@oh-my-pi/pi-catalog 16.2.9 → 16.2.11

package/CHANGELOG.md

@@ -2,6 +2,29 @@
 
 ## [Unreleased]
 
+## [16.2.11] - 2026-07-01
+
+### Fixed
+
+- Fixed a potential memory leak caused by dangling timeout timers during model discovery in OpenAI-compatible, vLLM, LiteLLM, and LM Studio catalogs.
+- Widened stream watchdogs for local OpenAI-compatible backends (including llama.cpp, LM Studio, vLLM, and Ollama) to prevent premature timeouts during cold model loads.
+
+## [16.2.10] - 2026-06-30
+
+### Added
+
+- Added Claude Sonnet 3.7, Claude Opus 3, and Claude Sonnet 3 model entries to the Anthropic catalog
+- Added Anthropic Claude Sonnet 5 model entry to the Kilo provider catalog
+- Added first-party catalog discovery support for the Anthropic provider
+- Added Gemini 3.1 Flash Lite Image model entry to the Kilo provider catalog
+- Added Anthropic Claude Sonnet 5 model variants with low, medium, high, xhigh, and max thinking efforts to the Devin provider catalog
+- Added Claude Sonnet 5 model entry to the Anthropic curated catalog.
+
+### Changed
+
+- Updated the base API URL for the Claude Sonnet 5 model in the Anthropic catalog
+- Updated pricing metrics for DeepSeek R1 and DeepSeek V3 model entries to reflect new rates
+
 ## [16.2.9] - 2026-06-30
 
 ### Added

package/dist/types/discovery/openai-compatible.d.ts

@@ -48,8 +48,10 @@ export interface FetchOpenAICompatibleModelsOptions<TApi extends Api> {
     apiKey?: string;
     /** Additional request headers. */
     headers?: Record<string, string>;
-    /** Optional AbortSignal for request cancellation. */
+    /** Optional AbortSignal for request cancellation; caller owns its lifecycle. */
     signal?: AbortSignal;
+    /** Optional cancellable request timeout used when `signal` is omitted. */
+    timeoutMs?: number;
     /** Optional fetch implementation override for testing/custom runtimes. */
     fetch?: FetchImpl;
     /**

package/dist/types/provider-models/descriptors.d.ts

@@ -29,7 +29,11 @@ export declare const CATALOG_PROVIDERS: readonly [{
 }, {
     readonly id: "anthropic";
     readonly defaultModel: "claude-opus-4-8";
+    readonly envVars: readonly ["ANTHROPIC_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("..").ModelManagerOptions<"anthropic-messages", unknown>;
+    readonly catalogDiscovery: {
+        readonly label: "Anthropic";
+    };
 }, {
     readonly id: "azure";
     readonly defaultModel: "gpt-5.5";

package/dist/types/provider-models/openai-compat.d.ts

@@ -29,8 +29,8 @@ export interface ModelsDevModel {
  * first-party `/v1/models` endpoint but that models.dev has not catalogued yet.
  * Seeded into model generation so the bundled catalog is never gated on
  * models.dev's update cadence; deduped behind upstream catalog / models.dev
- * entries once those appear. Token limits and pricing are pinned
- * authoritatively in `applyAnthropicCatalogPolicy`, and `thinking` is re-baked
+ * entries once those appear. Token limits and pricing are pinned either directly or
+ * in `applyAnthropicCatalogPolicy`, and `thinking` is re-baked
  * by the generator's policy pass (scripts/generated-policies.ts).
  */
 export declare const ANTHROPIC_CURATED_FALLBACK_MODELS: readonly ModelSpec<"anthropic-messages">[];
@@ -371,6 +371,7 @@ export interface FetchLiteLLMRichModelsOptions<TApi extends Api> {
     headers?: Record<string, string>;
     fetch?: FetchImpl;
     signal?: AbortSignal;
+    timeoutMs?: number;
     referenceResolver?: (modelId: string) => ModelSpec<TApi> | undefined;
 }
 export declare const OPENAI_COMPAT_DISCOVERY_DEFAULT_CONTEXT_WINDOW = 128000;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "16.2.9",
+	"version": "16.2.11",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -34,12 +34,12 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.0",
-		"@oh-my-pi/pi-utils": "16.2.9",
+		"@oh-my-pi/pi-utils": "16.2.11",
 		"arktype": "^2.2.0",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "16.2.9",
+		"@oh-my-pi/pi-ai": "16.2.11",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/openai.ts

