@oh-my-pi/pi-catalog 15.11.0 → 15.11.2

package/CHANGELOG.md

@@ -2,6 +2,13 @@
 
 ## [Unreleased]
 
+## [15.11.1] - 2026-06-11
+
+### Fixed
+
+- Fixed NVIDIA NIM Qwen turns failing with `400 Validation: Unsupported parameter(s): enable_thinking`. NIM's chat-completions schema is `additionalProperties: false` and exposes thinking via the vLLM convention `chat_template_kwargs.enable_thinking`; `buildOpenAICompat` was sending top-level `enable_thinking` for every `qwen/*` id regardless of host. Registered `nvidia` as a known host (`integrate.api.nvidia.com`) and routed NVIDIA-hosted Qwen models to `thinkingFormat: "qwen-chat-template"` ([#2299](https://github.com/can1357/oh-my-pi/issues/2299)).
+- Fixed Moonshot/Kimi native OpenAI-compatible request metadata so Kimi K2 uses `max_tokens` and omits OpenAI-only `store`, restoring first-turn output with `MOONSHOT_API_KEY` ([#2289](https://github.com/can1357/oh-my-pi/issues/2289)).
+
 ## [15.11.0] - 2026-06-10
 
 ### Fixed

package/dist/types/hosts.d.ts

@@ -98,6 +98,11 @@ export declare const KNOWN_HOSTS: {
         readonly providers: readonly ["qwen-portal"];
         readonly urlMarkers: readonly ["portal.qwen.ai"];
     };
+    /** NVIDIA NIM (`integrate.api.nvidia.com`). Qwen NIM endpoints take `chat_template_kwargs.enable_thinking`, never top-level `enable_thinking`. */
+    readonly nvidia: {
+        readonly providers: readonly ["nvidia"];
+        readonly urlMarkers: readonly ["integrate.api.nvidia.com"];
+    };
     readonly moonshotNative: {
         readonly providers: readonly ["moonshot", "kimi-code"];
         readonly urlMarkers: readonly ["api.moonshot.ai", "api.kimi.com"];

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "15.11.0",
+	"version": "15.11.2",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -34,11 +34,11 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.0",
-		"@oh-my-pi/pi-utils": "15.11.0",
+		"@oh-my-pi/pi-utils": "15.11.2",
 		"zod": "4.4.3"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "15.11.0",
+		"@oh-my-pi/pi-ai": "15.11.2",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/openai.ts

@@ -102,11 +102,13 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 	const isZhipu = modelMatchesHost(hostModel, "zhipu");
 	const isKilo = modelMatchesHost(hostModel, "kilo");
 	const isKimiModel = isKimiModelId(spec.id);
-	const isMoonshotKimi = isKimiModel && modelMatchesHost(hostModel, "moonshotNative");
+	const isMoonshotNative = modelMatchesHost(hostModel, "moonshotNative");
+	const isMoonshotKimi = isKimiModel && isMoonshotNative;
 	const usesMoonshotKimiPreservedThinking = isMoonshotKimi && isKimiK26ModelId(spec.id);
 	const isAnthropicModel =
 		modelMatchesHost(hostModel, "anthropic") || isClaudeModelId(spec.id) || isAnthropicNamespacedModelId(spec.id);
 	const isAlibaba = modelMatchesHost(hostModel, "alibabaDashscope");
+	const isNvidiaNim = modelMatchesHost(hostModel, "nvidia");
 	const isQwen = isQwenModelId(spec.id);
 	// DeepSeek V4 (and other reasoning-capable DeepSeek models) reject follow-up requests in
 	// thinking mode unless prior assistant tool-call turns include `reasoning_content`. The
@@ -145,11 +147,16 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		isKilo ||
 		isQwen ||
 		isXiaomiHost ||
+		isMoonshotNative ||
 		isOpenCodeHost;
 	const isOpenCodeProvider = provider === "opencode-go" || provider === "opencode-zen";
 
 	const useMaxTokens =
-		isMistral || hostMatchesUrl(baseUrl, "chutes") || hostMatchesUrl(baseUrl, "fireworks") || isDirectDeepseekApi;
+		isMistral ||
+		isMoonshotNative ||
+		hostMatchesUrl(baseUrl, "chutes") ||
+		hostMatchesUrl(baseUrl, "fireworks") ||
+		isDirectDeepseekApi;
 
 	// Hosts whose chat-completions endpoints are known to accept multiple
 	// leading `system`/`developer` messages (preferred for KV-cache reuse).
@@ -266,14 +273,20 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		// OpenAI-compatible proxies — Fireworks' Fire Pass router, OpenCode's gateway,
 		// etc. — drives reasoning via OpenAI-style `reasoning_effort`
 		// (low|medium|high|xhigh|max|none), so those stay on the "openai" path.
+		// NVIDIA NIM hosts Qwen with the vLLM convention
+		// (`chat_template_kwargs.enable_thinking`); top-level `enable_thinking`
+		// is rejected by NIM's `additionalProperties: false` request schema
+		// (issue #2299).
 		thinkingFormat:
 			isZai || isZhipu || isMoonshotKimi || isXiaomiMimo
 				? "zai"
 				: isOpenRouter
 					? "openrouter"
-					: isAlibaba || isQwen
-						? "qwen"
-						: "openai",
+					: isQwen && isNvidiaNim
+						? "qwen-chat-template"
+						: isAlibaba || isQwen
+							? "qwen"
+							: "openai",
 		thinkingKeep: usesMoonshotKimiPreservedThinking ? "all" : undefined,
 		reasoningContentField: "reasoning_content",
 		// Backends that 400 follow-up requests when prior assistant tool-call turns lack `reasoning_content`:

package/src/hosts.ts

@@ -54,6 +54,8 @@ export const KNOWN_HOSTS = {
 		urlMarkers: ["api.minimax.io", "api.minimaxi.com"],
 	},
 	qwenPortal: { providers: ["qwen-portal"], urlMarkers: ["portal.qwen.ai"] },
+	/** NVIDIA NIM (`integrate.api.nvidia.com`). Qwen NIM endpoints take `chat_template_kwargs.enable_thinking`, never top-level `enable_thinking`. */
+	nvidia: { providers: ["nvidia"], urlMarkers: ["integrate.api.nvidia.com"] },
 	moonshotNative: { providers: ["moonshot", "kimi-code"], urlMarkers: ["api.moonshot.ai", "api.kimi.com"] },
 	opencode: { providers: ["opencode-go", "opencode-zen"], urlMarkers: ["opencode.ai"] },
 	chutes: { urlMarkers: ["chutes.ai"] },