@nick3/copilot-api 1.2.1 → 1.2.4

package/dist/{server-B6nFgTkW.js → server-Cxlbm6kJ.js}

@@ -1,10 +1,10 @@
-import { HTTPError, PATHS, accountFromState, accountsManager, copilotBaseUrl, copilotHeaders, forwardError, getAliasTargetSet, getConfig, getCopilotUsage, getExtraPromptForModel, getModelAliases, getModelAliasesInfo, getModelRefreshIntervalMs, getReasoningEffortForModel, getSmallModel, isForceAgentEnabled, isFreeModelLoadBalancingEnabled, isMessageStartInputTokensFallbackEnabled, isNullish, listAccountsFromRegistry, mergeConfigWithDefaults, shouldCompactUseSmallModel, sleep, state } from "./accounts-manager-BY_xYGQ7.js";
+import { HTTPError, PATHS, accountFromState, accountsManager, copilotBaseUrl, copilotHeaders, forwardError, generateRequestIdFromPayload, getAliasTargetSet, getConfig, getCopilotUsage, getExtraPromptForModel, getModelAliases, getModelAliasesInfo, getModelRefreshIntervalMs, getReasoningEffortForModel, getRootSessionId, getSmallModel, getUUID, isForceAgentEnabled, isFreeModelLoadBalancingEnabled, isMessageStartInputTokensFallbackEnabled, isNullish, isResponsesApiContextManagementModel, listAccountsFromRegistry, mergeConfigWithDefaults, shouldCompactUseSmallModel, sleep, state } from "./accounts-manager-eec8Wj3_.js";
 import consola from "consola";
 import fs, { readFile } from "node:fs/promises";
 import * as path$1 from "node:path";
 import path from "node:path";
 import { randomUUID, timingSafeEqual } from "node:crypto";
-import fs$1, { existsSync, readFileSync } from "node:fs";
+import fs$1, { existsSync } from "node:fs";
 import { Hono } from "hono";
 import { cors } from "hono/cors";
 import { logger } from "hono/logger";
@@ -2227,13 +2227,18 @@ const getTokenCount = async (payload, model) => {
 
 //#endregion
 //#region src/services/copilot/create-responses.ts
-const createResponses = async (payload, { vision, initiator, upstreamRequestId }, account) => {
+const createResponses = async (payload, { vision, initiator, upstreamRequestId, subagentMarker, sessionId }, account) => {
 	const ctx = account ?? accountFromState();
 	if (!ctx.copilotToken) throw new Error("Copilot token not found");
 	const headers = {
 		...copilotHeaders(ctx, vision, upstreamRequestId),
-		"X-Initiator": initiator
+		"x-initiator": initiator
 	};
+	if (subagentMarker) {
+		headers["x-initiator"] = "agent";
+		headers["x-interaction-type"] = "conversation-subagent";
+	}
+	if (sessionId) headers["x-interaction-id"] = sessionId;
 	payload.service_tier = null;
 	const response = await fetch(`${copilotBaseUrl(ctx)}/responses`, {
 		method: "POST",
@@ -2251,12 +2256,14 @@ const createResponses = async (payload, { vision, initiator, upstreamRequestId }
 //#endregion
 //#region src/routes/messages/responses-translation.ts
 const MESSAGE_TYPE = "message";
-const CODEX_PHASE_MODEL = "gpt-5.3-codex";
+const COMPACTION_SIGNATURE_PREFIX = "cm1#";
+const COMPACTION_SIGNATURE_SEPARATOR = "@";
 const THINKING_TEXT$1 = "Thinking...";
 const translateAnthropicMessagesToResponsesPayload = (payload, modelOverride) => {
 	const model = modelOverride ?? payload.model;
 	const input = [];
-	for (const message of payload.messages) input.push(...translateMessage(message, payload.model));
+	const applyPhase = shouldApplyPhase(payload.model);
+	for (const message of payload.messages) input.push(...translateMessage(message, payload.model, applyPhase));
 	const translatedTools = convertAnthropicTools(payload.tools);
 	const toolChoice = convertAnthropicToolChoice(payload.tool_choice);
