pi-free 2.0.14 → 2.0.15

package/CHANGELOG.md

@@ -7,6 +7,20 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.0.15] - 2026-06-02
+
+### Fixed
+
+- **Qwen 3.7 reasoning compat** — `qwen/qwen3.7-max` on Cline/OpenRouter uses DeepSeek-style `reasoning_content` format. Added `DEEPSEEK_PROXY_COMPAT` so Pi preserves and replays reasoning tokens correctly, preventing plan-mode hangs ([#213](https://github.com/apmantza/pi-free/pull/213)).
+
+- **Kimi K2.6 reasoning compat** — Kimi models on NVIDIA/OpenRouter need `requiresReasoningContentOnAssistantMessages: true` to correctly replay reasoning tokens in assistant messages. Without it, the model gets stuck when trying to call tools or produce output after thinking. Refs [earendil-works/pi#5309](https://github.com/earendil-works/pi/issues/5309) ([#213](https://github.com/apmantza/pi-free/pull/213)).
+
+- **MiniMax reasoning compat** — MiniMax M3 and other MiniMax models now have full DeepSeek-style compat (`thinkingFormat: "deepseek"`, `requiresReasoningContentOnAssistantMessages: true`). Previously, models marked `reasoning: true` without `thinkingFormat` caused Pi to enter plan mode but couldn't parse the reasoning tokens, resulting in hangs ([#212](https://github.com/apmantza/pi-free/pull/212), [#213](https://github.com/apmantza/pi-free/pull/213)).
+
+### Added
+
+- **`/probe-routeway` command** — Tests each Routeway model with a minimal chat request and auto-hides models that return 5xx or 404 errors. Runs lazily on first `session_start` with 24h probe cache TTL. Follows the same pattern as `/probe-nvidia` ([#213](https://github.com/apmantza/pi-free/pull/213)).
+
 ## [2.0.14] - 2026-06-02
 
 ### Added
@@ -19,6 +33,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - **`_pricingKnown` / `_freeKnown` authoritatve flag** — Providers can now signal whether pricing data is authoritative via `_pricingKnown`. When `false`, `isFreeModel` falls back to name-based detection. Kilo's `isFree` API flag now flows through as `_freeKnown` ([#209](https://github.com/apmantza/pi-free/pull/209)).
 
+- **MiniMax reasoning compat** — MiniMax M3 and other MiniMax models now have `supportsReasoningEffort: true` compat settings. Previously, models marked `reasoning: true` without compat caused Pi to enter plan mode without knowing the thinking format, resulting in hangs.
+
 ## [2.0.13] - 2026-05-21
 
 ### Added

package/README.md

@@ -467,10 +467,11 @@ Each provider has toggle commands to switch between free and all models:
 
 Test models for 404/403 errors and auto-hide broken ones:
 
-| Command         | What it does                                                |
-| --------------- | ----------------------------------------------------------- |
-| `/probe-nvidia` | Test all NVIDIA models, auto-hide 404s in `~/.pi/free.json` |
-| `/probe-ollama` | Test all Ollama models, auto-hide 403s in `~/.pi/free.json` |
+| Command           | What it does                                                |
+| ----------------- | ----------------------------------------------------------- |
+| `/probe-nvidia`   | Test all NVIDIA models, auto-hide 404s in `~/.pi/free.json` |
+| `/probe-ollama`   | Test all Ollama models, auto-hide 403s in `~/.pi/free.json` |
+| `/probe-routeway` | Test all Routeway models, auto-hide 5xx/404s               |
 
 **How it works:**
 

package/lib/provider-compat.ts

@@ -23,6 +23,10 @@ export function isLikelyReasoningModel(model: ProviderModelIdentity): boolean {
 	const haystack = `${model.id} ${model.name ?? ""}`.toLowerCase();
 	return (
 		isDeepSeekModel(model) ||
+		haystack.includes("minimax") ||
+		haystack.includes("kimi") ||
+		haystack.includes("qwen3.7") ||
+		haystack.includes("qwen3-7") ||
 		haystack.includes("thinking") ||
 		haystack.includes("reasoning") ||
 		haystack.includes("reasoner") ||
@@ -42,5 +46,34 @@ export function getProxyModelCompat(
 		return DEEPSEEK_PROXY_COMPAT;
 	}
 