@@ -57,6 +57,8 @@ function matchesKimiK27CodeFamily(spec: ModelSpec<"openai-completions">): boolea
 const XIAOMI_MIMO_STREAM_IDLE_TIMEOUT_MS = 300_000;
 /** Alibaba Coding Plan (coding-intl.dashscope) qwen models idle before the first event (issue #1770). */
 const ALIBABA_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS = 600_000;
+/** Local OpenAI-compatible backends can spend minutes cold-loading a model before the first SSE event. */
+const LOCAL_OPENAI_COMPAT_STREAM_IDLE_TIMEOUT_MS = 300_000;
 const MINIMAX_PROVIDER_OR_ID_PATTERN = /minimax/i;
 const DSML_HEALING_PROVIDERS = new Set([
 	"ollama",
@@ -294,6 +296,9 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		isMoonshotNative ||
 		isOpenCodeHost;
 	const isOpenCodeProvider = provider === "opencode-go" || provider === "opencode-zen";
+	const isLocalOpenAICompatBackend =
+		!PROXY_OPENAI_COMPAT_PROVIDERS.has(provider) &&
+		(LOCAL_OPENAI_COMPAT_PROVIDERS.has(provider) || hasLocalLoopbackBaseUrl(baseUrl));
 
 	const useMaxTokens =
 		isMistral ||
@@ -348,9 +353,10 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 			isCopilotHost ||
 			isZenmuxHost);
 
-	// Stream-watchdog floor: GLM coding-plan SKUs, Kimi K2.6, and direct
-	// DeepSeek reasoning models can idle for minutes while reasoning; widen the
-	// idle timeout so warm-ups stop aborting and retrying.
+	// Stream-watchdog floor: GLM coding-plan SKUs, Kimi K2.6, direct
+	// DeepSeek reasoning models, and local OpenAI-compatible backends can idle
+	// for minutes while reasoning or cold-loading weights; widen the idle
+	// timeout so warm-ups stop aborting and retrying.
 	const streamIdleTimeoutMs =
 		GLM_CODING_PLAN_MODEL_PATTERN.test(spec.id) && (isZai || isZhipu)
 			? GLM_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS
@@ -362,7 +368,9 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 						? KIMI_K26_REASONING_STREAM_IDLE_TIMEOUT_MS
 						: spec.reasoning && isDirectDeepseekApi
 							? DEEPSEEK_REASONING_STREAM_IDLE_TIMEOUT_MS
-							: undefined;
+							: isLocalOpenAICompatBackend
+								? LOCAL_OPENAI_COMPAT_STREAM_IDLE_TIMEOUT_MS
+								: undefined;
 
 	// Fireworks "Fast" variants (`<id>-fast`) are served from the router
 	// namespace (`accounts/fireworks/routers/<id>-fast`), like Fire Pass, rather
@@ -482,9 +490,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		// when a thinking block actually exists on the turn
 		// (`nonEmptyThinkingBlocks.length > 0`), so the flag is a no-op on
 		// pure-text histories.
-		replayReasoningContent:
-			!PROXY_OPENAI_COMPAT_PROVIDERS.has(provider) &&
-			(LOCAL_OPENAI_COMPAT_PROVIDERS.has(provider) || hasLocalLoopbackBaseUrl(baseUrl)),
+		replayReasoningContent: isLocalOpenAICompatBackend,
 		// `preserve_thinking: true` makes the Qwen3.6+ chat template render
 		// `<think>...</think>` for older assistant turns too, instead of
 		// stripping it the moment a new user message moves them past
@@ -496,9 +502,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		// with `replayReasoningContent` above). Non-Qwen templates ignore the
 		// parameter, so the flag stays a no-op outside the Qwen path.
 		qwenPreserveThinking:
-			(thinkingFormat === "qwen" || thinkingFormat === "qwen-chat-template") &&
-			!PROXY_OPENAI_COMPAT_PROVIDERS.has(provider) &&
-			(LOCAL_OPENAI_COMPAT_PROVIDERS.has(provider) || hasLocalLoopbackBaseUrl(baseUrl)),
+			(thinkingFormat === "qwen" || thinkingFormat === "qwen-chat-template") && isLocalOpenAICompatBackend,
 		requiresAssistantContentForToolCalls: isKimiModel || isDirectDeepseekReasoning,
 		cacheControlFormat: isOpenRouter && spec.id.startsWith("anthropic/") ? "anthropic" : undefined,
 		openRouterRouting: undefined,
@@ -584,6 +588,9 @@ export function buildOpenAIResponsesCompat(spec: OpenAIResponsesSpecLike): Resol
 	const isAnthropicModel = id ? isClaudeModelId(id) || isAnthropicNamespacedModelId(id) : false;
 	const isDeepseekFamily = id ? isDeepseekModelIdOrName(id) || isDeepseekModelIdOrName(spec.name) : false;
 	const reasoningCapable = Boolean(spec.reasoning);
+	const isLocalOpenAICompatBackend =
+		!PROXY_OPENAI_COMPAT_PROVIDERS.has(spec.provider) &&
+		(LOCAL_OPENAI_COMPAT_PROVIDERS.has(spec.provider) || hasLocalLoopbackBaseUrl(baseUrl));
 
