pi-free 2.0.1 → 2.0.4

package/providers/ollama/ollama.ts

@@ -28,6 +28,8 @@ import {
 	applyHidden,
 	getOllamaApiKey,
 	getOllamaShowPaid,
+	loadConfigFile,
+	saveConfig,
 } from "../../config.ts";
 import {
 	BASE_URL_OLLAMA,
@@ -36,11 +38,21 @@ import {
 } from "../../constants.ts";
 import { createLogger } from "../../lib/logger.ts";
 import { registerWithGlobalToggle } from "../../lib/registry.ts";
-import { fetchWithRetry } from "../../lib/util.ts";
+import { fetchWithRetry, fetchWithTimeout } from "../../lib/util.ts";
 import { createReRegister, enhanceWithCI } from "../../provider-helper.ts";
 
 const _logger = createLogger("ollama-cloud");
 
+// =============================================================================
+// Known 403 models (listed but return "access denied" on /v1/chat/completions)
+// These are models that appear in /v1/models but aren't provisioned for chat.
+// Add new IDs here as they surface via /probe-ollama command.
+// =============================================================================
+const OLLAMA_KNOWN_403_MODELS: ReadonlySet<string> = new Set([
+	// Example entries - populate via probe-ollama.mjs
+	// "model-id-that-403s",
+]);
+
 // =============================================================================
 // Fetch + map
 // =============================================================================
@@ -79,12 +91,20 @@ async function fetchOllamaModels(
 	);
 
 	// Filter to chat/text generation models only
-	const chatModels = models.filter((m) => {
-		// Skip embedding-only models (typically have "embed" in name)
-		const name = m.id.toLowerCase();
-		if (name.includes("embed")) return false;
-		return true;
-	});
+	const chatModels = models
+		.filter((m) => {
+			// Skip embedding-only models (typically have "embed" in name)
+			const name = m.id.toLowerCase();
+			if (name.includes("embed")) return false;
+			return true;
+		})
+		// Filter out known 403 models (listed but not provisioned for chat)
+		.filter((m) => {
+			if (OLLAMA_KNOWN_403_MODELS.has(m.id)) {
+				return false;
+			}
+			return true;
+		});
 
 	const result = applyHidden(
 		chatModels.map(
@@ -110,6 +130,7 @@ async function fetchOllamaModels(
 				maxTokens: 4096, // Default, varies by model
 			}),
 		),
+		PROVIDER_OLLAMA,
 	);
 
 	return result;
@@ -169,4 +190,106 @@ export default async function (pi: ExtensionAPI) {
 	_logger.info(
 		`[ollama-cloud] Registered ${initialModels.length} models (usage-based free tier)`,
 	);
+
+	// ── Probe command: test all registered models for 403s ─────────────
+	pi.registerCommand("probe-ollama", {
+		description: "Test all Ollama Cloud models for 403 'access denied' errors",
+		handler: async (_args, ctx) => {
+			if (!apiKey) {
+				ctx.ui.notify("OLLAMA_API_KEY not set", "error");
+				return;
+			}
+
+			const modelsToTest = allModels;
+			ctx.ui.notify(`Probing ${modelsToTest.length} Ollama models…`, "info");
+
+			const notFound: string[] = [];
+			const batchSize = 5;
+
+			for (let i = 0; i < modelsToTest.length; i += batchSize) {
+				const batch = modelsToTest.slice(i, i + batchSize);
+				const results = await Promise.all(
+					batch.map(async (m) => {
+						const ok = await probeOllamaModel(apiKey, m.id);
+						return { id: m.id, ok };
+					}),
+				);
+				for (const r of results) {
+					if (!r.ok) notFound.push(r.id);
+				}
+			}
+
+			if (notFound.length === 0) {
+				ctx.ui.notify("All Ollama models are accessible ✅", "info");
+				return;
+			}
+
+			// Auto-hide 403 models in config (provider-scoped)
+			const config = loadConfigFile();
+			const existingHidden = new Set(config.hidden_models ?? []);
+			for (const id of notFound) existingHidden.add(`${PROVIDER_OLLAMA}/${id}`);
+			saveConfig({ hidden_models: Array.from(existingHidden) });
+
+			// Re-register so hidden models disappear immediately
+			const filtered = await fetchOllamaModels(apiKey);
+			stored.free = filtered;
+			stored.all = filtered;
+			reRegister(filtered);
+
+			ctx.ui.notify(
+				`Found ${notFound.length} broken models (auto-hidden):\n${notFound.join("\n")}`,
+				"warning",
+			);
+		},
+	});
+
+	// ── Status bar for provider selection ─────────────────────────
+
+	pi.on("model_select", (_event, ctx) => {
+		if (_event.model?.provider !== PROVIDER_OLLAMA) {
+			ctx.ui.setStatus(`${PROVIDER_OLLAMA}-status`, undefined);
+			return;
+		}
+
+		const count = allModels.length;
+		ctx.ui.setStatus(
+			`${PROVIDER_OLLAMA}-status`,
+			`ollama: ${count} models (usage-based)`,
+		);
+	});
+}
+
+/**
+ * Probe a single Ollama model with a minimal chat request.
+ * Returns true if the model is accessible (not 403), false if it 403s.
+ */
+async function probeOllamaModel(
+	apiKey: string,
+	modelId: string,
+): Promise<boolean> {
+	try {
+		const response = await fetchWithTimeout(
+			`${BASE_URL_OLLAMA}/chat/completions`,
+			{
+				method: "POST",
+				headers: {
+					Authorization: `Bearer ${apiKey}`,
+					"Content-Type": "application/json",
+					"User-Agent": "pi-free-providers",
+				},
+				body: JSON.stringify({
+					model: modelId,
+					messages: [{ role: "user", content: "hi" }],
+					max_tokens: 1,
+				}),
+			},
+			10000, // 10 second timeout
+		);
+		// 403 = access denied (model not provisioned)
+		// 200/400/401/etc = at least accessible
+		return response.status !== 403;
+	} catch {
+		// Network errors / timeouts are not "access denied"
+		return true;
+	}
 }

