librechat-data-provider 0.8.505 → 0.8.506

package/dist/index.js

@@ -1,63 +1,18 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
-const require_data_service = require("./data-service-CG5e1FOi.js");
+const require_data_service = require("./data-service-CCdWvcRd.js");
 let zod = require("zod");
 let dayjs = require("dayjs");
 dayjs = require_data_service.__toESM(dayjs);
+let dayjs_plugin_utc = require("dayjs/plugin/utc");
+dayjs_plugin_utc = require_data_service.__toESM(dayjs_plugin_utc);
+let dayjs_plugin_timezone = require("dayjs/plugin/timezone");
+dayjs_plugin_timezone = require_data_service.__toESM(dayjs_plugin_timezone);
 let axios = require("axios");
 axios = require_data_service.__toESM(axios);
 let url = require("url");
 let crypto = require("crypto");
 crypto = require_data_service.__toESM(crypto);
 let js_yaml = require("js-yaml");
-//#region src/types/runs.ts
-let ContentTypes = /* @__PURE__ */ function(ContentTypes) {
-	ContentTypes["TEXT"] = "text";
-	ContentTypes["THINK"] = "think";
-	ContentTypes["TEXT_DELTA"] = "text_delta";
-	ContentTypes["TOOL_CALL"] = "tool_call";
-	ContentTypes["IMAGE_FILE"] = "image_file";
-	ContentTypes["IMAGE_URL"] = "image_url";
-	ContentTypes["VIDEO_URL"] = "video_url";
-	ContentTypes["INPUT_AUDIO"] = "input_audio";
-	ContentTypes["AGENT_UPDATE"] = "agent_update";
-	ContentTypes["SUMMARY"] = "summary";
-	ContentTypes["ERROR"] = "error";
-	return ContentTypes;
-}({});
-let StepTypes = /* @__PURE__ */ function(StepTypes) {
-	StepTypes["TOOL_CALLS"] = "tool_calls";
-	StepTypes["MESSAGE_CREATION"] = "message_creation";
-	return StepTypes;
-}({});
-let ToolCallTypes = /* @__PURE__ */ function(ToolCallTypes) {
-	ToolCallTypes["FUNCTION"] = "function";
-	ToolCallTypes["RETRIEVAL"] = "retrieval";
-	ToolCallTypes["FILE_SEARCH"] = "file_search";
-	ToolCallTypes["CODE_INTERPRETER"] = "code_interpreter";
-	ToolCallTypes["TOOL_CALL"] = "tool_call";
-	return ToolCallTypes;
-}({});
-/** Event names dispatched by the agent graph and consumed by step handlers. */
-let StepEvents = /* @__PURE__ */ function(StepEvents) {
-	StepEvents["ON_RUN_STEP"] = "on_run_step";
-	StepEvents["ON_AGENT_UPDATE"] = "on_agent_update";
-	StepEvents["ON_MESSAGE_DELTA"] = "on_message_delta";
-	StepEvents["ON_REASONING_DELTA"] = "on_reasoning_delta";
-	StepEvents["ON_RUN_STEP_DELTA"] = "on_run_step_delta";
-	StepEvents["ON_RUN_STEP_COMPLETED"] = "on_run_step_completed";
-	StepEvents["ON_SUMMARIZE_START"] = "on_summarize_start";
-	StepEvents["ON_SUMMARIZE_DELTA"] = "on_summarize_delta";
-	StepEvents["ON_SUMMARIZE_COMPLETE"] = "on_summarize_complete";
-	StepEvents["ON_SUBAGENT_UPDATE"] = "on_subagent_update";
-	return StepEvents;
-}({});
-/** Token-tracking event names streamed to the client (separate from StepEvents dispatch). */
-let UsageEvents = /* @__PURE__ */ function(UsageEvents) {
-	UsageEvents["ON_CONTEXT_USAGE"] = "on_context_usage";
-	UsageEvents["ON_TOKEN_USAGE"] = "on_token_usage";
-	return UsageEvents;
-}({});
-//#endregion
 //#region src/bedrock.ts
 const DEFAULT_ENABLED_MAX_TOKENS = 8192;
 const DEFAULT_THINKING_BUDGET = 2e3;
@@ -217,6 +172,7 @@ const bedrockInputSchema = require_data_service.tConversationSchema.pick({
 	thinkingDisplay: true,
 	reasoning_effort: true,
 	promptCache: true,
+	promptCacheTtl: true,
 	topK: true,
