@oh-my-pi/pi-ai 14.7.8 → 14.8.1

package/CHANGELOG.md

@@ -2,6 +2,13 @@
 
 ## [Unreleased]
 
+## [14.8.0] - 2026-05-09
+
+### Fixed
+- Fixed Gemini 3 Pro thinking metadata so `medium` effort is rejected with the expected error instead of being silently accepted: `ThinkingConfig` now carries an optional explicit `levels` list that survives `expandEffortRange`, letting non-contiguous supported sets (e.g. `[low, high]`) round-trip through enrichment.
+- Fixed Kimi Code OAuth expiry handling to refresh access tokens 5 minutes before server expiry, avoiding daily 401s from using tokens right up to the cutoff.
+- Fixed OpenAI Responses custom tool replay to preserve custom tool call item IDs with the `ctc_` prefix instead of rewriting them as `fc_` function-call IDs ([#977](https://github.com/can1357/oh-my-pi/issues/977)).
+
 ## [14.7.6] - 2026-05-07
 
 ### Added

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "14.7.8",
+	"version": "14.8.1",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://github.com/can1357/oh-my-pi",
 	"author": "Can Boluk",
@@ -46,8 +46,8 @@
 		"@aws-sdk/credential-provider-node": "^3.972.39",
 		"@bufbuild/protobuf": "^2.12.0",
 		"@google/genai": "^1.52.0",
-		"@oh-my-pi/pi-natives": "14.7.8",
-		"@oh-my-pi/pi-utils": "14.7.8",
+		"@oh-my-pi/pi-natives": "14.8.1",
+		"@oh-my-pi/pi-utils": "14.8.1",
 		"@sinclair/typebox": "^0.34.49",
 		"@smithy/node-http-handler": "^4.6.1",
 		"ajv": "^8.20.0",

package/src/model-thinking.ts

@@ -182,8 +182,11 @@ export function linkOpenAIPromotionTargets(models: ApiModel<Api>[]): void {
 }
 
 /**
- * Returns supported thinking efforts from canonical model rules constrained by
- * explicit model metadata.
+ * Returns the supported thinking efforts declared on the model metadata.
+ *
+ * Catalog enrichment is responsible for normalizing bundled model metadata up front.
+ * Runtime callers must treat explicit `model.thinking` on custom models as authoritative
+ * so proxy-specific overrides from `models.yml` survive request construction.
  *
  * @throws Error when a reasoning-capable model is missing thinking metadata
  */
@@ -194,12 +197,7 @@ export function getSupportedEfforts<TApi extends Api>(model: ApiModel<TApi>): re
 	if (!model.thinking) {
 		throw new Error(`Model ${model.provider}/${model.id} is missing thinking metadata`);
 	}
-	const configuredEfforts = expandEffortRange(model.thinking);
-	const parsedModel = parseKnownModel(model.id);
-	if (parsedModel.family === "unknown") {
-		return configuredEfforts;
-	}
-	return intersectEfforts(configuredEfforts, inferSupportedEfforts(parsedModel, model));
+	return expandEffortRange(model.thinking);
 }
 
 /**
@@ -316,6 +314,19 @@ function applyGeneratedModelPolicy(model: ApiModel<Api>): void {
 		model.maxTokens = copilotLimits.maxTokens;
 	}
 
+	if (
+		model.api === "openai-completions" &&
+		(model.provider === "minimax-code" || model.provider === "minimax-code-cn")
+	) {
+		model.compat = {
+			...model.compat,
+			supportsStore: false,
+			supportsDeveloperRole: false,
+			supportsReasoningEffort: false,
+			reasoningContentField: "reasoning_content",
+		};
+		delete model.compat.thinkingFormat;
+	}
 	const parsedModel = parseKnownModel(model.id);
 	const applyPatchToolType = inferGeneratedApplyPatchToolType(model, parsedModel);
 	if (applyPatchToolType) {
@@ -392,11 +403,19 @@ function inferModelThinking<TApi extends Api>(model: ApiModel<TApi>): ThinkingCo
 	if (!minLevel || !maxLevel) {
 		throw new Error(`Model ${model.provider}/${model.id} resolved to an empty thinking range`);
 	}
-	return {
+	const config: ThinkingConfig = {
 		mode: inferThinkingControlMode(model, parsedModel),
 		minLevel,
 		maxLevel,
 	};
+	// Encode explicit levels only when the inferred set has gaps the min..max range cannot represent.
+	const minIndex = THINKING_EFFORTS.indexOf(minLevel);
+	const maxIndex = THINKING_EFFORTS.indexOf(maxLevel);
+	const expandedRange = THINKING_EFFORTS.slice(minIndex, maxIndex + 1);
+	if (expandedRange.length !== efforts.length) {
+		config.levels = efforts;
+	}
+	return config;
 }
 
