pi-free 2.0.9 → 2.0.10

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "pi-free",
-	"version": "2.0.9",
+	"version": "2.0.10",
 	"type": "module",
 	"description": "AI model providers for Pi with free model filtering. Shows only $0 cost models by default. Supports Kilo (free OAuth), Cline (free), NVIDIA (freemium), ZenMux, CrofAI, Ollama Cloud, and more.",
 	"keywords": [
@@ -41,6 +41,8 @@
 		"LICENSE",
 		"CHANGELOG.md",
 		"banner.svg",
+		"banner.png",
+		"banner.jpg",
 		"scripts/check-extensions.mjs"
 	],
 	"scripts": {

package/providers/crofai/crofai.ts

@@ -1,99 +1,190 @@
-/**
- * CrofAI Provider Extension
- *
- * Provides access to CrofAI API - OpenAI-compatible LLM inference service.
- *
- * Setup:
- *   1. Get API key from https://ai.nahcrof.com
- *   2. Set CROFAI_API_KEY env var or add to ~/.pi/free.json
- *
- * Responds to global free-only filter.
- *
- * Usage:
- *   pi install git:github.com/apmantza/pi-free
- *   # Set CROFAI_API_KEY env var
- *   # Models appear in /model selector
- */
-
-import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
-import { getCrofaiApiKey, getCrofaiShowPaid } from "../../config.ts";
-import { BASE_URL_CROFAI, PROVIDER_CROFAI } from "../../constants.ts";
-import { createLogger } from "../../lib/logger.ts";
-import { isFreeModel, registerWithGlobalToggle } from "../../lib/registry.ts";
-import { fetchOpenAICompatibleModels } from "../../lib/util.ts";
-import { createReRegister, setupProvider } from "../../provider-helper.ts";
-
-const _logger = createLogger("crofai");
-
-// =============================================================================
-// Extension Entry Point
-// =============================================================================
-
-export default async function crofaiProvider(pi: ExtensionAPI) {
-	const apiKey = getCrofaiApiKey();
-
-	if (!apiKey) {
-		_logger.info(
-			"[crofai] Skipping - CROFAI_API_KEY not set (env var or ~/.pi/free.json)",
-		);
-		return;
-	}
-
-	// Fetch models via shared OpenAI-compatible helper
-	const allModels = await fetchOpenAICompatibleModels(
-		"crofai",
-		BASE_URL_CROFAI,
-		apiKey,
-	);
-
-	if (allModels.length === 0) {
-		_logger.warn("[crofai] No models available");
-		return;
-	}
-
-	// Use isFreeModel with allModels for proper detection
-	// CrofAI doesn't expose pricing (all costs are $0), so Route B will be used:
-	// FREE only if "free" in name
-	const freeModels = allModels.filter((m) =>
-		isFreeModel({ ...m, provider: PROVIDER_CROFAI }, allModels),
-	);
-
-	const stored = { free: freeModels, all: allModels };
-
-	_logger.info(
-		`[crofai] Registered ${allModels.length} models (${freeModels.length} free)`,
-	);
-
-	// Create re-register function
-	const reRegister = createReRegister(pi, {
-		providerId: PROVIDER_CROFAI,
-		baseUrl: BASE_URL_CROFAI,
-		apiKey,
-	});
-
-	// Register with global toggle
-	registerWithGlobalToggle(PROVIDER_CROFAI, stored, reRegister, true);
-
-	// Setup provider with toggle command
-	setupProvider(
-		pi,
-		{
-			providerId: PROVIDER_CROFAI,
-			initialShowPaid: getCrofaiShowPaid(),
-			reRegister: (models, _stored) => {
-				if (_stored) {
-					stored.free = _stored.free;
-					stored.all = _stored.all;
-				}
-				reRegister(models);
-			},
-		},
-		stored,
-	);
-
-	// Initial registration — respect persisted toggle state
-	const showPaid = getCrofaiShowPaid();
-	const initialModels =
-		showPaid && stored.all.length > 0 ? stored.all : freeModels;
-	reRegister(initialModels);
-}
+/**
+ * CrofAI Provider Extension
+ *
+ * Provides access to CrofAI API - OpenAI-compatible LLM inference service
+ * hosting DeepSeek, Qwen, and other open-source models.
+ *
+ * NOTE: CrofAI's /v1/models returns per-model context_length, max_completion_tokens,
+ * name, custom_reasoning, and reasoning_effort. Pricing is per-MILLION tokens.
+ *
+ * Setup:
+ *   1. Get API key from https://ai.nahcrof.com
+ *   2. Set CROFAI_API_KEY env var or add to ~/.pi/free.json
+ *
+ * Usage:
+ *   pi install git:github.com/apmantza/pi-free
+ *   # Set CROFAI_API_KEY env var
+ *   # Models appear in /model selector
+ */
+
+import type {
+	ExtensionAPI,
+	ProviderModelConfig,
+} from "@earendil-works/pi-coding-agent";
+import { getCrofaiApiKey, getCrofaiShowPaid } from "../../config.ts";
+import {
+	BASE_URL_CROFAI,
+	DEFAULT_FETCH_TIMEOUT_MS,
+	PROVIDER_CROFAI,
+} 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("crofai");
+
+// =============================================================================
+// Types
+// =============================================================================
+
+interface CrofaiModel {
+	id: string;
+	name?: string;
+	context_length?: number;
+	max_completion_tokens?: number;