 	const { safetyIdentifier, promptCacheKey } = parseUserId(payload.metadata?.user_id);
@@ -2282,9 +2289,26 @@ const translateAnthropicMessagesToResponsesPayload = (payload, modelOverride) =>
 		include: ["reasoning.encrypted_content"]
 	};
 };
-const translateMessage = (message, model) => {
+const encodeCompactionCarrierSignature = (compaction) => {
+	return `${COMPACTION_SIGNATURE_PREFIX}${compaction.encrypted_content}${COMPACTION_SIGNATURE_SEPARATOR}${compaction.id}`;
+};
+const decodeCompactionCarrierSignature = (signature) => {
+	if (signature.startsWith(COMPACTION_SIGNATURE_PREFIX)) {
+		const raw = signature.slice(4);
+		const separatorIndex = raw.indexOf(COMPACTION_SIGNATURE_SEPARATOR);
+		if (separatorIndex <= 0 || separatorIndex === raw.length - 1) return;
+		const encrypted_content = raw.slice(0, separatorIndex);
+		const id = raw.slice(separatorIndex + 1);
+		if (!encrypted_content) return;
+		return {
+			id,
+			encrypted_content
+		};
+	}
+};
+const translateMessage = (message, model, applyPhase) => {
 	if (message.role === "user") return translateUserMessage(message);
-	return translateAssistantMessage(message, model);
+	return translateAssistantMessage(message, model, applyPhase);
 };
 const translateUserMessage = (message) => {
 	if (typeof message.content === "string") return [createMessage("user", message.content)];
@@ -2303,8 +2327,8 @@ const translateUserMessage = (message) => {
 	flushPendingContent(pendingContent, items, { role: "user" });
 	return items;
 };
-const translateAssistantMessage = (message, model) => {
-	const assistantPhase = resolveAssistantPhase(model, message.content);
+const translateAssistantMessage = (message, model, applyPhase) => {
+	const assistantPhase = resolveAssistantPhase(model, message.content, applyPhase);
 	if (typeof message.content === "string") return [createMessage("assistant", message.content, assistantPhase)];
 	if (!Array.isArray(message.content)) return [];
 	const items = [];
@@ -2318,13 +2342,24 @@ const translateAssistantMessage = (message, model) => {
 			items.push(createFunctionToolCall(block));
 			continue;
 		}
-		if (block.type === "thinking" && block.signature && block.signature.includes("@")) {
-			flushPendingContent(pendingContent, items, {
-				role: "assistant",
-				phase: assistantPhase
-			});
-			items.push(createReasoningContent(block));
-			continue;
+		if (block.type === "thinking" && block.signature) {
+			const compactionContent = createCompactionContent(block);
+			if (compactionContent) {
+				flushPendingContent(pendingContent, items, {
+					role: "assistant",
+					phase: assistantPhase
+				});
+				items.push(compactionContent);
+				continue;
+			}
+			if (block.signature.includes("@")) {
+				flushPendingContent(pendingContent, items, {
+					role: "assistant",
+					phase: assistantPhase
+				});
+				items.push(createReasoningContent(block));
+				continue;
+			}
 		}
 		const converted = translateAssistantContentBlock(block);
 		if (converted) pendingContent.push(converted);
@@ -2360,14 +2395,16 @@ const createMessage = (role, content, phase) => ({
 	content,
 	...role === "assistant" && phase ? { phase } : {}
 });
-const resolveAssistantPhase = (model, content) => {
-	if (!shouldApplyCodexPhase(model)) return;
+const resolveAssistantPhase = (_model, content, applyPhase) => {
+	if (!applyPhase) return;
 	if (typeof content === "string") return "final_answer";
 	if (!Array.isArray(content)) return;
 	if (!content.some((block) => block.type === "text")) return;
