@jeffreycao/copilot-api 2.7.36 → 2.7.38

package/dist/{server-BogSoFE5.js → server-CEF4Sh-g.js}

@@ -1,9 +1,9 @@
 import { D as state, E as isGitHubCopilotEnabled, O as HTTPError, S as PATHS, a as getAnthropicApiKey, c as getConfig, d as getUpstreamTransportConfig, g as isResponsesApiWebSocketEnabled, h as isResponsesApiWebSearchEnabled, i as getAlphaSearchModel, j as forwardError, l as getMessageApiWebSearchModel, m as isMessagesApiEnabled, o as getClaudeAutoModel, p as isAlphaSearchCodexPriorityEnabled, s as getClaudeTokenMultiplier, u as getOpencodeModelContextWindow, w as GITHUB_COPILOT_PROVIDER, y as reloadConfig } from "./config-store-CYG9aCTL.js";
-import { A as resolveMappedModel, C as getModelResponsesApiCompactThreshold$1, D as isContextManagementEnabledForMessages, E as getSmallModelForProvider, O as isContextManagementEnabledForResponses, S as getModelMappings, T as getSmallModel, _ as getModelsDevModelMaxOutputTokens, a as normalizeProviderBaseUrl, d as isProviderModelVisible, f as resolveProviderModelsDevId, g as getModelsDevModel, h as getBuiltinProviderModelRecords, i as listEnabledProviders, j as setModelMappings, k as isGpt56OrAbove, l as getLocalProviderModelRecords, m as builtinProviderModelRegistry, n as getRawProviderConfig, o as resolveEffectiveProviderType, p as usesModelsDevModelCatalog, s as resolveProviderConfigForModel, t as getProviderConfig, u as getProviderModelConfig, v as getModelsDevModelPricing, w as getReasoningEffortForModel, x as getExtraPromptForModel } from "./config-D8pUyhJp.js";
-import { $ as forwardCodexResponses, B as getXaiAccessToken, C as copilotHeaders, D as prepareForCompact, E as isOpencodeOauthApp, F as compactTextOnlyGuard, I as createAuthMiddleware, J as XAI_API_BASE_URL, K as readGitHubToken, L as getConfiguredAdminApiKeys, M as compactMessageSections, O as prepareInteractionHeaders, P as compactSystemPromptStarts, Q as buildCodexRequestHeaders, S as copilotBaseUrl, T as copilotWebSocketHeaders, Z as CODEX_API_BASE_URL, _ as isAsyncIterable, at as createResponsesSafeStream, b as parseUserIdMetadata, ct as createPooledWebSocketStream, d as generateRequestIdFromPayload, dt as stripInternalRequestHeaders, et as generateTraceId, f as getPromptCacheKey, g as getUUID, h as getSessionAffinityFromHeaders, it as createResponsesHttpEventStream, j as compactAutoContinuePromptStarts, k as prepareMessageProxyHeaders, lt as createWebSocketUrl, m as getRootSessionIdFromHeaders, n as invalidateCodexRuntime, nt as resolveTraceId$1, ot as encodePoolKeyPart, p as getRootSessionId, q as readGitHubTokenFromEnv, rt as fetchUpstreamWithLifecycle, s as setupCodexToken, st as getResponsesStreamTerminalDisposition, tt as requestContext, u as stopCopilotRefreshLoop, v as isNullish, x as getCopilotUsage, y as isResponsesStream } from "./token-BL6FigKp.js";
+import { A as resolveMappedModel, C as getModelResponsesApiCompactThreshold$1, D as isContextManagementEnabledForMessages, E as getSmallModelForProvider, O as isContextManagementEnabledForResponses, S as getModelMappings, T as getSmallModel, _ as getModelsDevModelMaxOutputTokens, a as normalizeProviderBaseUrl, d as isProviderModelVisible, f as resolveProviderModelsDevId, g as getModelsDevModel, h as getBuiltinProviderModelRecords, i as listEnabledProviders, j as setModelMappings, k as isGpt56OrAbove, l as getLocalProviderModelRecords, m as builtinProviderModelRegistry, n as getRawProviderConfig, o as resolveEffectiveProviderType, p as usesModelsDevModelCatalog, s as resolveProviderConfigForModel, t as getProviderConfig, u as getProviderModelConfig, v as getModelsDevModelPricing, w as getReasoningEffortForModel, x as getExtraPromptForModel } from "./config-BzhriV_y.js";