 function normalizeThinkingConfig(thinking: ThinkingConfig | undefined): ThinkingConfig | undefined {
@@ -409,10 +428,19 @@ function normalizeThinkingConfig(thinking: ThinkingConfig | undefined): Thinking
 function thinkingsEqual(left: ThinkingConfig | undefined, right: ThinkingConfig | undefined): boolean {
 	if (left === right) return true;
 	if (!left || !right) return false;
-	return left.mode === right.mode && left.minLevel === right.minLevel && left.maxLevel === right.maxLevel;
+	if (left.mode !== right.mode || left.minLevel !== right.minLevel || left.maxLevel !== right.maxLevel) return false;
+	const leftLevels = left.levels;
+	const rightLevels = right.levels;
+	if (leftLevels === rightLevels) return true;
+	if (!leftLevels || !rightLevels) return false;
+	if (leftLevels.length !== rightLevels.length) return false;
+	return leftLevels.every((level, index) => level === rightLevels[index]);
 }
 
 function expandEffortRange(thinking: ThinkingConfig): readonly Effort[] {
+	if (thinking.levels && thinking.levels.length > 0) {
+		return thinking.levels;
+	}
 	const minIndex = THINKING_EFFORTS.indexOf(thinking.minLevel);
 	const maxIndex = THINKING_EFFORTS.indexOf(thinking.maxLevel);
 	if (minIndex === -1 || maxIndex === -1 || minIndex > maxIndex) {
@@ -421,10 +449,6 @@ function expandEffortRange(thinking: ThinkingConfig): readonly Effort[] {
 	return THINKING_EFFORTS.slice(minIndex, maxIndex + 1);
 }
 
-function intersectEfforts(left: readonly Effort[], right: readonly Effort[]): readonly Effort[] {
-	return left.filter(effort => right.includes(effort));
-}
-
 function inferSupportedEfforts<TApi extends Api>(parsedModel: ParsedModel, model: ApiModel<TApi>): readonly Effort[] {
 	switch (parsedModel.family) {
 		case "openai":

package/src/models.json

@@ -18451,7 +18451,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {
@@ -18481,7 +18481,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {
@@ -18508,7 +18508,7 @@
 			},
 			"compat": {
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"contextWindow": 1000000,
@@ -18540,7 +18540,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {
@@ -18570,7 +18570,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {
@@ -18597,7 +18597,7 @@
 			},
 			"compat": {
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"contextWindow": 204800,
@@ -18629,7 +18629,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {
@@ -18659,7 +18659,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {
@@ -18691,7 +18691,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {
@@ -18721,7 +18721,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {
@@ -18748,7 +18748,7 @@
 			},
 			"compat": {
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"contextWindow": 1000000,
@@ -18780,7 +18780,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {
@@ -18810,7 +18810,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {
@@ -18837,7 +18837,7 @@
 			},
 			"compat": {
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"contextWindow": 204800,
@@ -18869,7 +18869,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {
@@ -18899,7 +18899,7 @@
 			"compat": {
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
-				"thinkingFormat": "zai",
+				"supportsReasoningEffort": false,
 				"reasoningContentField": "reasoning_content"
 			},
 			"thinking": {

package/src/provider-models/openai-compat.ts

@@ -2112,7 +2112,7 @@ const MODELS_DEV_PROVIDER_DESCRIPTORS_CODING_PLANS: readonly ModelsDevProviderDe
 		compat: {
 			supportsStore: false,
 			supportsDeveloperRole: false,
-			thinkingFormat: "zai",
+			supportsReasoningEffort: false,
 			reasoningContentField: "reasoning_content",
 		},
 	}),
@@ -2120,7 +2120,7 @@ const MODELS_DEV_PROVIDER_DESCRIPTORS_CODING_PLANS: readonly ModelsDevProviderDe
 		compat: {
 			supportsStore: false,
 			supportsDeveloperRole: false,
-			thinkingFormat: "zai",
+			supportsReasoningEffort: false,
 			reasoningContentField: "reasoning_content",
 		},
 	}),

package/src/providers/openai-codex-responses.ts

@@ -2438,7 +2438,7 @@ function convertMessages(model: Model<"openai-codex-responses">, context: Contex
 				}
 				if (block.type === "toolCall") {
 					const toolCall = block as ToolCall;
-					const normalized = normalizeResponsesToolCallId(toolCall.id);
+					const normalized = normalizeResponsesToolCallId(toolCall.id, toolCall.customWireName ? "ctc" : "fc");
 					if (toolCall.customWireName) {
 						const rawInput = typeof toolCall.arguments?.input === "string" ? toolCall.arguments.input : "";
 						customCallIds.add(normalized.callId);

