@oh-my-pi/pi-catalog 17.1.2 → 17.1.3

package/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 ## [Unreleased]
 
+## [17.1.3] - 2026-07-24
+
+### Fixed
+
+- Disabled the first-event watchdog for local OpenAI-compatible backends while retaining the 300-second inter-event watchdog, so long llama.cpp prompt prefill is not canceled and retried ([#6524](https://github.com/can1357/oh-my-pi/issues/6524)).
+
 ## [17.1.1] - 2026-07-24
 
 ### Added

package/dist/types/types.d.ts

@@ -303,6 +303,12 @@ export interface OpenAICompat {
      * to opt a host out.
      */
     toolSchemaFlavor?: "moonshot-mfjs" | "grammar" | "none";
+    /**
+     * Stream-watchdog first-event timeout in ms.
+     * Set to `0` to allow unbounded prompt processing. Default: auto-detected
+     * (disabled for local OpenAI-compatible backends).
+     */
+    streamFirstEventTimeoutMs?: number;
     /**
      * Stream-watchdog idle-timeout floor in ms for slow reasoning hosts.
      * Default: auto-detected (GLM coding-plan hosts, direct DeepSeek reasoning).
@@ -520,6 +526,8 @@ export interface ResolvedOpenAISharedCompat {
     requiresAssistantContentForToolCalls: boolean;
     stripDeepseekSpecialTokens: boolean;
     streamMarkupHealingPattern?: OpenAIStreamMarkupHealingPattern;
+    /** See {@link OpenAICompat.streamFirstEventTimeoutMs}. */
+    streamFirstEventTimeoutMs?: number;
     reasoningDeltasMayBeCumulative: boolean;
     emptyLengthFinishIsContextError: boolean;
     usesOpenAIToolCallIdLimit: boolean;
@@ -550,7 +558,7 @@ export interface ResolvedOpenAISharedCompat {
  * `buildModel`; request handlers read fields and never detect, resolve, or
  * allocate.
  */
-export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsSamplingParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "replayReasoningContent" | "qwenPreserveThinking" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamIdleTimeoutMs" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "enableGeminiThinkingLoopGuard" | "whenThinking">> & {
+export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsSamplingParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "replayReasoningContent" | "qwenPreserveThinking" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "enableGeminiThinkingLoopGuard" | "whenThinking">> & {
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     extraBody?: OpenAICompat["extraBody"];
     cacheControlFormat?: OpenAICompat["cacheControlFormat"];

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "17.1.2",
+	"version": "17.1.3",
 	"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": "^2.12.1",
-		"@oh-my-pi/pi-utils": "17.1.2",
+		"@oh-my-pi/pi-utils": "17.1.3",
 		"arktype": "2.2.3",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "17.1.2",
+		"@oh-my-pi/pi-ai": "17.1.3",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/openai.ts

@@ -574,6 +574,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		// to Moonshot verbatim, which 400s on non-MFJS constructs.
 		toolSchemaFlavor:
 			isMoonshotNative || isKimiModel ? "moonshot-mfjs" : isLocalOpenAICompatBackend ? "grammar" : undefined,
+		streamFirstEventTimeoutMs: isLocalServingBackend ? 0 : undefined,
 		streamIdleTimeoutMs,
 		stripDeepseekSpecialTokens:
 			isDeepseekModelIdOrName(spec.id) && (provider === "nvidia" || provider === "deepseek"),
@@ -723,6 +724,7 @@ export function buildOpenAIResponsesCompat(spec: OpenAIResponsesSpecLike): Resol
 		emptyLengthFinishIsContextError: spec.provider === "ollama",
 		usesOpenAIToolCallIdLimit: spec.provider === "openai",
 		promptCacheSessionHeader: spec.provider === "xai-oauth" ? "x-grok-conv-id" : undefined,
+		streamFirstEventTimeoutMs: isLocalServingBackend ? 0 : spec.compat?.streamFirstEventTimeoutMs,
 		streamIdleTimeoutMs: isLocalServingBackend
 			? LOCAL_OPENAI_COMPAT_STREAM_IDLE_TIMEOUT_MS
 			: spec.compat?.streamIdleTimeoutMs,

package/src/types.ts

@@ -341,6 +341,12 @@ export interface OpenAICompat {
 	 * to opt a host out.
 	 */
 	toolSchemaFlavor?: "moonshot-mfjs" | "grammar" | "none";
+	/**
+	 * Stream-watchdog first-event timeout in ms.
+	 * Set to `0` to allow unbounded prompt processing. Default: auto-detected
+	 * (disabled for local OpenAI-compatible backends).
+	 */
+	streamFirstEventTimeoutMs?: number;
 	/**
 	 * Stream-watchdog idle-timeout floor in ms for slow reasoning hosts.
 	 * Default: auto-detected (GLM coding-plan hosts, direct DeepSeek reasoning).
@@ -565,6 +571,8 @@ export interface ResolvedOpenAISharedCompat {
 	requiresAssistantContentForToolCalls: boolean;
 	stripDeepseekSpecialTokens: boolean;
 	streamMarkupHealingPattern?: OpenAIStreamMarkupHealingPattern;
+	/** See {@link OpenAICompat.streamFirstEventTimeoutMs}. */
+	streamFirstEventTimeoutMs?: number;
 	reasoningDeltasMayBeCumulative: boolean;
 	emptyLengthFinishIsContextError: boolean;
 	usesOpenAIToolCallIdLimit: boolean;
@@ -645,6 +653,7 @@ export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat &
 			| "extraBody"
 			| "toolStrictMode"
 			| "toolSchemaFlavor"
+			| "streamFirstEventTimeoutMs"
 			| "streamIdleTimeoutMs"
 			| "cacheControlFormat"
 			| "thinkingKeep"