pi-free 2.0.8 → 2.0.10

package/providers/ollama/thinking-levels.ts

@@ -0,0 +1,96 @@
+/**
+ * Thinking level mapping for Ollama Cloud models.
+ *
+ * Maps Pi's thinking levels to Ollama Cloud's OpenAI-compatible
+ * `reasoning_effort` values. The API accepts "none", "low", "medium",
+ * "high", and "max". On simple prompts, "max" can be a no-op over
+ * "high", but on harder prompts it can increase thinking substantially
+ * (e.g. deepseek-v4-pro: ~32k tokens on high vs ~55k on max).
+ *
+ * A `null` value means the level is hidden in Pi's UI.
+ *
+ * Model-specific behavior discovered through testing
+ * (see https://github.com/fgrehm/pi-ollama-cloud/blob/main/docs/think-experiment.md):
+ *   - Most models: all levels work, "none" disables thinking
+ *   - GPT-OSS: no off mode, only low/medium/high
+ *   - Qwen 3.x (non-VL): binary-only (think/nothink) - off works
+ *   - Qwen 3 VL: "none" doesn't disable thinking - off is hidden
+ *   - Kimi K2 Thinking: "none" doesn't disable thinking - off is hidden
+ *   - MiniMax M2.x: "none" doesn't disable thinking - off is hidden
+ *
+ * Reference: https://docs.ollama.com/api/openai-compatibility
+ */
+
+import type { ProviderModelConfig } from "@earendil-works/pi-coding-agent";
+
+export type ThinkingLevelMap = NonNullable<
+	ProviderModelConfig["thinkingLevelMap"]
+>;
+
+/** Default: off/low/medium/high/xhigh with minimal hidden. */
+export const DEFAULT: ThinkingLevelMap = {
+	off: "none",
+	minimal: null,
+	low: "low",
+	medium: "medium",
+	high: "high",
+	xhigh: "max",
+};
+
+/**
+ * GPT-OSS: can't disable thinking, only low/medium/high.
+ * https://ollama.com/library/gpt-oss
+ */
+export const GPT_OSS: ThinkingLevelMap = {
+	off: null,
+	minimal: null,
+	low: "low",
+	medium: "medium",
+	high: "high",
+	xhigh: null,
+};
+
+/**
+ * Qwen 3.x: binary-only (think/nothink), no gradation.
+ * https://docs.ollama.com/capabilities/thinking
+ */
+export const QWEN3: ThinkingLevelMap = {
+	off: "none",
+	minimal: null,
+	low: null,
+	medium: "medium",
+	high: null,
+	xhigh: null,
+};
+
+/**
+ * "none" doesn't disable thinking - off is hidden.
+ * Used by kimi-k2-thinking, minimax family, qwen3-vl.
+ */
+export const NO_OFF: ThinkingLevelMap = {
+	off: null,
+	minimal: null,
+	low: "low",
+	medium: "medium",
+	high: "high",
+	xhigh: "max",
+};
+
+/**
+ * Resolve the thinking level map for a model.
+ * Matches by model ID prefix (case-sensitive, checks first chars).
+ */
+export function resolveThinkingMap(
+	id: string,
+	capabilities: string[],
+): ThinkingLevelMap | undefined {
+	if (!capabilities.includes("thinking")) return undefined;
+
+	if (id.startsWith("gpt-oss")) return GPT_OSS;
+	if (id.startsWith("qwen3-vl")) return NO_OFF;
+	if (id.startsWith("qwen3")) return QWEN3;
+	if (id === "kimi-k2-thinking") return NO_OFF;
+	if (id.startsWith("minimax")) return NO_OFF;
+
+	return DEFAULT;
+}

package/providers/qwen/qwen-auth.ts

@@ -16,7 +16,7 @@ import crypto from "node:crypto";
 import type {
 	OAuthCredentials,
 	OAuthLoginCallbacks,
-} from "@mariozechner/pi-ai";
+} from "@earendil-works/pi-ai";
 import { createLogger } from "../../lib/logger.ts";
 import { openBrowser } from "../../lib/open-browser.ts";
 

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 "@earendil-works/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/qwen/qwen.ts

@@ -10,8 +10,8 @@
  * 1,000 free API calls/day — run /login qwen to authenticate.~~
  */
 
-import type { Api, Model, OAuthCredentials } from "@mariozechner/pi-ai";
-import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
+import type { Api, Model, OAuthCredentials } from "@earendil-works/pi-ai";
+import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
 import { PROVIDER_QWEN, URL_QWEN_TOS } from "../../constants.ts";
 import { createLogger } from "../../lib/logger.ts";
 import { logWarning } from "../../lib/util.ts";

package/providers/sambanova/sambanova.ts

@@ -27,7 +27,7 @@
  *   # Models appear in /model selector as "sambanova/Meta-Llama-3.3-70B-Instruct"
  */
 