+import { $ as forwardCodexResponses, B as getXaiAccessToken, C as copilotHeaders, D as prepareForCompact, E as isOpencodeOauthApp, F as compactTextOnlyGuard, I as createAuthMiddleware, J as XAI_API_BASE_URL, K as readGitHubToken, L as getConfiguredAdminApiKeys, M as compactMessageSections, O as prepareInteractionHeaders, P as compactSystemPromptStarts, Q as buildCodexRequestHeaders, S as copilotBaseUrl, T as copilotWebSocketHeaders, Z as CODEX_API_BASE_URL, _ as isAsyncIterable, at as createResponsesHttpEventStream, b as parseUserIdMetadata, ct as getResponsesStreamTerminalDisposition, d as generateRequestIdFromPayload, et as generateTraceId, f as getPromptCacheKey, ft as stripInternalRequestHeaders, g as getUUID, h as getSessionAffinityFromHeaders, it as fetchUpstreamWithLifecycle, j as compactAutoContinuePromptStarts, k as prepareMessageProxyHeaders, lt as createPooledWebSocketStream, m as getRootSessionIdFromHeaders, n as invalidateCodexRuntime, nt as resolveTraceId$1, ot as createResponsesSafeStream, p as getRootSessionId, q as readGitHubTokenFromEnv, rt as setPluginRequestActualModel, s as setupCodexToken, st as encodePoolKeyPart, tt as requestContext, u as stopCopilotRefreshLoop, ut as createWebSocketUrl, v as isNullish, x as getCopilotUsage, y as isResponsesStream } from "./token-XNKmjmqH.js";
 import { a as isDeferredToolName, c as parseMcpToolSearchSentinel, d as shouldEnableResponsesToolSearch, i as isBridgeToolSearchName, l as resolveBridgeToolSearchName, o as listDeferredToolNames, r as formatToolSearchBridgeArguments, s as normalizeToolSearchBridgeArguments, t as BRIDGE_TOOL_SEARCH_NAME, u as selectDeferredToolsByNames } from "./tool-search-CcvaqM03.js";
 import { n as isProviderAgentModelVisible, t as getProviderAgentModels } from "./provider-management-B9lCOTka.js";
-import { n as stopModelsRefreshLoop, t as setupCopilotRuntime } from "./copilot-runtime-RW0tbgv3.js";
+import { i as registerProcessCleanup, n as stopModelsRefreshLoop, r as EventBus, t as setupCopilotRuntime } from "./copilot-runtime-8IpywGxw.js";
 import consola from "consola";
 import { createHash, randomUUID } from "node:crypto";
 import fs, { readFileSync } from "node:fs";
@@ -35,605 +35,184 @@ const traceIdMiddleware = async (c, next) => {
 	});
 };
 //#endregion
-//#region src/lib/process-cleanup.ts
-const cleanupHandlers = /* @__PURE__ */ new Set();
-let cleanupPromise = null;
-let cleanupState = "idle";
-let runtimeInitialized$1 = false;
-function initializeProcessCleanupRuntime() {
-	if (runtimeInitialized$1) return;
-	runtimeInitialized$1 = true;
-	process.once("beforeExit", () => {
-		runProcessCleanups();
-	});
-	process.once("exit", runProcessCleanupsSync);
-	process.once("SIGINT", () => {
-		shutdownProcess(0);
-	});
-	process.once("SIGTERM", () => {
-		shutdownProcess(0);
-	});
-}
-function runProcessCleanupsSync() {
-	if (cleanupState !== "idle") return;
-	cleanupState = "done";
-	for (const handler of Array.from(cleanupHandlers)) try {
-		handler();
-	} catch {}
-}
-async function runProcessCleanups() {
-	if (cleanupPromise) return cleanupPromise;
-	if (cleanupState === "done") return;
-	cleanupState = "running";
-	cleanupPromise = (async () => {
-		for (const handler of Array.from(cleanupHandlers)) await handler();
-		cleanupState = "done";
-	})();
-	return cleanupPromise;
+//#region src/lib/sqlite.ts