package/providers/opencode-session.ts

@@ -1,3 +1,5 @@
+import { randomUUID } from "node:crypto";
+
 /**
  * Shared OpenCode session/request tracking.
  *
@@ -9,10 +11,7 @@ export function createOpenCodeSessionTracker() {
 	let requestCount = 0;
 
 	function generateId(): string {
-		return (
-			Math.random().toString(36).substring(2, 15) +
-			Math.random().toString(36).substring(2, 15)
-		);
+		return randomUUID().replace(/-/g, "");
 	}
 
 	function getSessionId(): string {

package/providers/qwen/qwen-models.ts

@@ -1,101 +1,101 @@
-/**
- * Qwen OAuth model definitions.
- *
- * @deprecated The 1,000 req/day free tier is no longer available. Auth is broken.
- * This provider remains for backward compatibility but should not be used.
- */
-
-import type { ProviderModelConfig } from "@mariozechner/pi-coding-agent";
-import { createLogger } from "../../lib/logger.ts";
-
-const _logger = createLogger("qwen-models");
-
-/**
- * portal.qwen.ai compatibility settings.
- *
- * portal.qwen.ai's OpenAI-compatible API does not support several parameters
- * that the pi framework sends by default.
- */
-export const PORTAL_COMPAT: NonNullable<ProviderModelConfig["compat"]> = {
-	supportsStore: false,
-	supportsDeveloperRole: false,
-	supportsReasoningEffort: false,
-	supportsUsageInStreaming: false,
-	supportsStrictMode: false,
-	maxTokensField: "max_tokens",
-};
-
-/**
- * Fallback model used before OAuth completes or if model discovery fails.
- * The real model ID is resolved dynamically via fetchQwenLiveModels() after auth.
- */
-export const QWEN_FREE_MODELS: ProviderModelConfig[] = [
-	{
-		id: "coder-model",
-		name: "Qwen Coder — DEPRECATED (free tier discontinued)",
-		reasoning: false,
-		input: ["text"],
-		cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
-		contextWindow: 131_072,
-		maxTokens: 16_384,
-		compat: PORTAL_COMPAT,
-	},
-];
-
-/**
- * Fetch Qwen models. Returns static model list for backward compatibility.
- * @deprecated Qwen free tier is discontinued.
- */
-export async function fetchQwenModels(): Promise<ProviderModelConfig[]> {
-	_logger.info("Qwen provider is deprecated, returning placeholder models");
-	return QWEN_FREE_MODELS;
-}
-
-/**
- * Fetch live model list from the Qwen API using the OAuth access token.
- * Returns updated models with real IDs from the server, or the original
- * models unchanged if the request fails.
- */
-export async function fetchQwenLiveModels(
-	baseUrl: string,
-	accessToken: string,
-	templateModels: ProviderModelConfig[],
-): Promise<ProviderModelConfig[]> {
-	try {
-		const response = await fetch(`${baseUrl}/models`, {
-			headers: {
-				Authorization: `Bearer ${accessToken}`,
-				Accept: "application/json",
-			},
-		});
-
-		if (!response.ok) {
-			_logger.info("Qwen /v1/models fetch failed, keeping current model IDs", {
-				status: response.status,
-			});
-			return templateModels;
-		}
-
-		interface ModelEntry {
-			id: string;
-		}
-		const data = (await response.json()) as { data?: ModelEntry[] };
-		const ids: string[] = (data.data ?? [])
-			.map((m: ModelEntry) => m.id)
-			.filter(Boolean);
-
-		_logger.info("Qwen live models discovered", { ids });
-
-		if (ids.length === 0) return templateModels;
-
-		// Prefer a coder model if available, otherwise use the first model