 	return content.some((block) => block.type === "tool_use") ? "commentary" : "final_answer";
 };
-const shouldApplyCodexPhase = (model) => model === CODEX_PHASE_MODEL;
+const shouldApplyPhase = (model) => {
+	return getExtraPromptForModel(model).includes("## Intermediary updates");
+};
 const createTextContent = (text) => ({
 	type: "input_text",
 	text
@@ -2382,9 +2419,7 @@ const createImageContent = (block) => ({
 	detail: "auto"
 });
 const createReasoningContent = (block) => {
-	const array = block.signature.split("@");
-	const signature = array[0];
-	const id = array[1];
+	const { encryptedContent, id } = parseReasoningSignature(block.signature);
 	const thinking = block.thinking === THINKING_TEXT$1 ? "" : block.thinking;
 	return {
 		id,
@@ -2393,7 +2428,27 @@ const createReasoningContent = (block) => {
 			type: "summary_text",
 			text: thinking
 		}] : [],
-		encrypted_content: signature
+		encrypted_content: encryptedContent
+	};
+};
+const createCompactionContent = (block) => {
+	const compaction = decodeCompactionCarrierSignature(block.signature);
+	if (!compaction) return;
+	return {
+		id: compaction.id,
+		type: "compaction",
+		encrypted_content: compaction.encrypted_content
+	};
+};
+const parseReasoningSignature = (signature) => {
+	const splitIndex = signature.lastIndexOf("@");
+	if (splitIndex <= 0 || splitIndex === signature.length - 1) return {
+		encryptedContent: signature,
+		id: ""
+	};
+	return {
+		encryptedContent: signature.slice(0, splitIndex),
+		id: signature.slice(splitIndex + 1)
 	};
 };
 const createFunctionToolCall = (block) => ({
@@ -2484,6 +2539,11 @@ const mapOutputToAnthropicContent = (output) => {
 			});
 			break;
 		}
+		case "compaction": {
+			const compactionBlock = createCompactionThinkingBlock(item);
+			if (compactionBlock) contentBlocks.push(compactionBlock);
+			break;
+		}
 		default: {
 			const combinedText = combineMessageTextContent(item.content);
 			if (combinedText.length > 0) contentBlocks.push({
@@ -2541,6 +2601,17 @@ const createToolUseContentBlock = (call) => {
 		input
 	};
 };
+const createCompactionThinkingBlock = (item) => {
+	if (!item.id || !item.encrypted_content) return null;
+	return {
+		type: "thinking",
+		thinking: THINKING_TEXT$1,
+		signature: encodeCompactionCarrierSignature({
+			id: item.id,
+			encrypted_content: item.encrypted_content
+		})
+	};
+};
 const parseFunctionCallArguments = (rawArguments) => {
 	if (typeof rawArguments !== "string" || rawArguments.trim().length === 0) return {};
 	try {
@@ -2645,8 +2716,10 @@ const createChatCompletions = async (payload, account, options) => {
 	const initiator = options?.initiator ?? getChatInitiator(payload.messages);
 	const headers = {
 		...copilotHeaders(ctx, enableVision, options?.upstreamRequestId),
-		"X-Initiator": initiator
+		"x-initiator": options?.subagentMarker ? "agent" : initiator
 	};
+	if (options?.subagentMarker) headers["x-interaction-type"] = "conversation-subagent";
+	if (options?.sessionId) headers["x-interaction-id"] = options.sessionId;
 	const upstreamPayload = applyDefaultReasoningEffort(payload);
 	const response = await fetch(`${copilotBaseUrl(ctx)}/chat/completions`, {
 		method: "POST",
@@ -2724,7 +2797,10 @@ async function handleCompletion$1(c) {
 	if (state.manualApprove) await awaitApproval();
 	const payloadWithMaxTokens = applyDefaultMaxTokens(upstreamPayload, selectedModel);
 	const accountCtx = toAccountContext(account);
-	request.upstreamRequestId = randomUUID();