-import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
+import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
 import { getSambanovaApiKey, getSambanovaShowPaid } from "../../config.ts";
 import { BASE_URL_SAMBANOVA, PROVIDER_SAMBANOVA } from "../../constants.ts";
 import { createLogger } from "../../lib/logger.ts";

package/providers/together/together.ts

@@ -0,0 +1,197 @@
+/**
+ * Together AI Provider Extension
+ *
+ * Together AI provides fast inference on 200+ open-source models through an
+ * OpenAI-compatible API. Known for Llama, DeepSeek, Qwen, Mixtral, and other
+ * popular models at competitive per-token pricing.
+ *
+ * Free tier:
+ *   - $1 one-time credit on signup (no credit card)
+ *   - 60 RPM, 600 RPD (varies by model)
+ *   - Sign up at https://api.together.ai/
+ *
+ * Paid: pay-per-token after credits exhaust
+ *
+ * NOTE: Together AI's /v1/models returns a plain array (not { data: [...] }),
+ * uses per-million-token pricing (not per-token), and includes a "type" field
+ * we use to filter to chat models only.
+ *
+ * Endpoint:
+ *   Chat: https://api.together.xyz/v1/chat/completions
+ *
+ * Setup:
+ *   1. Sign up at https://api.together.ai/
+ *   2. Get API key from https://api.together.ai/settings/api-keys
+ *   3. Set TOGETHER_AI_API_KEY env var (or add to ~/.pi/free.json)
+ *
+ * Usage:
+ *   pi install git:github.com/apmantza/pi-free
+ *   # Set TOGETHER_AI_API_KEY env var
+ *   # Models appear in /model selector as "together/deepseek-ai/..."
+ */
+
+import type {
+	ExtensionAPI,
+	ProviderModelConfig,
+} from "@earendil-works/pi-coding-agent";
+import { getTogetherApiKey, getTogetherShowPaid } from "../../config.ts";
+import {
+	BASE_URL_TOGETHER,
+	DEFAULT_FETCH_TIMEOUT_MS,
+	PROVIDER_TOGETHER,
+} 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("together");
+
+// =============================================================================
+// Types
+// =============================================================================
+
+interface TogetherModel {
+	id: string;
+	display_name?: string;
+	type?: string;
+	context_length?: number;
+	pricing?: {
+		input?: number;
+		output?: number;
+		cached_input?: number;
+	};
+}
+
+// =============================================================================
+// Fetch
+// =============================================================================
+
+async function fetchTogetherModels(
+	apiKey: string,
+): Promise<ProviderModelConfig[]> {
+	const response = await fetchWithRetry(
+		`${BASE_URL_TOGETHER}/models`,
+		{
+			headers: {
+				Authorization: `Bearer ${apiKey}`,
+				"Content-Type": "application/json",
+			},
+		},
+		3,
+		1000,
+		DEFAULT_FETCH_TIMEOUT_MS,
+	);
+
+	if (!response.ok) {
+		throw new Error(
+			`Together AI API error: ${response.status} ${response.statusText}`,
+		);
+	}
+
+	// Together AI returns a plain array (not { data: [...] })
+	const models = (await response.json()) as TogetherModel[];
+
+	_logger.info(`[together] Fetched ${models.length} models`);
+
+	return models
+		.filter((m) => m.type === "chat" && m.id && !m.id.includes("embed"))
+		.map((m): ProviderModelConfig => {
+			const name = m.display_name || m.id.split("/").pop() || m.id;
+
+			// Together AI pricing is per-MILLION tokens.
+			// Divide by 1_000_000 to get per-token cost (Pi convention).
+			const inputCost = (m.pricing?.input ?? 0) / 1_000_000;
+			const outputCost = (m.pricing?.output ?? 0) / 1_000_000;
+			const cacheReadCost = (m.pricing?.cached_input ?? 0) / 1_000_000;
+
+			return {
+				id: m.id,
+				name,
+				reasoning: isLikelyReasoningModel({ id: m.id, name }),
+				input: ["text"],
+				cost: {
+					input: inputCost,
+					output: outputCost,
+					cacheRead: cacheReadCost,
+					cacheWrite: 0,
+				},
+				contextWindow: m.context_length ?? 128_000,
+				maxTokens: 16_384,
+				compat: getProxyModelCompat({ id: m.id, name }),
+			};
+		});
+}
+
+// =============================================================================
+// Extension Entry Point
+// =============================================================================
+
+export default async function togetherProvider(pi: ExtensionAPI) {
+	const apiKey = getTogetherApiKey();
+
+	if (!apiKey) {
+		_logger.info(
+			"[together] Skipping — TOGETHER_AI_API_KEY not set. Sign up at https://api.together.ai/",
+		);
+		return;
+	}
+
+	// Fetch models
+	const allModels = await fetchTogetherModels(apiKey);
+
+	if (allModels.length === 0) {
+		_logger.warn("[together] No chat models available");
+		return;
+	}
+
+	// Together AI is a pay-per-token provider with $1 trial credit.
+	// Zero-cost models (if any) are marked free; all others are paid.
+	const freeModels = allModels.filter(
+		(m) =>
+			m.cost.input === 0 &&
+			m.cost.output === 0 &&
+			m.cost.cacheRead === 0 &&
+			m.cost.cacheWrite === 0,
+	);
+	const stored = { free: freeModels, all: allModels };
+
+	_logger.info(
+		`[together] ${allModels.length} chat models (${freeModels.length} free)`,
+	);
+
+	// Create re-register function
+	const reRegister = createReRegister(pi, {
+		providerId: PROVIDER_TOGETHER,
+		baseUrl: BASE_URL_TOGETHER,
+		apiKey,
+	});
+
+	// Register with global toggle
+	registerWithGlobalToggle(PROVIDER_TOGETHER, stored, reRegister, true);
+
+	// Setup provider with toggle command
+	setupProvider(
+		pi,
+		{
+			providerId: PROVIDER_TOGETHER,
+			initialShowPaid: getTogetherShowPaid(),
+			tosUrl: "https://api.together.ai/",
+			reRegister: (models, _stored) => {
+				if (_stored) {
+					stored.free = _stored.free;
+					stored.all = _stored.all;
+				}
+				reRegister(models);
+			},
+		},
+		stored,
+	);
+
+	// Initial registration — show all models (trial credit provider)
+	reRegister(stored.all);
+}