+const MINIMUM_NODE_SQLITE_VERSION = "22.13.0";
+const isBunRuntime = () => Boolean(globalThis.Bun);
+function parseNodeVersion(version) {
+	return version.split(".", 3).map((part) => {
+		const parsed = Number.parseInt(part, 10);
+		return Number.isFinite(parsed) && parsed >= 0 ? parsed : 0;
+	});
 }
-async function shutdownProcess(exitCode) {
-	try {
-		await runProcessCleanups();
-	} finally {
-		process.exit(exitCode);
+function isNodeSqliteSupportedVersion(version) {
+	const current = parseNodeVersion(version);
+	const minimum = parseNodeVersion(MINIMUM_NODE_SQLITE_VERSION);
+	for (const [index, minimumPart] of minimum.entries()) {
+		const currentPart = current[index] ?? 0;
+		if (currentPart > minimumPart) return true;
+		if (currentPart < minimumPart) return false;
 	}
+	return true;
 }
-function registerProcessCleanup(handler) {
-	initializeProcessCleanupRuntime();
-	cleanupHandlers.add(handler);
-	return () => {
-		cleanupHandlers.delete(handler);
-	};
+function isSqliteRuntimeSupported(input = {}) {
+	if (input.isBun ?? isBunRuntime()) return true;
+	return isNodeSqliteSupportedVersion(input.nodeVersion ?? process.versions.node);
 }
-//#endregion
-//#region src/lib/logger.ts
-const LOG_RETENTION_MS = 6048e5;
-const CLEANUP_INTERVAL_MS = 864e5;
-const LOG_DIR_ENV = "COPILOT_API_LOG_DIR";
-const DEFAULT_LOG_DIR = path.join(PATHS.APP_DIR, "logs");
-const FLUSH_INTERVAL_MS = 1e3;
-const MAX_BUFFER_SIZE = 100;
-const getLogDir = () => process.env[LOG_DIR_ENV]?.trim() || DEFAULT_LOG_DIR;
-const logStreams = /* @__PURE__ */ new Map();
-const logBuffers = /* @__PURE__ */ new Map();
-let runtimeInitialized = false;
-let flushInterval;
-let cleanupInterval;
-let currentLogDateKey;
-const ensureLogDirectory = () => {
-	const logDir = getLogDir();
-	if (!fs.existsSync(logDir)) fs.mkdirSync(logDir, { recursive: true });
-};
-const formatArgs = (args) => args.map((arg) => typeof arg === "string" ? arg : util.inspect(arg, {
-	depth: null,
-	colors: false
-})).join(" ");
-const sanitizeName = (name) => {
-	const normalized = name.toLowerCase().replaceAll(/[^a-z0-9]+/g, "-").replaceAll(/^-+|-+$/g, "");
-	return normalized === "" ? "handler" : normalized;
-};
-const maybeUnref = (timer) => {
-	timer.unref();
-};
-const flushBuffer = (filePath) => {
-	const buffer = logBuffers.get(filePath);
-	if (!buffer) return;
-	logBuffers.delete(filePath);
-	if (buffer.length === 0) return;
-	const stream = getLogStream(filePath);
-	const content = buffer.join("\n") + "\n";
-	stream.write(content, (error) => {
-		if (error) console.warn("Failed to write handler log", error);
-	});
-};
-const flushAllBuffers = () => {
-	for (const filePath of logBuffers.keys()) flushBuffer(filePath);
-};
-const closeLogFile = async (filePath) => {
-	flushBuffer(filePath);
-	const stream = logStreams.get(filePath);
-	logStreams.delete(filePath);
-	logBuffers.delete(filePath);
-	if (!stream || stream.closed) return;
-	await new Promise((resolve) => {
-		stream.once("close", resolve);
-		stream.end();
-	});
-};
-const closePreviousDateLogFiles = (dateKey) => {
-	if (currentLogDateKey === void 0) {
-		currentLogDateKey = dateKey;
-		return;
+function getUnsupportedNodeSqliteMessage(nodeVersion) {
+	return `SQLite-backed token usage requires Bun or Node.js >= ${MINIMUM_NODE_SQLITE_VERSION}. Detected Node.js ${nodeVersion}. Upgrade Node.js or run the CLI with Bun, for example \`bunx --bun @jeffreycao/copilot-api@latest start\` or \`bun run start start\`.`;
+}
+var UnsupportedNodeSqliteRuntimeError = class extends Error {