 	const compat: ResolvedOpenAIResponsesCompat = {
 		supportsDeveloperRole: isAzure || isOpenAIUrl || hostMatchesUrl(baseUrl, "githubCopilot"),
@@ -645,7 +652,9 @@ export function buildOpenAIResponsesCompat(spec: OpenAIResponsesSpecLike): Resol
 		emptyLengthFinishIsContextError: spec.provider === "ollama",
 		usesOpenAIToolCallIdLimit: spec.provider === "openai",
 		promptCacheSessionHeader: spec.provider === "xai-oauth" ? "x-grok-conv-id" : undefined,
-		streamIdleTimeoutMs: spec.compat?.streamIdleTimeoutMs,
+		streamIdleTimeoutMs: isLocalOpenAICompatBackend
+			? LOCAL_OPENAI_COMPAT_STREAM_IDLE_TIMEOUT_MS
+			: spec.compat?.streamIdleTimeoutMs,
 	};
 	applyCompatOverrides(compat, spec.compat);
 	if (spec.compat?.reasoningDisableMode === undefined) {

package/src/discovery/openai-compatible.ts

@@ -3,6 +3,27 @@ import type { Api, FetchImpl, ModelSpec, Provider } from "../types";
 
 const MODELS_PATH = "/models";
 
+/**
+ * Uses a cancellable timer rather than the native abort-timeout helper so
+ * successful fast discovery requests do not leave armed timeout signals for
+ * concurrent GC to trip over later.
+ */
+async function withOpenAICompatibleDiscoveryTimeout<T>(
+	timeoutMs: number,
+	run: (signal: AbortSignal) => Promise<T>,
+): Promise<T> {
+	const controller = new AbortController();
+	const timer = setTimeout(
+		() => controller.abort(new DOMException("The operation timed out.", "TimeoutError")),
+		timeoutMs,
+	);
+	try {
+		return await run(controller.signal);
+	} finally {
+		clearTimeout(timer);
+	}
+}
+
 /**
  * Minimal OpenAI-style model entry shape consumed by discovery.
  *
@@ -72,8 +93,10 @@ export interface FetchOpenAICompatibleModelsOptions<TApi extends Api> {
 	apiKey?: string;
 	/** Additional request headers. */
 	headers?: Record<string, string>;
-	/** Optional AbortSignal for request cancellation. */
+	/** Optional AbortSignal for request cancellation; caller owns its lifecycle. */
 	signal?: AbortSignal;
+	/** Optional cancellable request timeout used when `signal` is omitted. */
+	timeoutMs?: number;
 	/** Optional fetch implementation override for testing/custom runtimes. */
 	fetch?: FetchImpl;
 	/**
@@ -115,25 +138,35 @@ export async function fetchOpenAICompatibleModels<TApi extends Api>(
 	}
 
 	const fetchImpl = options.fetch ?? globalThis.fetch;
-	let response: Response;
-	try {
-		response = await fetchImpl(`${baseUrl}${MODELS_PATH}`, {
-			method: "GET",
-			headers: requestHeaders,
-			signal: options.signal,
-		});
-	} catch {
-		return null;
-	}
+	const fetchPayload = async (signal?: AbortSignal): Promise<unknown | null> => {
+		let response: Response;
+		try {
+			response = await fetchImpl(`${baseUrl}${MODELS_PATH}`, {
+				method: "GET",
+				headers: requestHeaders,
+				signal,
+			});
+		} catch {
+			return null;
+		}
 
-	if (!response.ok) {
-		return null;
-	}
+		if (!response.ok) {
+			return null;
+		}
 
-	let payload: unknown;
-	try {
-		payload = await response.json();
-	} catch {
+		try {
+			return await response.json();
+		} catch {
+			return null;
+		}
+	};
+	const payload =
+		options.signal !== undefined
+			? await fetchPayload(options.signal)
+			: options.timeoutMs !== undefined
+				? await withOpenAICompatibleDiscoveryTimeout(options.timeoutMs, fetchPayload)
+				: await fetchPayload();
+	if (payload === null) {
 		return null;
 	}