuiltInProviders(
 		fetchers.push(discoverAndRegisterHF(pi, hfKey));
 	}
 
+	// FastRouter: always discovered (model listing needs no auth), but Pi
+	// requires a non-empty apiKey/env-var name when replacing a provider's models.
+	// Use the real configured key when present; otherwise register with the env
+	// var name so startup does not fail for users who have not configured it yet.
+	const fastrouterApiKey = getFastrouterApiKey();
+	fetchers.push(
+		discoverAndRegister(
+			pi,
+			{
+				providerId: "fastrouter",
+				getApiKey: getFastrouterApiKey,
+				baseUrl: "https://api.fastrouter.ai/api/v1",
+				api: "openai-completions",
+				defaultShowPaid: getFastrouterShowPaid,
+				fetchModels: () =>
+					fetchOpenRouterCompatibleModels({
+						baseUrl: "https://api.fastrouter.ai/api/v1",
+						apiKey: fastrouterApiKey,
+						freeOnly: false,
+					}),
+			},
+			fastrouterApiKey ?? "FASTROUTER_API_KEY",
+		),
+	);
+
 	if (fetchers.length === 0) return;
 
 	_logger.info(
-		`[dynamic] Kicking off discovery for ${fetchers.length} providers (1s timeout each, concurrent)...`,
+		`[dynamic] Kicking off discovery for ${fetchers.length} providers (concurrent)...`,
 	);
 
-	// Fire-and-forget: log results, never block init
-	void Promise.allSettled(fetchers).then((results) => {
-		const succeeded = results.filter((r) => r.status === "fulfilled").length;
-		const failed = results.filter((r) => r.status === "rejected").length;
-		_logger.info(
-			`[dynamic] Discovery complete: ${succeeded} succeeded, ${failed} failed/rejected`,
-		);
-	});
+	const results = await Promise.allSettled(fetchers);
+	const succeeded = results.filter((r) => r.status === "fulfilled").length;
+	const failed = results.filter((r) => r.status === "rejected").length;
+	_logger.info(
+		`[dynamic] Discovery complete: ${succeeded} succeeded, ${failed} failed/rejected`,
+	);
 }