+	constructor(nodeVersion, cause) {
+		super(getUnsupportedNodeSqliteMessage(nodeVersion), { cause });
+		this.name = "UnsupportedNodeSqliteRuntimeError";
 	}
-	if (currentLogDateKey === dateKey) return;
-	const filePaths = /* @__PURE__ */ new Set([...logBuffers.keys(), ...logStreams.keys()]);
-	currentLogDateKey = dateKey;
-	for (const filePath of filePaths) closeLogFile(filePath).catch((error) => {
-		console.warn("Failed to rotate handler log", error);
-	});
 };
-const cleanupOldLogs = async () => {
-	const logDir = getLogDir();
-	if (!fs.existsSync(logDir)) return;
-	const now = Date.now();
-	for (const entry of fs.readdirSync(logDir)) {
-		const filePath = path.join(logDir, entry);
-		let stats;
-		try {
-			stats = fs.statSync(filePath);
-		} catch {
-			continue;
-		}
-		if (!stats.isFile() || now - stats.mtimeMs <= LOG_RETENTION_MS) continue;
-		try {
-			await closeLogFile(filePath);
-		} catch (error) {
-			console.warn(`Failed to close old handler log: ${filePath}`, error);
-			continue;
-		}
-		try {
-			fs.rmSync(filePath);
-		} catch (error) {
-			console.warn(`Failed to remove old handler log: ${filePath}`, error);
-		}
+async function openBunDatabase(dbPath) {
+	return new (await (import(["bun", "sqlite"].join(":")))).Database(dbPath);
+}
+async function loadNodeSqliteModule() {
+	const nodeVersion = process.versions.node;
+	if (!isNodeSqliteSupportedVersion(nodeVersion)) throw new UnsupportedNodeSqliteRuntimeError(nodeVersion);
+	const emitWarning = process.emitWarning.bind(process);
+	process.emitWarning = (warning, ...args) => {
+		if (typeof warning === "string" && warning.includes("SQLite is an experimental feature")) return;
+		return emitWarning(warning, ...args);
+	};
+	const specifier = ["node", "sqlite"].join(":");
+	try {
+		return await import(specifier);
+	} catch (error) {
+		throw new UnsupportedNodeSqliteRuntimeError(nodeVersion, error);
+	} finally {
+		process.emitWarning = emitWarning;
 	}
-};
-const scheduleLogCleanup = () => {
-	cleanupOldLogs().catch((error) => {
-		console.warn("Failed to clean up handler logs", error);
-	});
-};
-const flushAndCloseLogStreams = () => {
-	flushAllBuffers();
-	for (const stream of logStreams.values()) stream.end();
-	logStreams.clear();
-	logBuffers.clear();
-	currentLogDateKey = void 0;
-};
-const shutdownLoggerRuntime = () => {
-	if (flushInterval) {
-		clearInterval(flushInterval);
-		flushInterval = void 0;
+}
+async function openNodeDatabase(dbPath) {
+	return new (await (loadNodeSqliteModule())).DatabaseSync(dbPath);
+}
+async function openSqliteDatabase(dbPath) {
+	const dir = path.dirname(dbPath);
+	if (dbPath !== ":memory:" && dir !== ".") await fs$1.mkdir(dir, { recursive: true });
+	return isBunRuntime() ? openBunDatabase(dbPath) : openNodeDatabase(dbPath);
+}
+var SqliteDbStore = class {
+	dbPromise = null;
+	options;
+	constructor(options) {
+		this.options = options;
 	}
-	if (cleanupInterval) {
-		clearInterval(cleanupInterval);
-		cleanupInterval = void 0;
+	getDb() {
+		this.dbPromise ??= this.open();
+		return this.dbPromise;
 	}
-	flushAndCloseLogStreams();
-	runtimeInitialized = false;
-};
-const initializeLoggerRuntime = () => {
-	if (runtimeInitialized) return;
-	runtimeInitialized = true;
-	ensureLogDirectory();
-	scheduleLogCleanup();
-	flushInterval = setInterval(flushAllBuffers, FLUSH_INTERVAL_MS);
-	maybeUnref(flushInterval);
-	cleanupInterval = setInterval(scheduleLogCleanup, CLEANUP_INTERVAL_MS);
-	maybeUnref(cleanupInterval);
-	registerProcessCleanup(shutdownLoggerRuntime);