 	additionalModelRequestFields: true
 }).transform((obj) => {
@@ -255,6 +211,7 @@ const bedrockInputParser = require_data_service.tConversationSchema.pick({
 	thinkingDisplay: true,
 	reasoning_effort: true,
 	promptCache: true,
+	promptCacheTtl: true,
 	topK: true,
 	additionalModelRequestFields: true
 }).catchall(zod.z.any()).transform((data) => {
@@ -275,7 +232,8 @@ const bedrockInputParser = require_data_service.tConversationSchema.pick({
 		"temperature",
 		"topP",
 		"stop",
-		"promptCache"
+		"promptCache",
+		"promptCacheTtl"
 	];
 	const additionalFields = {};
 	const typedData = data;
@@ -381,6 +339,12 @@ const bedrockInputParser = require_data_service.tConversationSchema.pick({
 	if (typeof typedData.model === "string" && (typedData.model.includes("claude") || typedData.model.includes("nova"))) {
 		if (typedData.promptCache === void 0) typedData.promptCache = true;
 	} else if (typedData.promptCache === true) typedData.promptCache = void 0;
+	/**
+	* A cache TTL is meaningless without caching — tie it to promptCache. When
+	* caching is off or unsupported for the model (cleared above), drop the TTL
+	* so an unsupported `1h` is never sent on a non-caching Bedrock request.
+	*/
+	if (typedData.promptCache !== true) typedData.promptCacheTtl = void 0;
 	if (Object.keys(additionalFields).length > 0) typedData.additionalModelRequestFields = {
 		...typedData.additionalModelRequestFields || {},
 		...additionalFields
@@ -438,7 +402,100 @@ const bedrockOutputParser = (data) => {
 	return result;
 };
 //#endregion
+//#region src/types/runs.ts
+let ContentTypes = /* @__PURE__ */ function(ContentTypes) {
+	ContentTypes["TEXT"] = "text";
+	ContentTypes["THINK"] = "think";
+	ContentTypes["TEXT_DELTA"] = "text_delta";
+	ContentTypes["TOOL_CALL"] = "tool_call";
+	ContentTypes["IMAGE_FILE"] = "image_file";
+	ContentTypes["IMAGE_URL"] = "image_url";
+	ContentTypes["VIDEO_URL"] = "video_url";
+	ContentTypes["INPUT_AUDIO"] = "input_audio";
+	ContentTypes["AGENT_UPDATE"] = "agent_update";
+	ContentTypes["SUMMARY"] = "summary";
+	ContentTypes["ERROR"] = "error";
+	return ContentTypes;
+}({});
+let StepTypes = /* @__PURE__ */ function(StepTypes) {
+	StepTypes["TOOL_CALLS"] = "tool_calls";
+	StepTypes["MESSAGE_CREATION"] = "message_creation";
+	return StepTypes;
+}({});
+let ToolCallTypes = /* @__PURE__ */ function(ToolCallTypes) {
+	ToolCallTypes["FUNCTION"] = "function";
+	ToolCallTypes["RETRIEVAL"] = "retrieval";
+	ToolCallTypes["FILE_SEARCH"] = "file_search";
+	ToolCallTypes["CODE_INTERPRETER"] = "code_interpreter";
+	ToolCallTypes["TOOL_CALL"] = "tool_call";
+	return ToolCallTypes;
+}({});
+/** Event names dispatched by the agent graph and consumed by step handlers. */
+let StepEvents = /* @__PURE__ */ function(StepEvents) {
+	StepEvents["ON_RUN_STEP"] = "on_run_step";
+	StepEvents["ON_AGENT_UPDATE"] = "on_agent_update";
+	StepEvents["ON_MESSAGE_DELTA"] = "on_message_delta";
+	StepEvents["ON_REASONING_DELTA"] = "on_reasoning_delta";
+	StepEvents["ON_RUN_STEP_DELTA"] = "on_run_step_delta";
+	StepEvents["ON_RUN_STEP_COMPLETED"] = "on_run_step_completed";
+	StepEvents["ON_SUMMARIZE_START"] = "on_summarize_start";
+	StepEvents["ON_SUMMARIZE_DELTA"] = "on_summarize_delta";
+	StepEvents["ON_SUMMARIZE_COMPLETE"] = "on_summarize_complete";