+	const upstreamRequestId = generateRequestIdFromPayload(payloadWithMaxTokens);
+	const upstreamSessionId = getUUID(upstreamRequestId);
+	request.upstreamRequestId = upstreamRequestId;
+	request.upstreamSessionId = upstreamSessionId;
 	if (streamRequested) return handleStreamingRequest({
 		c,
 		store,
@@ -2827,7 +2903,10 @@ async function handleStreamingRequest(params) {
 	const { c, store, request, payload, selection, accountCtx, clientModel, premiumRemainingBefore, premiumUnlimitedBefore } = params;
 	let response;
 	try {
-		response = await createChatCompletions(payload, accountCtx, { upstreamRequestId: request.upstreamRequestId });
+		response = await createChatCompletions(payload, accountCtx, {
+			upstreamRequestId: request.upstreamRequestId,
+			sessionId: request.upstreamSessionId
+		});
 	} catch (error) {
 		return handleUpstreamCreateError$1({
 			store,
@@ -3011,7 +3090,10 @@ async function handleNonStreamingRequest(params) {
 	let errorMessage;
 	let finishedAtMs;
 	try {
-		const response = await createChatCompletions(payload, accountCtx, { upstreamRequestId: request.upstreamRequestId });
+		const response = await createChatCompletions(payload, accountCtx, {
+			upstreamRequestId: request.upstreamRequestId,
+			sessionId: request.upstreamSessionId
+		});
 		finishedAtMs = Date.now();
 		if (!isNonStreaming$1(response)) {
 			logger$3.debug("Unexpected streaming response");
@@ -3714,8 +3796,34 @@ const handleOutputItemAdded$1 = (rawEvent, state$1) => {
 const handleOutputItemDone$1 = (rawEvent, state$1) => {
 	const events$1 = new Array();
 	const item = rawEvent.item;
-	if (item.type !== "reasoning") return events$1;
+	const itemType = item.type;
 	const outputIndex = rawEvent.output_index;
+	if (itemType === "compaction") {
+		if (!item.id || !item.encrypted_content) return events$1;
+		const blockIndex$1 = openThinkingBlockIfNeeded(state$1, outputIndex, events$1);
+		if (!state$1.blockHasDelta.has(blockIndex$1)) events$1.push({
+			type: "content_block_delta",
+			index: blockIndex$1,
+			delta: {
+				type: "thinking_delta",
+				thinking: THINKING_TEXT$1
+			}
+		});
+		events$1.push({
+			type: "content_block_delta",
+			index: blockIndex$1,
+			delta: {
+				type: "signature_delta",
+				signature: encodeCompactionCarrierSignature({
+					id: item.id,
+					encrypted_content: item.encrypted_content
+				})
+			}
+		});
+		state$1.blockHasDelta.add(blockIndex$1);
+		return events$1;
+	}
+	if (itemType !== "reasoning") return events$1;
 	const blockIndex = openThinkingBlockIfNeeded(state$1, outputIndex, events$1);
 	const signature = (item.encrypted_content ?? "") + "@" + item.id;
 	if (signature) {
@@ -4063,6 +4171,31 @@ const isAgentRole = (item) => {
 const hasVisionInput = (payload) => {
 	return getPayloadItems(payload).some((item) => containsVisionContent(item));
 };
+const resolveResponsesCompactThreshold = (maxPromptTokens) => {
+	if (typeof maxPromptTokens === "number" && maxPromptTokens > 0) return Math.floor(maxPromptTokens * .9);
+	return 5e4;
+};
+const createCompactionContextManagement = (compactThreshold) => [{
+	type: "compaction",
+	compact_threshold: compactThreshold
+}];
+const applyResponsesApiContextManagement = (payload, maxPromptTokens) => {
+	if (payload.context_management !== void 0) return;
+	if (!isResponsesApiContextManagementModel(payload.model)) return;
+	payload.context_management = createCompactionContextManagement(resolveResponsesCompactThreshold(maxPromptTokens));
+};
+const compactInputByLatestCompaction = (payload) => {