-};
-const getLogStream = (filePath) => {
-	initializeLoggerRuntime();
-	let stream = logStreams.get(filePath);
-	if (!stream || stream.destroyed) {
-		const createdStream = fs.createWriteStream(filePath, { flags: "a" });
-		stream = createdStream;
-		logStreams.set(filePath, createdStream);
-		createdStream.on("error", (error) => {
-			console.warn("Log stream error", error);
-			if (logStreams.get(filePath) === createdStream) logStreams.delete(filePath);
-		});
+	async close(input) {
+		const currentDbPromise = this.dbPromise;
+		this.dbPromise = null;
+		if (!currentDbPromise) return;
+		const db = await currentDbPromise;
+		input?.beforeClose?.(db);
+		db.close?.();
 	}
-	return stream;
-};
-const appendLine = (filePath, line) => {
-	let buffer = logBuffers.get(filePath);
-	if (!buffer) {
-		buffer = [];
-		logBuffers.set(filePath, buffer);
+	async open() {
+		const db = await openSqliteDatabase(this.options.getPath());
+		this.options.initialize?.(db);
+		return db;
 	}
-	buffer.push(line);
-	if (buffer.length >= MAX_BUFFER_SIZE) flushBuffer(filePath);
-};
-const debugLazy = (logger, factory) => {
-	if (!state.verbose) return;
-	logger.debug(...factory());
-};
-const debugJson = (logger, label, value) => {
-	debugLazy(logger, () => [label, JSON.stringify(value)]);
-};
-const debugJsonAsync = async (logger, label, factory) => {
-	if (!state.verbose) return;
-	logger.debug(label, JSON.stringify(await factory()));
-};
-const debugJsonTail = (logger, label, { value, tailLength = 400 }) => {
-	debugLazy(logger, () => [label, JSON.stringify(value).slice(-tailLength)]);
-};
-const createHandlerLogger = (name) => {
-	const sanitizedName = sanitizeName(name);
-	const instance = consola.withTag(name);
-	if (state.verbose) instance.level = 5;
-	instance.setReporters([]);
-	instance.addReporter({ log(logObj) {
-		initializeLoggerRuntime();
-		const traceId = requestContext.getStore()?.traceId;
-		const date = logObj.date;
-		const dateKey = date.toLocaleDateString("sv-SE");
-		const timestamp = date.toLocaleString("sv-SE", { hour12: false });
-		const filePath = path.join(getLogDir(), `${sanitizedName}-${dateKey}.log`);
-		const message = formatArgs(logObj.args);
-		const traceIdStr = traceId ? ` [${traceId}]` : "";
-		const line = `[${timestamp}] [${logObj.type}] [${logObj.tag || name}]${traceIdStr}${message ? ` ${message}` : ""}`;
-		closePreviousDateLogFiles(dateKey);
-		appendLine(filePath, line);
-	} });
-	return instance;
 };
 //#endregion
-//#region src/lib/models.ts
-/**
-* Converts a Copilot upstream model ID to a client-friendly ID that Claude Code
-* and Claude Desktop recognize (dots in version replaced with hyphens).
-* e.g. "claude-sonnet-4.6" -> "claude-sonnet-4-6"
-* Non-Claude models are returned unchanged.
-*/
-const toClientModelId = (modelId) => {
-	const normalized = normalizeSdkModelId(modelId);
-	if (!normalized) return modelId;
-	const versionHyphenated = normalized.version.replaceAll(".", "-");
-	return `claude-${normalized.family}-${versionHyphenated}`;
-};
-const findEndpointModel = (sdkModelId) => {
-	const models = state.models?.data ?? [];
-	const exactMatch = models.find((m) => m.id === sdkModelId);
-	if (exactMatch) return exactMatch;
-	const normalized = normalizeSdkModelId(sdkModelId);
-	if (!normalized) return;
-	const modelName = `claude-${normalized.family}-${normalized.version}`;
-	const model = models.find((m) => m.id === modelName);
-	if (model) return model;
-};
-/**
-* Normalizes an SDK model ID to extract the model family and version.