+	StepEvents["ON_SUBAGENT_UPDATE"] = "on_subagent_update";
+	return StepEvents;
+}({});
+/** Token-tracking event names streamed to the client (separate from StepEvents dispatch). */
+let UsageEvents = /* @__PURE__ */ function(UsageEvents) {
+	UsageEvents["ON_CONTEXT_USAGE"] = "on_context_usage";
+	UsageEvents["ON_TOKEN_USAGE"] = "on_token_usage";
+	return UsageEvents;
+}({});
+/**
+* Full prompt token count for one completed model call — the EXACT context the
+* model saw, provider-aware: additive providers (Bedrock) report `input_tokens`
+* excluding cache, so cache reads/writes are added back; subset providers
+* (Anthropic, OpenAI, …) already fold cache into `input_tokens`. When the
+* provider is absent (custom/OpenAI-compatible payloads), fall back to the same
+* magnitude heuristic `normalizeUsageUnits` uses — cache ≤ input means it's
+* already included — so cached events aren't re-inflated. The ground truth the
+* gauge reconciles its calibrated estimate to.
+*/
+const promptTokensFromUsage = (event) => {
+	const input = event.input_tokens ?? 0;
+	const details = event.input_token_details ?? {};
+	const cacheRead = details.cache_read ?? 0;
+	const cacheCreation = details.cache_creation ?? 0;
+	return (event.provider != null ? require_data_service.inputTokensIncludesCache(event.provider) : cacheRead + cacheCreation <= input) ? input : input + cacheRead + cacheCreation;
+};
+/**
+* Reconciles a pre-invoke context snapshot's CALIBRATED estimate to a call's
+* ACTUAL prompt tokens. The SDK's calibration multiplier scales only
+* `messageTokens` (instructions/summary are raw tiktoken counts), and it can
+* over-shoot badly when a provider injects server-side content the SDK never
+* counted (e.g. Anthropic web search) — pinning the gauge several× too high and
+* persisting it. Trust the provider's own prompt count: keep the raw
+* instruction/summary rows, set `messageTokens` to the remainder, and recompute
+* the free space. No-op when `promptTokens` is unusable.
+*/
+const reconcileContextUsage = (snapshot, promptTokens) => {
+	if (!Number.isFinite(promptTokens) || promptTokens <= 0) return snapshot;
+	const { breakdown } = snapshot;
+	const budget = snapshot.contextBudget ?? breakdown.maxContextTokens;
+	const nonMessageTokens = (breakdown.instructionTokens ?? 0) + (breakdown.summaryTokens ?? 0);
+	const messageTokens = Math.max(0, promptTokens - nonMessageTokens);
+	return {
+		...snapshot,
+		breakdown: {
+			...breakdown,
+			messageTokens
+		},
+		remainingContextTokens: budget != null ? Math.max(0, budget - promptTokens) : snapshot.remainingContextTokens
+	};
+};
+//#endregion
 //#region src/parsers.ts
+dayjs.default.extend(dayjs_plugin_utc.default);
+dayjs.default.extend(dayjs_plugin_timezone.default);
 const endpointSchemas = {
 	["openAI"]: require_data_service.openAISchema,
 	["azureOpenAI"]: require_data_service.openAISchema,
@@ -637,10 +694,24 @@ function findLastSeparatorIndex(text, separators = SEPARATORS) {
 	}
 	return lastIndex;
 }
-function replaceSpecialVars({ text, user, now: inputNow }) {
+/**
+* Anchors a dayjs instant to the user's IANA timezone when one is supplied,
+* so local-time special vars reflect the user's wall clock rather than the
+* server's. Falls back to the original instant for missing or invalid zones.