+	custom_reasoning?: boolean;
+	reasoning_effort?: boolean;
+	pricing?: {
+		prompt?: string;
+		completion?: string;
+		cache_prompt?: string;
+	};
+}
+
+// =============================================================================
+// Fetch
+// =============================================================================
+
+function parseCrofaiPrice(priceStr: string | undefined): number {
+	if (priceStr === undefined) return 0;
+	const num = Number.parseFloat(priceStr);
+	if (Number.isNaN(num)) return 0;
+	// CrofAI pricing is per-MILLION tokens. Divide to get per-token (Pi convention).
+	return num / 1_000_000;
+}
+
+async function fetchCrofaiModels(
+	apiKey: string,
+): Promise<ProviderModelConfig[]> {
+	const response = await fetchWithRetry(
+		`${BASE_URL_CROFAI}/models`,
+		{
+			headers: {
+				Authorization: `Bearer ${apiKey}`,
+				"Content-Type": "application/json",
+			},
+		},
+		3,
+		1000,
+		DEFAULT_FETCH_TIMEOUT_MS,
+	);
+
+	if (!response.ok) {
+		throw new Error(
+			`CrofAI API error: ${response.status} ${response.statusText}`,
+		);
+	}
+
+	// CrofAI returns { data: [...] }
+	const json = (await response.json()) as {
+		data?: CrofaiModel[];
+	};
+	const models = json.data ?? [];
+
+	_logger.info(`[crofai] Fetched ${models.length} models`);
+
+	return models
+		.filter((m) => m.id)
+		.map((m): ProviderModelConfig => {
+			const name = m.name || m.id;
+			const reasoning =
+				m.custom_reasoning ?? isLikelyReasoningModel({ id: m.id, name });
+
+			return {
+				id: m.id,
+				name,
+				reasoning,
+				input: ["text"],
+				cost: {
+					input: parseCrofaiPrice(m.pricing?.prompt),
+					output: parseCrofaiPrice(m.pricing?.completion),
+					cacheRead: parseCrofaiPrice(m.pricing?.cache_prompt),
+					cacheWrite: 0,
+				},
+				contextWindow: m.context_length ?? 128_000,
+				maxTokens: m.max_completion_tokens ?? 16_384,
+				compat: getProxyModelCompat({ id: m.id, name }),
+			};
+		});
+}
+
+// =============================================================================
+// Extension Entry Point
+// =============================================================================
+
+export default async function crofaiProvider(pi: ExtensionAPI) {
+	const apiKey = getCrofaiApiKey();
+
+	if (!apiKey) {
+		_logger.info(
+			"[crofai] Skipping - CROFAI_API_KEY not set (env var or ~/.pi/free.json)",
+		);
+		return;
+	}
+
+	// Fetch models
+	const allModels = await fetchCrofaiModels(apiKey);
+
+	if (allModels.length === 0) {
+		_logger.warn("[crofai] No models available");
+		return;
+	}
+
+	const freeModels = allModels.filter((m) =>
+		isFreeModel({ ...m, provider: PROVIDER_CROFAI }, allModels),
+	);
+
+	const stored = { free: freeModels, all: allModels };
+
+	_logger.info(
+		`[crofai] Registered ${allModels.length} models (${freeModels.length} free)`,
+	);
+
+	// Create re-register function
+	const reRegister = createReRegister(pi, {
+		providerId: PROVIDER_CROFAI,
+		baseUrl: BASE_URL_CROFAI,
+		apiKey,
+	});
+
+	// Register with global toggle
+	registerWithGlobalToggle(PROVIDER_CROFAI, stored, reRegister, true);
+
+	// Setup provider with toggle command
+	setupProvider(
+		pi,
+		{
+			providerId: PROVIDER_CROFAI,
+			initialShowPaid: getCrofaiShowPaid(),
+			reRegister: (models, _stored) => {
+				if (_stored) {
+					stored.free = _stored.free;
+					stored.all = _stored.all;
+				}
+				reRegister(models);
+			},
+		},
+		stored,
+	);
+
+	// Initial registration — respect persisted toggle state
+	const showPaid = getCrofaiShowPaid();
+	const initialModels =
+		showPaid && stored.all.length > 0 ? stored.all : freeModels;
+	reRegister(initialModels);
+}

package/providers/deepinfra/deepinfra.ts

@@ -1,109 +1,206 @@
-/**
- * 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.).
- *
- * 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, PROVIDER_DEEPINFRA } from "../../constants.ts";
-import { createLogger } from "../../lib/logger.ts";
-import { registerWithGlobalToggle } from "../../lib/registry.ts";
-import { fetchOpenAICompatibleModels } from "../../lib/util.ts";
-import { createReRegister, setupProvider } from "../../provider-helper.ts";
-
-const _logger = createLogger("deepinfra");
-
-// =============================================================================
-// 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 via shared OpenAI-compatible helper
-	const allModels = await fetchOpenAICompatibleModels(
-		"deepinfra",
-		BASE_URL_DEEPINFRA,
-		apiKey,
-		{ cost: { input: 0.3, output: 0.9 } },
-	);
-
-	if (allModels.length === 0) {
-		_logger.warn("[deepinfra] No 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} 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 { 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);
+}