-* this method from github copilot extension
-* Examples:
-* - "claude-opus-4-5-20251101" -> { family: "opus", version: "4.5" }
-* - "claude-3-5-sonnet-20241022" -> { family: "sonnet", version: "3.5" }
-* - "claude-sonnet-4-20250514" -> { family: "sonnet", version: "4" }
-* - "claude-haiku-3-5-20250514" -> { family: "haiku", version: "3.5" }
-* - "claude-haiku-4.5" -> { family: "haiku", version: "4.5" }
-*/
-const normalizeSdkModelId = (sdkModelId) => {
-	const withoutDate = sdkModelId.toLowerCase().replace(/-\d{8}$/, "");
-	const pattern1 = withoutDate.match(/^claude-(\w+)-(\d+)\.(\d+)$/);
-	if (pattern1) return {
-		family: pattern1[1],
-		version: `${pattern1[2]}.${pattern1[3]}`
-	};
-	const pattern2 = withoutDate.match(/^claude-(\w+)-(\d+)-(\d+)$/);
-	if (pattern2) return {
-		family: pattern2[1],
-		version: `${pattern2[2]}.${pattern2[3]}`
-	};
-	const pattern3 = withoutDate.match(/^claude-(\d+)-(\d+)-(\w+)$/);
-	if (pattern3) return {
-		family: pattern3[3],
-		version: `${pattern3[1]}.${pattern3[2]}`
-	};
-	const pattern4 = withoutDate.match(/^claude-(\w+)-(\d+)$/);
-	if (pattern4) return {
-		family: pattern4[1],
-		version: pattern4[2]
-	};
-	const pattern5 = withoutDate.match(/^claude-(\d+)-(\w+)$/);
-	if (pattern5) return {
-		family: pattern5[2],
-		version: pattern5[1]
-	};
-};
-//#endregion
-//#region src/lib/provider-model.ts
-const parseProviderModelAlias = (model) => {
-	const separatorIndex = model.indexOf("/");
-	if (separatorIndex <= 0 || separatorIndex === model.length - 1) return null;
-	const provider = model.slice(0, separatorIndex).trim();
-	const providerModel = model.slice(separatorIndex + 1).trim();
-	if (!provider || !providerModel) return null;
-	return {
-		model: providerModel,
-		provider
-	};
-};
-const createFallbackModel = (modelId) => ({
-	capabilities: {
-		family: "provider",
-		limits: {},
-		object: "model_capabilities",
-		supports: {},
-		tokenizer: "o200k_base",
-		type: "chat"
-	},
-	id: modelId,
-	model_picker_enabled: false,
-	name: modelId,
-	object: "model",
-	preview: false,
-	vendor: "provider",
-	version: "unknown"
+//#region src/lib/token-usage/store.ts
+const DB_PATH_ENV = "COPILOT_API_SQLITE_DB_PATH";
+const DEFAULT_DB_FILENAME = "copilot-api.sqlite";
+const COST_NANOS_PER_UNIT$1 = 1e9;
+let writeQueue = Promise.resolve();
+function getDbPath() {
+	return process.env[DB_PATH_ENV] ?? path.join(PATHS.APP_DIR, DEFAULT_DB_FILENAME);
+}
+const tokenUsageDbStore = new SqliteDbStore({
+	getPath: getDbPath,
+	initialize: initializeTokenUsageDb
 });
-//#endregion
-//#region src/lib/provider-resolver.ts
-const AZURE_OPENAI_SCOPE = "https://cognitiveservices.azure.com/.default";
-let cachedAzureAccessTokenProvider = null;
-/**
-* Lazily creates the Entra token provider on first use and reuses it
-* afterwards. Importing @azure/identity eagerly adds ~0.5s to every process
-* startup, and constructing a new DefaultAzureCredential per call would drop
-* the token cache, so both the import and the credential are singletons.