+*/
+function applyTimezone(value, timezone) {
+	if (!timezone) return value;
+	try {
+		const zoned = value.tz(timezone);
+		return zoned.isValid() ? zoned : value;
+	} catch {
+		return value;
+	}
+}
+function replaceSpecialVars({ text, user, now: inputNow, timezone }) {
 	let result = text;
 	if (!result) return result;
-	const now = inputNow != null ? (0, dayjs.default)(inputNow) : (0, dayjs.default)();
+	const now = applyTimezone(inputNow != null ? (0, dayjs.default)(inputNow) : (0, dayjs.default)(), timezone);
 	const weekdayName = now.format("dddd");
 	const currentDate = now.format("YYYY-MM-DD");
 	result = result.replace(/{{\s*current_date\s*}}/gi, `${currentDate} (${weekdayName})`);
@@ -4992,6 +5063,14 @@ function validateAndParseOpenAPISpec(specString) {
 }
 //#endregion
 //#region src/createPayload.ts
+/** Resolves the browser's IANA timezone so the server can localize prompt variables. */
+function getUserTimezone() {
+	try {
+		return Intl.DateTimeFormat().resolvedOptions().timeZone || void 0;
+	} catch {
+		return;
+	}
+}
 function createPayload(submission) {
 	const { isEdited, addedConvo, userMessage, isContinued, isTemporary, isRegenerate, conversation, editedContent, ephemeralAgent, endpointOption, manualSkills } = submission;
 	const { conversationId } = require_data_service.tConvoUpdateSchema.parse(conversation);
@@ -5010,7 +5089,8 @@ function createPayload(submission) {
 		conversationId,
 		isContinued: !!(isEdited && isContinued),
 		ephemeralAgent: require_data_service.isAssistantsEndpoint(endpoint) ? void 0 : ephemeralAgent,
-		manualSkills: require_data_service.isAssistantsEndpoint(endpoint) ? void 0 : manualSkills
+		manualSkills: require_data_service.isAssistantsEndpoint(endpoint) ? void 0 : manualSkills,
+		timezone: getUserTimezone()
 	};
 	return {
 		server,
@@ -5414,6 +5494,22 @@ const anthropic = {
 		showDefault: false,
 		columnSpan: 2
 	},
+	promptCacheTtl: {
+		key: "promptCacheTtl",
+		label: "com_endpoint_prompt_cache_ttl",
+		labelCode: true,
+		description: "com_endpoint_anthropic_prompt_cache_ttl",
+		descriptionCode: true,
+		type: "enum",
+		default: require_data_service.anthropicSettings.promptCacheTtl.default,
+		options: ["5m", "1h"],
+		component: "dropdown",
+		optionType: "conversation",
+		showDefault: false,
+		placeholder: "com_endpoint_prompt_cache_ttl_default",
+		placeholderCode: true,
+		columnSpan: 2
+	},
 	thinking: {
 		key: "thinking",
 		label: "com_endpoint_thinking",
@@ -5569,6 +5665,22 @@ const bedrock = {
 		showDefault: false,
 		columnSpan: 2
 	},
+	promptCacheTtl: {
+		key: "promptCacheTtl",
+		label: "com_endpoint_prompt_cache_ttl",
+		labelCode: true,
+		description: "com_endpoint_anthropic_prompt_cache_ttl",
+		descriptionCode: true,
+		type: "enum",
+		default: void 0,
+		options: ["5m", "1h"],
+		component: "dropdown",