package/src/providers/openai-completions.ts

@@ -287,10 +287,18 @@ function getTrailingPartialTag(text: string, tags: readonly string[]): string {
 // Body is restricted to identifier-like chars (with the DeepSeek tokenizer's `▁`),
 // capped at a sane length to avoid swallowing legitimate angle-bracket text.
 const DEEPSEEK_SPECIAL_TOKEN_REGEX = /<(?:｜|\|)[A-Za-z0-9_.｜|▁]{1,64}(?:｜|\|)>/g;
+const DEEPSEEK_SPECIAL_TOKEN_AT_START_REGEX = /^\s*<(?:｜|\|)[A-Za-z0-9_.｜|▁]{1,64}(?:｜|\|)>/;
+const DEEPSEEK_SPECIAL_TOKEN_AT_END_REGEX = /<(?:｜|\|)[A-Za-z0-9_.｜|▁]{1,64}(?:｜|\|)>\s*$/;
 const DEEPSEEK_OPEN_DELIMS = ["<｜", "<|"] as const;
 
 function stripDeepseekSpecialTokens(text: string): string {
-	return text.replace(DEEPSEEK_SPECIAL_TOKEN_REGEX, "");
+	const stripped = text.replace(DEEPSEEK_SPECIAL_TOKEN_REGEX, "");
+	if (stripped === text) return text;
+
+	let normalized = stripped;
+	if (DEEPSEEK_SPECIAL_TOKEN_AT_START_REGEX.test(text)) normalized = normalized.replace(/^\s+/u, "");
+	if (DEEPSEEK_SPECIAL_TOKEN_AT_END_REGEX.test(text)) normalized = normalized.replace(/\s+$/u, "");
+	return normalized;
 }
 
 // Find any trailing partial `<｜...` (or `<|...`) that has not yet been closed by a
@@ -431,10 +439,12 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions"> = (
 			stream.push({ type: "start", partial: output });
 
 			const parseMiniMaxThinkTags = model.provider === "minimax-code";
-			// NVIDIA NIM and similar OpenAI-compatible hosts return DeepSeek's chat-template
-			// tool-call markers in `delta.content` even though tool calls are also surfaced
-			// structurally. Strip the leaked markers so users don't see raw `<｜...｜>` tokens.
-			const stripDeepseekChatTemplateTokens = model.provider === "nvidia" && /deepseek/i.test(model.id);
+			// Some OpenAI-compatible DeepSeek hosts (including NVIDIA NIM and DeepSeek's
+			// native API) leak chat-template tool-call markers in `delta.content` even
+			// though tool calls are also surfaced structurally. Strip the leaked markers
+			// so users don't see raw `<｜...｜>` tokens.
+			const stripDeepseekChatTemplateTokens =
+				/deepseek/i.test(model.id) && (model.provider === "nvidia" || model.provider === "deepseek");
 			type OpenAIStreamBlock = TextContent | ThinkingContent | (ToolCall & { partialArgs: string });
 			let currentBlock: OpenAIStreamBlock | undefined;
 			const blockIndex = (block: OpenAIStreamBlock | undefined): number => {
@@ -568,7 +578,7 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions"> = (
 					deepseekStripBuffer = trailing;
 				}
 				const stripped = stripDeepseekSpecialTokens(flushable);
-				if (stripped) appendTextDelta(stripped);
+				if (stripped && (stripped === flushable || stripped.trim().length > 0)) appendTextDelta(stripped);
 			};
 
 			for await (const chunk of iterateWithIdleTimeout(openaiStream, {

package/src/providers/openai-responses-shared.ts

@@ -187,9 +187,9 @@ export function convertResponsesAssistantMessage<TApi extends Api>(
 			continue;
 		}
 
-		const normalized = normalizeResponsesToolCallId(block.id);
+		const normalized = normalizeResponsesToolCallId(block.id, block.customWireName ? "ctc" : "fc");
 		let itemId: string | undefined = normalized.itemId;
-		if (isDifferentModel && (itemId?.startsWith("fc_") || itemId?.startsWith("fcr_"))) {
+		if (isDifferentModel && (itemId?.startsWith("fc_") || itemId?.startsWith("fcr_") || itemId?.startsWith("ctc_"))) {
 			itemId = undefined;
 		}
 		knownCallIds.add(normalized.callId);

package/src/types.ts

@@ -83,6 +83,12 @@ export interface ThinkingConfig {
 	minLevel: Effort;
 	/** Most intensive supported user-facing effort level. */
 	maxLevel: Effort;
+	/**
+	 * Optional explicit list of supported levels. When present, takes precedence over
+	 * the `minLevel`..`maxLevel` range — used to encode discrete sets with gaps
+	 * (e.g. Gemini 3 Pro supports `low` and `high` but not `medium`).
+	 */
+	levels?: readonly Effort[];
 	/** Optional default effort applied when this model is selected. Falls back to global default if absent. */
 	defaultLevel?: Effort;
 	/** Provider-specific transport used to encode the selected effort. */