-		const preferred = ids.find((id) => /coder/i.test(id)) ?? ids[0];
-
-		return templateModels.map((m) => ({ ...m, id: preferred }));
-	} catch (err) {
-		_logger.info("Qwen live model fetch error, keeping current model IDs", {
-			error: String(err),
-		});
-		return templateModels;
-	}
-}
+/**
+ * Qwen OAuth model definitions.
+ *
+ * @deprecated The 1,000 req/day free tier is no longer available. Auth is broken.
+ * This provider remains for backward compatibility but should not be used.
+ */
+
+import type { ProviderModelConfig } from "@mariozechner/pi-coding-agent";
+import { createLogger } from "../../lib/logger.ts";
+
+const _logger = createLogger("qwen-models");
+
+/**
+ * portal.qwen.ai compatibility settings.
+ *
+ * portal.qwen.ai's OpenAI-compatible API does not support several parameters
+ * that the pi framework sends by default.
+ */
+export const PORTAL_COMPAT: NonNullable<ProviderModelConfig["compat"]> = {
+	supportsStore: false,
+	supportsDeveloperRole: false,
+	supportsReasoningEffort: false,
+	supportsUsageInStreaming: false,
+	supportsStrictMode: false,
+	maxTokensField: "max_tokens",
+};
+
+/**
+ * Fallback model used before OAuth completes or if model discovery fails.
+ * The real model ID is resolved dynamically via fetchQwenLiveModels() after auth.
+ */
+export const QWEN_FREE_MODELS: ProviderModelConfig[] = [
+	{
+		id: "coder-model",
+		name: "Qwen Coder — DEPRECATED (free tier discontinued)",
+		reasoning: false,
+		input: ["text"],
+		cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+		contextWindow: 131_072,
+		maxTokens: 16_384,
+		compat: PORTAL_COMPAT,
+	},
+];
+
+/**
+ * Fetch Qwen models. Returns static model list for backward compatibility.
+ * @deprecated Qwen free tier is discontinued.
+ */
+export async function fetchQwenModels(): Promise<ProviderModelConfig[]> {
+	_logger.info("Qwen provider is deprecated, returning placeholder models");
+	return QWEN_FREE_MODELS;
+}
+
+/**
+ * Fetch live model list from the Qwen API using the OAuth access token.
+ * Returns updated models with real IDs from the server, or the original
+ * models unchanged if the request fails.
+ */
+export async function fetchQwenLiveModels(
+	baseUrl: string,
+	accessToken: string,
+	templateModels: ProviderModelConfig[],
+): Promise<ProviderModelConfig[]> {
+	try {
+		const response = await fetch(`${baseUrl}/models`, {
+			headers: {
+				Authorization: `Bearer ${accessToken}`,
+				Accept: "application/json",
+			},
+		});
+
+		if (!response.ok) {
+			_logger.info("Qwen /v1/models fetch failed, keeping current model IDs", {
+				status: response.status,
+			});
+			return templateModels;
+		}
+
+		interface ModelEntry {
+			id: string;
+		}
+		const data = (await response.json()) as { data?: ModelEntry[] };
+		const ids: string[] = (data.data ?? [])
+			.map((m: ModelEntry) => m.id)
+			.filter(Boolean);
+
+		_logger.info("Qwen live models discovered", { ids });
+
+		if (ids.length === 0) return templateModels;
+
+		// Prefer a coder model if available, otherwise use the first model
+		const preferred = ids.find((id) => /coder/i.test(id)) ?? ids[0];
+
+		return templateModels.map((m) => ({ ...m, id: preferred }));
+	} catch (err) {
+		_logger.info("Qwen live model fetch error, keeping current model IDs", {
+			error: String(err),
+		});
+		return templateModels;
+	}
+}