+		optionType: "conversation",
+		showDefault: false,
+		placeholder: "com_endpoint_prompt_cache_ttl_default",
+		placeholderCode: true,
+		columnSpan: 2
+	},
 	reasoning_effort: {
 		key: "reasoning_effort",
 		label: "com_endpoint_reasoning_effort",
@@ -5822,7 +5934,11 @@ const openAI = [
 	openAIParams.disableStreaming,
 	librechat.fileTokenLimit
 ];
-const openRouter = [...openAI, anthropic.promptCache];
+const openRouter = [
+	...openAI,
+	anthropic.promptCache,
+	anthropic.promptCacheTtl
+];
 const openAICol1 = [
 	baseDefinitions.model,
 	librechat.modelLabel,
@@ -5856,6 +5972,7 @@ const anthropicConfig = [
 	anthropic.topK,
 	librechat.resendFiles,
 	anthropic.promptCache,
+	anthropic.promptCacheTtl,
 	anthropic.thinking,
 	anthropic.thinkingBudget,
 	anthropic.effort,
@@ -5876,6 +5993,7 @@ const anthropicCol2 = [
 	anthropic.topK,
 	librechat.resendFiles,
 	anthropic.promptCache,
+	anthropic.promptCacheTtl,
 	anthropic.thinking,
 	anthropic.thinkingBudget,
 	anthropic.effort,
@@ -5895,6 +6013,7 @@ const bedrockAnthropic = [
 	librechat.resendFiles,
 	bedrock.region,
 	bedrock.promptCache,
+	bedrock.promptCacheTtl,
 	anthropic.thinking,
 	anthropic.thinkingBudget,
 	anthropic.effort,
@@ -5932,6 +6051,7 @@ const bedrockGeneral = [
 	librechat.resendFiles,
 	bedrock.region,
 	bedrock.promptCache,
+	bedrock.promptCacheTtl,
 	librechat.fileTokenLimit
 ];
 const bedrockAnthropicCol1 = [
@@ -5949,6 +6069,7 @@ const bedrockAnthropicCol2 = [
 	librechat.resendFiles,
 	bedrock.region,
 	bedrock.promptCache,
+	bedrock.promptCacheTtl,
 	anthropic.thinking,
 	anthropic.thinkingBudget,
 	anthropic.effort,
@@ -5995,6 +6116,7 @@ const bedrockGeneralCol2 = [
 	librechat.resendFiles,
 	bedrock.region,
 	bedrock.promptCache,
+	bedrock.promptCacheTtl,
 	librechat.fileTokenLimit
 ];
 const bedrockZAI = [
@@ -6084,7 +6206,11 @@ const presetSettings = {
 	["custom"]: openAIColumns,
 	["openrouter"]: {
 		col1: openAICol1,
-		col2: [...openAICol2, anthropic.promptCache]
+		col2: [
+			...openAICol2,
+			anthropic.promptCache,
+			anthropic.promptCacheTtl
+		]
 	},
 	["anthropic"]: {
 		col1: anthropicCol1,
@@ -6515,9 +6641,11 @@ exports.presetSettings = presetSettings;
 exports.principalSchema = require_data_service.principalSchema;
 exports.progress = progress;
 exports.promptPermissionsSchema = promptPermissionsSchema;
+exports.promptTokensFromUsage = promptTokensFromUsage;
 exports.providerEndpointMap = require_data_service.providerEndpointMap;
 exports.radioGroup = radioGroup;
 exports.rateLimitSchema = require_data_service.rateLimitSchema;
+exports.reconcileContextUsage = reconcileContextUsage;
 exports.registerPage = require_data_service.registerPage;
 exports.remoteAgentsPermissionsSchema = remoteAgentsPermissionsSchema;
 exports.removeNullishValues = require_data_service.removeNullishValues;