+	if (!Array.isArray(payload.input) || payload.input.length === 0) return;
+	const latestCompactionMessageIndex = getLatestCompactionMessageIndex(payload.input);
+	if (latestCompactionMessageIndex === void 0) return;
+	payload.input = payload.input.slice(latestCompactionMessageIndex);
+};
+const getLatestCompactionMessageIndex = (input) => {
+	for (let index = input.length - 1; index >= 0; index -= 1) if (isCompactionInputItem(input[index])) return index;
+};
+const isCompactionInputItem = (value) => {
+	return "type" in value && typeof value.type === "string" && value.type === "compaction";
+};
 const getPayloadItems = (payload) => {
 	const result = [];
 	const { input } = payload;
@@ -4111,8 +4244,10 @@ const createMessages = async (payload, account, options) => {
 	const initiator = options?.initiator ?? getMessagesInitiator(payload);
 	const headers = {
 		...copilotHeaders(ctx, enableVision, options?.upstreamRequestId),
-		"X-Initiator": initiator
+		"x-initiator": options?.subagentMarker ? "agent" : initiator
 	};
+	if (options?.subagentMarker) headers["x-interaction-type"] = "conversation-subagent";
+	if (options?.sessionId) headers["x-interaction-id"] = options.sessionId;
 	const anthropicBeta = buildAnthropicBetaHeader(options?.anthropicBetaHeader, payload.thinking);
 	if (anthropicBeta) headers["anthropic-beta"] = anthropicBeta;
 	const response = await fetch(`${copilotBaseUrl(ctx)}/v1/messages`, {
@@ -4451,6 +4586,8 @@ async function handleCompletion(c) {
 	const subagentMarker = parseSubagentMarkerFromFirstUser(anthropicPayload);
 	const initiatorOverride = subagentMarker ? "agent" : void 0;
 	if (subagentMarker) logger$2.debug("Detected Subagent marker:", JSON.stringify(subagentMarker));
+	const sessionId = getRootSessionId(anthropicPayload, c);
+	logger$2.debug("Extracted session ID:", sessionId);
 	const anthropicBeta = c.req.header("anthropic-beta");
 	const isCompact = isCompactRequest(anthropicPayload);
 	if (anthropicBeta && isWarmupProbeRequest(anthropicPayload)) anthropicPayload.model = getSmallModel();
@@ -4458,6 +4595,8 @@ async function handleCompletion(c) {
 		logger$2.debug("Is compact request:", isCompact);
 		if (shouldCompactUseSmallModel()) anthropicPayload.model = getSmallModel();
 	} else mergeToolResultForClaude(anthropicPayload);
+	const upstreamRequestId = generateRequestIdFromPayload(anthropicPayload, sessionId);
+	logger$2.debug("Generated request ID:", upstreamRequestId);
 	const clientModel = anthropicPayload.model;
 	const streamRequested = Boolean(anthropicPayload.stream);
 	const rawUserId = anthropicPayload.metadata?.user_id;
@@ -4541,6 +4680,7 @@ async function handleCompletion(c) {
 		upstreamEndpoint: endpoint,
 		upstreamModel: selectedModel.id,
 		costUnits,
+		upstreamRequestId,
 		premiumRemainingBefore,
 		premiumUnlimitedBefore
 	};
@@ -4549,6 +4689,8 @@ async function handleCompletion(c) {
 		anthropicPayload,
 		anthropicBetaHeader: anthropicBeta ?? void 0,
 		initiatorOverride,
+		subagentMarker,
+		sessionId,
 		instr,
 		selectedModel
 	});
@@ -4557,6 +4699,8 @@ async function handleCompletion(c) {
 		anthropicPayload,
 		openAIPayload,
 		initiatorOverride,
+		subagentMarker,
+		sessionId,
 		selectedModel,
 		instr
 	});
@@ -4564,23 +4708,25 @@ async function handleCompletion(c) {
 		c,
 		openAIPayload,
 		initiatorOverride,
+		subagentMarker,
+		sessionId,
 		selectedModel,
 		instr
 	});
 }
 const handleWithChatCompletions = async (params) => {
-	const { c, openAIPayload, initiatorOverride, selectedModel, instr } = params;