package/src/utils/oauth/kimi.ts

@@ -15,6 +15,7 @@ const DEFAULT_OAUTH_HOST = "https://auth.kimi.com";
 const DEVICE_ID_FILENAME = "kimi-device-id";
 const DEFAULT_POLL_INTERVAL_MS = 5000;
 const DEFAULT_DEVICE_FLOW_TTL_MS = 15 * 60 * 1000;
+const OAUTH_EXPIRY_SKEW_MS = 5 * 60 * 1000;
 
 interface DeviceAuthorizationResponse {
 	user_code?: string;
@@ -146,7 +147,7 @@ function parseTokenPayload(payload: TokenResponse, refreshTokenFallback?: string
 	return {
 		access: payload.access_token,
 		refresh,
-		expires: Date.now() + payload.expires_in * 1000,
+		expires: Date.now() + payload.expires_in * 1000 - OAUTH_EXPIRY_SKEW_MS,
 	};
 }
 

package/src/utils.ts

@@ -5,8 +5,10 @@ import type { CacheRetention, OpenAIResponsesHistoryPayload, ProviderPayload } f
 type OpenAIResponsesReplayItem = ResponseInput[number];
 
 export { isRecord } from "@oh-my-pi/pi-utils";
-export function normalizeSystemPrompts(systemPrompt: readonly string[] | undefined): string[] {
-	return systemPrompt?.map(prompt => prompt.toWellFormed()).filter(prompt => prompt.length > 0) ?? [];
+export function normalizeSystemPrompts(systemPrompt: readonly string[] | string | undefined | null): string[] {
+	if (systemPrompt === undefined || systemPrompt === null) return [];
+	const prompts = Array.isArray(systemPrompt) ? systemPrompt : typeof systemPrompt === "string" ? [systemPrompt] : [];
+	return prompts.map(prompt => prompt.toWellFormed()).filter(prompt => prompt.length > 0);
 }
 
 export function toNumber(value: unknown): number | undefined {
@@ -34,16 +36,21 @@ export function normalizeToolCallId(id: string): string {
 	return sanitized.length > 64 ? sanitized.slice(0, 64) : sanitized;
 }
 
-export function normalizeResponsesToolCallId(id: string): { callId: string; itemId: string } {
+type ResponsesToolItemIdPrefix = "fc" | "ctc";
+
+export function normalizeResponsesToolCallId(
+	id: string,
+	itemPrefix: ResponsesToolItemIdPrefix = "fc",
+): { callId: string; itemId: string } {
 	const [callId, itemId] = id.split("|");
 	if (callId && itemId) {
 		const normalizedCallId = truncateResponseItemId(callId, getIdPrefix(callId, "call"));
-		const normalizedItemId = normalizeResponsesItemId(itemId);
+		const normalizedItemId = normalizeResponsesItemId(itemId, itemPrefix);
 		return { callId: normalizedCallId, itemId: normalizedItemId };
 	}
 	const hash = Bun.hash(id).toString(36);
 	const normalizedCallId = id.startsWith("call_") ? truncateResponseItemId(id, "call") : `call_${hash}`;
-	return { callId: normalizedCallId, itemId: `fc_${hash}` };
+	return { callId: normalizedCallId, itemId: `${itemPrefix}_${hash}` };
 }
 
 function getIdPrefix(id: string, fallback: string): string {
@@ -51,10 +58,19 @@ function getIdPrefix(id: string, fallback: string): string {
 	return prefix || fallback;
 }
 
-function normalizeResponsesItemId(itemId: string): string {
-	const prefix = getIdPrefix(itemId, "fc");
-	if (prefix !== "fc" && prefix !== "fcr") {
-		return `fc_${Bun.hash(itemId).toString(36)}`;
+function getExplicitIdPrefix(id: string): string | undefined {
+	return id.match(/^([a-zA-Z][a-zA-Z0-9]*)_/)?.[1];
+}
+
+function normalizeResponsesItemId(itemId: string, fallbackPrefix: ResponsesToolItemIdPrefix): string {
+	const prefix = getExplicitIdPrefix(itemId);
+	const isAllowedPrefix = prefix
+		? fallbackPrefix === "ctc"
+			? prefix === "ctc"
+			: prefix === "fc" || prefix === "fcr"
+		: false;
+	if (!prefix || !isAllowedPrefix) {
+		return `${fallbackPrefix}_${Bun.hash(itemId).toString(36)}`;
 	}
 	return truncateResponseItemId(itemId, prefix);
 }