+	// MiniMax on OpenRouter/Cline uses reasoning_content (DeepSeek format)
+	if (model.id.toLowerCase().includes("minimax")) {
+		return {
+			supportsStore: false,
+			supportsDeveloperRole: false,
+			supportsReasoningEffort: true,
+			requiresReasoningContentOnAssistantMessages: true,
+			thinkingFormat: "deepseek",
+		};
+	}
+
+	// Qwen 3.7+ on OpenRouter/Cline uses reasoning_content (DeepSeek format)
+	if (
+		model.id.toLowerCase().includes("qwen3.7") ||
+		model.id.toLowerCase().includes("qwen3-7")
+	) {
+		return DEEPSEEK_PROXY_COMPAT;
+	}
+
+	// Kimi K2.6 needs reasoning_content on assistant messages (OpenRouter issue #5309)
+	if (model.id.toLowerCase().includes("kimi")) {
+		return {
+			supportsStore: false,
+			supportsDeveloperRole: false,
+			supportsReasoningEffort: true,
+			requiresReasoningContentOnAssistantMessages: true,
+		};
+	}
+
 	return undefined;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "pi-free",
-	"version": "2.0.14",
+	"version": "2.0.15",
 	"type": "module",
 	"description": "AI model providers for Pi with free model filtering and dynamic model fetching",
 	"keywords": [

package/providers/cline/cline-models.ts

@@ -14,6 +14,7 @@ import {
 	PROVIDER_CLINE,
 } from "../../constants.ts";
 import type { ProviderModelConfig } from "../../lib/types.ts";
+import { getProxyModelCompat } from "../../lib/provider-compat.ts";
 import { cleanModelName, fetchWithRetry } from "../../lib/util.ts";
 
 interface ClineRaw {
@@ -164,8 +165,11 @@ function modelFromCatalog(
 		contextWindow:
 			info.context_length ?? info.top_provider?.context_length ?? 128_000,
 		maxTokens: info.top_provider?.max_completion_tokens ?? 8_192,
+		...(getProxyModelCompat({ id: info.id, name: info.name })
+			? { compat: getProxyModelCompat({ id: info.id, name: info.name }) }
+			: {}),
 		_pricingKnown: info.pricing !== null && info.pricing !== undefined,
-	};
+	} as ProviderModelConfig & { _pricingKnown?: boolean; compat?: any };
 }
 
 async function fetchClineRecommendedFreeModels(): Promise<

package/providers/cline/cline.ts

@@ -73,9 +73,9 @@ function toApiKey(credentials: OAuthCredentials): string {
 // =============================================================================
 
 const TASK_PROGRESS_BLOCK = `
-# task_progress List (Optional - Plan Mode)
+# task_progress List (Optional)
 
-While in PLAN MODE, if you've outlined concrete steps or requirements for the user, you may include a preliminary todo list using the task_progress parameter.
+You may include a todo list using the task_progress parameter to track progress on multi-step tasks.
 
 1. To create or update a todo list, include the task_progress parameter in the next tool call
 2. Review each item and update its status:
@@ -100,7 +100,7 @@ function buildEnvironmentDetails(): string {
 0 / 204.8K tokens used (0%)
 