package/providers/zenmux/zenmux.ts

@@ -0,0 +1,176 @@
+/**
+ * ZenMux Provider Extension
+ *
+ * Provides access to ZenMux AI gateway - unified API for 200+ models from
+ * OpenAI, Anthropic, Google, and other providers.
+ *
+ * Setup:
+ *   1. Get API key from https://zenmux.ai
+ *   2. Set ZENMUX_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 ZENMUX_API_KEY env var
+ *   # Models appear in /model selector
+ */
+
+import type {
+	ExtensionAPI,
+	ProviderModelConfig,
+} from "@mariozechner/pi-coding-agent";
+import { getZenmuxApiKey, getZenmuxShowPaid } from "../../config.ts";
+import {
+	BASE_URL_ZENMUX,
+	DEFAULT_FETCH_TIMEOUT_MS,
+	PROVIDER_ZENMUX,
+} 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("zenmux");
+
+// =============================================================================
+// Fetch ZenMux models
+// =============================================================================
+
+interface ZenMuxModel {
+	id: string;
+	name?: string;
+	context_length?: number;
+	pricing?: {
+		prompt?: number;
+		completion?: number;
+	};
+}
+
+function isZenmuxReasoningModel(model: Pick<ZenMuxModel, "id" | "name">) {
+	const haystack = `${model.id} ${model.name ?? ""}`.toLowerCase();
+	return isLikelyReasoningModel(model) || haystack.includes("claude");
+}
+
+async function fetchZenmuxModels(
+	apiKey: string,
+): Promise<ProviderModelConfig[]> {
+	_logger.info("[zenmux] Fetching models from ZenMux API...");
+
+	try {
+		const response = await fetchWithRetry(
+			`${BASE_URL_ZENMUX}/models`,
+			{
+				headers: {
+					Authorization: `Bearer ${apiKey}`,
+					"Content-Type": "application/json",
+				},
+			},
+			3,
+			1000,
+			DEFAULT_FETCH_TIMEOUT_MS,
+		);
+
+		if (!response.ok) {
+			throw new Error(`ZenMux API error: ${response.status}`);
+		}
+
+		const data = (await response.json()) as { data?: ZenMuxModel[] };
+		const models = data.data ?? [];
+
+		_logger.info(`[zenmux] Fetched ${models.length} models`);
+
+		return models.map(
+			(m): ProviderModelConfig => ({
+				id: m.id,
+				name: m.name || m.id,
+				reasoning: isZenmuxReasoningModel(m),
+				input: ["text"],
+				cost: {
+					input: m.pricing?.prompt || 0,
+					output: m.pricing?.completion || 0,
+					cacheRead: 0,
+					cacheWrite: 0,
+				},
+				contextWindow: m.context_length || 128000,
+				maxTokens: m.context_length ? Math.floor(m.context_length / 2) : 4096,
+				compat: getProxyModelCompat(m),
+			}),
+		);
+	} catch (error) {
+		_logger.error("[zenmux] Failed to fetch models:", {
+			error: error instanceof Error ? error.message : String(error),
+		});
+		return [];
+	}
+}
+
+// =============================================================================
+// Extension Entry Point
+// =============================================================================
+
+export default async function zenmuxProvider(pi: ExtensionAPI) {
+	const apiKey = getZenmuxApiKey();
+
+	if (!apiKey) {
+		_logger.info(
+			"[zenmux] Skipping - ZENMUX_API_KEY not set (env var or ~/.pi/free.json)",
+		);
+		return;
+	}
+
+	// Fetch models
+	const allModels = await fetchZenmuxModels(apiKey);
+
+	if (allModels.length === 0) {
+		_logger.warn("[zenmux] No models available");
+		return;
+	}
+
+	// Use isFreeModel with allModels for proper detection
+	// ZenMux exposes pricing, so Route A (OR logic) will be used:
+	// FREE if cost=0 OR "free" in name
+	const freeModels = allModels.filter((m) =>
+		isFreeModel({ ...m, provider: PROVIDER_ZENMUX }, allModels),
+	);
+
+	const stored = { free: freeModels, all: allModels };
+
+	_logger.info(
+		`[zenmux] Registered ${allModels.length} models (${freeModels.length} free)`,
+	);
+
+	// Create re-register function
+	const reRegister = createReRegister(pi, {
+		providerId: PROVIDER_ZENMUX,
+		baseUrl: BASE_URL_ZENMUX,
+		apiKey,
+	});
+
+	// Register with global toggle
+	registerWithGlobalToggle(PROVIDER_ZENMUX, stored, reRegister, true);
+
+	// Setup provider with toggle command
+	setupProvider(
+		pi,
+		{
+			providerId: PROVIDER_ZENMUX,
+			initialShowPaid: getZenmuxShowPaid(),
+			reRegister: (models, _stored) => {
+				if (_stored) {
+					stored.free = _stored.free;
+					stored.all = _stored.all;
+				}
+				reRegister(models);
+			},
+		},
+		stored,
+	);
+
+	// Initial registration
+	reRegister(freeModels);
+}