+	const { c, openAIPayload, initiatorOverride, subagentMarker, sessionId, selectedModel, instr } = params;
 	logger$2.debug("Translated OpenAI request payload:", JSON.stringify(openAIPayload));
 	const ctx = toAccountContext(instr.account);
 	const initiator = initiatorOverride ?? getChatInitiator(openAIPayload.messages);
-	const upstreamRequestId = randomUUID();
 	instr.initiator = initiator;
-	instr.upstreamRequestId = upstreamRequestId;
 	let response;
 	try {
 		response = await createChatCompletions(openAIPayload, ctx, {
-			upstreamRequestId,
-			initiator
+			upstreamRequestId: instr.upstreamRequestId,
+			initiator,
+			subagentMarker,
+			sessionId
 		});
 	} catch (error) {
 		return await handleChatCompletionsCreateError({
@@ -4611,21 +4757,23 @@ const handleWithChatCompletions = async (params) => {
 	}));
 };
 const handleWithResponsesApi = async (params) => {
-	const { c, anthropicPayload, openAIPayload, initiatorOverride, selectedModel, instr } = params;
+	const { c, anthropicPayload, openAIPayload, initiatorOverride, subagentMarker, sessionId, selectedModel, instr } = params;
 	const responsesPayload = translateAnthropicMessagesToResponsesPayload(anthropicPayload, selectedModel.id);
+	applyResponsesApiContextManagement(responsesPayload, selectedModel.capabilities.limits.max_prompt_tokens);
+	compactInputByLatestCompaction(responsesPayload);
 	logger$2.debug("Translated Responses payload:", JSON.stringify(responsesPayload));
 	const { vision, initiator } = getResponsesRequestOptions(responsesPayload);
 	const resolvedInitiator = initiatorOverride ?? initiator;
 	const ctx = toAccountContext(instr.account);
-	const upstreamRequestId = randomUUID();
 	instr.initiator = resolvedInitiator;
-	instr.upstreamRequestId = upstreamRequestId;
 	let response;
 	try {
 		response = await createResponses(responsesPayload, {
 			vision,
 			initiator: resolvedInitiator,
-			upstreamRequestId
+			upstreamRequestId: instr.upstreamRequestId,
+			subagentMarker,
+			sessionId
 		}, ctx);
 	} catch (error) {
 		return await handleResponsesCreateError({
@@ -5076,27 +5224,31 @@ async function streamMessagesAndLog(params) {
 	}
 }
 const handleWithMessagesApi = async (params) => {
-	const { c, anthropicPayload, anthropicBetaHeader, initiatorOverride, instr, selectedModel } = params;
+	const { c, anthropicPayload, anthropicBetaHeader, initiatorOverride, subagentMarker, sessionId, instr, selectedModel } = params;
 	for (const msg of anthropicPayload.messages) if (msg.role === "assistant" && Array.isArray(msg.content)) msg.content = msg.content.filter((block) => {
 		if (block.type !== "thinking") return true;
 		return block.thinking && block.thinking !== "Thinking..." && block.signature && !block.signature.includes("@");
 	});
-	if (selectedModel.capabilities.supports.adaptive_thinking) {
+	const toolChoice = anthropicPayload.tool_choice;
+	if (toolChoice?.type === "any" || toolChoice?.type === "tool") {
+		delete anthropicPayload.thinking;
+		delete anthropicPayload.output_config;
+	} else if (selectedModel.capabilities.supports.adaptive_thinking) {
 		anthropicPayload.thinking = { type: "adaptive" };
 		anthropicPayload.output_config = { effort: getAnthropicEffortForModel(anthropicPayload.model) };
 	}
 	logger$2.debug("Translated Messages payload:", JSON.stringify(anthropicPayload));
 	const ctx = toAccountContext(instr.account);
-	const upstreamRequestId = randomUUID();
 	const initiator = initiatorOverride ?? getMessagesInitiator(anthropicPayload);