 # Current Mode
-PLAN MODE
+ACT MODE
 </environmentDetails>`;
 }
 

package/providers/nvidia/nvidia.ts

@@ -35,6 +35,10 @@ import {
 	getModelsDueForProbe,
 	recordModelProbeResults,
 } from "../../lib/probe-cache.ts";
+import {
+	getProxyModelCompat,
+	isLikelyReasoningModel,
+} from "../../lib/provider-compat.ts";
 import { registerWithGlobalToggle } from "../../lib/registry.ts";
 import type { ModelsDevModel, ModelsDevProvider } from "../../lib/types.ts";
 import {
@@ -155,7 +159,8 @@ function inferModelFromId(id: string): ModelsDevModel | null {
 		.replaceAll(/\b(\d+(?:\.\d+)?)b\b/gi, "$1B");
 
 	const hasVision = /vision|multimodal|vl/i.test(id);
-	const hasReasoning = /reason|r1|thinking/i.test(id);
+	const hasReasoning =
+		/reason|r1|thinking/i.test(id) || isLikelyReasoningModel({ id, name });
 
 	return {
 		id,
@@ -277,6 +282,7 @@ async function fetchNvidiaModels(
 					},
 					contextWindow: m.limit.context,
 					maxTokens: m.limit.output,
+					compat: getProxyModelCompat({ id: m.id, name: m.name }),
 				}),
 			),
 		PROVIDER_NVIDIA,

package/providers/routeway/routeway.ts

@@ -18,7 +18,12 @@ import type {
 	ExtensionAPI,
 	ProviderModelConfig,
 } from "@earendil-works/pi-coding-agent";
-import { getRoutewayApiKey, getRoutewayShowPaid } from "../../config.ts";
+import {
+	getRoutewayApiKey,
+	getRoutewayShowPaid,
+	loadConfigFile,
+	saveConfig,
+} from "../../config.ts";
 import {
 	BASE_URL_ROUTEWAY,
 	DEFAULT_FETCH_TIMEOUT_MS,
@@ -30,8 +35,13 @@ import {
 	getProxyModelCompat,
 	isLikelyReasoningModel,
 } from "../../lib/provider-compat.ts";
+import {
+	getModelsDueForProbe,
+	recordModelProbeResults,
+} from "../../lib/probe-cache.ts";
 import { isFreeModel, registerWithGlobalToggle } from "../../lib/registry.ts";
 import { cleanModelName, fetchWithRetry } from "../../lib/util.ts";
+import { fetchWithTimeout } from "../../lib/util.ts";
 import { createReRegister, setupProvider } from "../../provider-helper.ts";
 
 const _logger = createLogger("routeway");
@@ -155,6 +165,125 @@ async function fetchRoutewayModels(
 	}
 }
 
+// =============================================================================
+// Probe
+// =============================================================================
+
+async function probeRoutewayModel(
+	apiKey: string,
+	modelId: string,
+): Promise<"ok" | "broken" | "unknown"> {
+	try {
+		const response = await fetchWithTimeout(
+			`${BASE_URL_ROUTEWAY}/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
+		);
+
+		// 5xx = upstream server error (model unavailable)
+		if (response.status >= 500) return "broken";
+		// 404 = model not found / not provisioned
+		if (response.status === 404) return "broken";
+		// 429 = rate limited (model works)
+		if (response.status === 429) return "ok";
+		// 401 = auth issue (model exists, key issue)
+		if (response.status === 401) return "ok";
+		// 400 = bad request (model exists, param issue)
+		if (response.status === 400) return "ok";
+		// 200 = success
+		if (response.ok) return "ok";
+		return "ok";
+	} catch {
+		return "unknown";
+	}
+}
+
+async function runRoutewayProbe(
+	apiKey: string,
+	modelsToTest: ProviderModelConfig[],
+	stored: { free: ProviderModelConfig[]; all: ProviderModelConfig[] },
+	reRegister: (models: ProviderModelConfig[]) => void,
+	options: { useCache?: boolean } = {},
+): Promise<string[]> {
+	const modelIdsToProbe = options.useCache
+		? new Set(
+				getModelsDueForProbe(
+					PROVIDER_ROUTEWAY,
+					modelsToTest.map((m) => m.id),
+				),
+			)
+		: undefined;
+	const probeCandidates = modelIdsToProbe
+		? modelsToTest.filter((m) => modelIdsToProbe.has(m.id))
+		: modelsToTest;
+
+	if (probeCandidates.length === 0) {
+		_logger.info("Auto-probe: Routeway probe cache is fresh");
+		return [];
+	}
+
+	const broken: string[] = [];
+	const cacheableResults: Array<{ modelId: string; status: "ok" | "broken" }> =
+		[];
+	const batchSize = 5;
+
+	for (let i = 0; i < probeCandidates.length; i += batchSize) {
+		const batch = probeCandidates.slice(i, i + batchSize);
+		const results = await Promise.all(
+			batch.map(async (m) => {
+				const status = await probeRoutewayModel(apiKey, m.id);
+				return { id: m.id, status };
+			}),
+		);
+		for (const r of results) {
+			if (r.status === "broken") broken.push(r.id);
+			if (r.status !== "unknown") {
+				cacheableResults.push({ modelId: r.id, status: r.status });
+			}
+		}
+	}
+
+	recordModelProbeResults(PROVIDER_ROUTEWAY, cacheableResults);
+
+	if (broken.length === 0) {
+		_logger.info("Auto-probe: all checked Routeway models are routable");
+		return [];
+	}
+
+	// Auto-hide broken models in config (provider-scoped)
+	const cfg = loadConfigFile();
+	const existingHidden = new Set(cfg.hidden_models ?? []);
+	for (const id of broken) existingHidden.add(`${PROVIDER_ROUTEWAY}/${id}`);
+	saveConfig({ hidden_models: Array.from(existingHidden) });
+
+	// Re-register so hidden models disappear immediately
+	const filtered = await fetchRoutewayModels(apiKey);
+	stored.free = filtered;
+	stored.all = filtered;
+	reRegister(filtered);
+
+	_logger.info(
+		`Auto-probe: found ${broken.length} broken models (auto-hidden)`,
+	);
+	return broken;
+}
+
+// =============================================================================
+// Extension Entry Point
+// =============================================================================
+
 export default async function routewayProvider(pi: ExtensionAPI) {
 	const apiKey = getRoutewayApiKey();
 
@@ -206,6 +335,55 @@ export default async function routewayProvider(pi: ExtensionAPI) {
 		stored,
 	);
 
+	// ── Lazy auto-probe on first session_start ──────────────────────
+	let _autoProbeDone = false;
+	pi.on("session_start", async () => {
+		if (_autoProbeDone || !apiKey) return;
+		_autoProbeDone = true;
+		_logger.info("Starting lazy auto-probe of Routeway models...");
+		runRoutewayProbe(apiKey, allModels, stored, reRegister, {
+			useCache: true,
+		}).catch((err) => {
+			_logger.warn("Auto-probe failed", {
+				error: err instanceof Error ? err.message : String(err),
+			});
+		});
+	});
+
+	// ── Probe command: test all registered models for 5xx ─────────────
+	pi.registerCommand("probe-routeway", {
+		description:
+			"Test all Routeway models for server errors and auto-hide broken ones",
+		handler: async (_args, ctx) => {
+			if (!apiKey) {
+				ctx.ui.notify("ROUTEWAY_API_KEY not set", "error");
+				return;
+			}
+
+			const modelsToTest = allModels;
+			ctx.ui.notify(
+				`Probing ${modelsToTest.length} Routeway models…`,
+				"info",
+			);
+
+			await runRoutewayProbe(apiKey, modelsToTest, stored, reRegister);
+
+			// Check if any were hidden (re-read config)
+			const cfgAfter = loadConfigFile();
+			const newHidden = (cfgAfter.hidden_models ?? []).filter((h) =>
+				h.startsWith(`${PROVIDER_ROUTEWAY}/`),
+			);
+			if (newHidden.length > 0) {
+				ctx.ui.notify(
+					`Found ${newHidden.length} broken models (auto-hidden):\n${newHidden.join("\n")}`,
+					"warning",
+				);
+			} else {
+				ctx.ui.notify("All Routeway models are routable ✅", "info");
+			}
+		},
+	});
+
 	const showPaid = getRoutewayShowPaid();
 	const initialModels =
 		showPaid && stored.all.length > 0 ? stored.all : freeModels;