 	instr.initiator = initiator;
-	instr.upstreamRequestId = upstreamRequestId;
 	let response;
 	try {
 		response = await createMessages(anthropicPayload, ctx, {
 			anthropicBetaHeader,
-			upstreamRequestId,
-			initiator
+			upstreamRequestId: instr.upstreamRequestId,
+			initiator,
+			subagentMarker,
+			sessionId
 		});
 	} catch (error) {
 		return await handleMessagesCreateError({
@@ -5239,6 +5391,7 @@ const handleResponses = async (c) => {
 	const clientModel = payload.model;
 	logger$1.debug("Responses request payload:", JSON.stringify(payload));
 	removeWebSearchTool(payload);
+	compactInputByLatestCompaction(payload);
 	const streamRequested = Boolean(payload.stream);
 	const { initiator: initialInitiator } = getResponsesRequestOptions(payload);
 	const userId = payload.metadata?.user_id;
@@ -5280,13 +5433,18 @@ const handleResponses = async (c) => {
 		model: selectedModel.id
 	};
 	useFunctionApplyPatch(upstreamPayload);
+	applyResponsesApiContextManagement(upstreamPayload, selectedModel.capabilities.limits.max_prompt_tokens);
+	compactInputByLatestCompaction(upstreamPayload);
 	const premiumRemainingBefore = account.premiumRemaining;
 	const premiumUnlimitedBefore = account.unlimited;
 	const { vision, initiator } = getResponsesRequestOptions(upstreamPayload);
 	request.initiator = initiator;
 	if (state.manualApprove) await awaitApproval();
 	const accountCtx = toAccountContext(account);
-	request.upstreamRequestId = randomUUID();
+	const upstreamRequestId = generateRequestIdFromPayload({ messages: upstreamPayload.input });
+	const upstreamSessionId = getUUID(upstreamRequestId);
+	request.upstreamRequestId = upstreamRequestId;
+	request.upstreamSessionId = upstreamSessionId;
 	if (streamRequested) return handleStreamingResponses({
 		c,
 		store,
@@ -5397,7 +5555,8 @@ async function handleStreamingResponses(params) {
 		response = await createResponses(payload, {
 			vision,
 			initiator,
-			upstreamRequestId: request.upstreamRequestId
+			upstreamRequestId: request.upstreamRequestId,
+			sessionId: request.upstreamSessionId
 		}, accountCtx);
 	} catch (error) {
 		return handleUpstreamCreateError({
@@ -5582,7 +5741,8 @@ async function handleNonStreamingResponses(params) {
 		const response = await createResponses(payload, {
 			vision,
 			initiator,
-			upstreamRequestId: request.upstreamRequestId
+			upstreamRequestId: request.upstreamRequestId,
+			sessionId: request.upstreamSessionId
 		}, accountCtx);
 		finishedAtMs = Date.now();
 		const streamResponse = handleUnexpectedResponsesStream(c, response);
@@ -5745,19 +5905,10 @@ const server = new Hono();
 server.use(logger());
 server.use(cors());
 server.use("*", createAuthMiddleware({
-	allowUnauthenticatedPaths: [
-		"/",
-		"/usage-viewer",
-		"/usage-viewer/"
-	],
+	allowUnauthenticatedPaths: ["/"],
 	allowUnauthenticatedPathPrefixes: ["/admin", "/api/admin"]
 }));
 server.get("/", (c) => c.text("Server running"));
-server.get("/usage-viewer", (c) => {
-	const usageViewerFileUrl = new URL("../pages/index.html", import.meta.url);
-	return c.html(readFileSync(usageViewerFileUrl, "utf8"));
-});
-server.get("/usage-viewer/", (c) => c.redirect("/usage-viewer", 301));
 server.route("/chat/completions", completionRoutes);
 server.route("/models", modelRoutes);
 server.route("/embeddings", embeddingRoutes);
@@ -5774,4 +5925,4 @@ server.route("/v1/messages", messageRoutes);
 
 //#endregion
 export { server };
-//# sourceMappingURL=server-B6nFgTkW.js.map
+//# sourceMappingURL=server-Cxlbm6kJ.js.map