-*/
-async function getAzureAccessTokenProvider() {
-	cachedAzureAccessTokenProvider ??= (async () => {
-		try {
-			const { DefaultAzureCredential, getBearerTokenProvider } = await import("@azure/identity");
-			return getBearerTokenProvider(new DefaultAzureCredential(), AZURE_OPENAI_SCOPE);
-		} catch (error) {
-			cachedAzureAccessTokenProvider = null;
-			throw error;
-		}
-	})();
-	return cachedAzureAccessTokenProvider;
+function getDb() {
+	return tokenUsageDbStore.getDb();
 }
-async function getDefaultAzureAccessToken() {
-	return (await getAzureAccessTokenProvider())();
+function isTokenUsageStorageEnabled() {
+	return isSqliteRuntimeSupported();
 }
-function isMissingCodexCredentialsError(error) {
-	return error instanceof Error && error.message === "Codex credentials not found. Run `copilot-api auth login --provider codex` first.";
+function initializeTokenUsageDb(db) {
+	db.exec("PRAGMA journal_mode = WAL");
+	db.exec("PRAGMA busy_timeout = 5000");
+	db.exec(`
+    CREATE TABLE IF NOT EXISTS token_usage_events (
+      id INTEGER PRIMARY KEY AUTOINCREMENT,
+      created_at_ms INTEGER NOT NULL,
+      created_at_utc TEXT NOT NULL,
+      trace_id TEXT NOT NULL,
+      session_id TEXT NOT NULL DEFAULT '',
+      user_id TEXT NOT NULL DEFAULT '',
+      user_agent TEXT NOT NULL DEFAULT '',
+      source TEXT NOT NULL,
+      endpoint TEXT NOT NULL,
+      provider_name TEXT,
+      model TEXT NOT NULL,
+      input_tokens INTEGER NOT NULL DEFAULT 0,
+      output_tokens INTEGER NOT NULL DEFAULT 0,
+      cache_read_input_tokens INTEGER NOT NULL DEFAULT 0,
+      cache_creation_input_tokens INTEGER NOT NULL DEFAULT 0,
+      total_tokens INTEGER NOT NULL DEFAULT 0,
+      total_nano_aiu INTEGER,
+      cost_currency TEXT,
+      total_cost_nanos INTEGER,
+      cost_source TEXT
+    )
+  `);
+	ensureColumn(db, "user_id", "TEXT NOT NULL DEFAULT ''");
+	ensureColumn(db, "user_agent", "TEXT NOT NULL DEFAULT ''");
+	ensureColumn(db, "total_tokens", "INTEGER NOT NULL DEFAULT 0");
+	ensureColumn(db, "total_nano_aiu", "INTEGER");
+	ensureColumn(db, "cost_currency", "TEXT");
+	ensureColumn(db, "total_cost_nanos", "INTEGER");
+	ensureColumn(db, "cost_source", "TEXT");
+	ensureColumn(db, "request_id", "TEXT");
+	ensureColumn(db, "first_token_ms", "REAL");
+	ensureColumn(db, "duration_ms", "REAL");
+	ensureColumn(db, "request_outcome", "TEXT");
+	db.exec(`
+    CREATE INDEX IF NOT EXISTS idx_token_usage_events_request_id
+    ON token_usage_events(request_id)
+  `);
+	db.exec(`
+    CREATE INDEX IF NOT EXISTS idx_token_usage_events_created_at_ms
+    ON token_usage_events(created_at_ms)
+  `);
+	db.exec(`
+    CREATE INDEX IF NOT EXISTS idx_token_usage_events_model
+    ON token_usage_events(model)
+  `);
+	db.exec(`
+    CREATE INDEX IF NOT EXISTS idx_token_usage_events_trace_id
+    ON token_usage_events(trace_id)
+  `);
+	db.exec(`
+    CREATE INDEX IF NOT EXISTS idx_token_usage_events_session_id
+    ON token_usage_events(session_id)
+  `);
+	db.exec(`
+    CREATE INDEX IF NOT EXISTS idx_token_usage_events_user_id
+    ON token_usage_events(user_id)
+  `);
 }
-async function resolveProviderConfig(providerName, getAzureAccessToken = getDefaultAzureAccessToken) {
-	const normalizedProviderName = providerName.trim();
-	if (!normalizedProviderName) return null;
-	if (normalizedProviderName === "codex") {
-		if (getRawProviderConfig(normalizedProviderName)?.enabled === false) return null;
-		try {
-			await setupCodexToken();
-		} catch (error) {
-			if (isMissingCodexCredentialsError(error)) return null;
-			throw error;
-		}
-		const providerConfig = getProviderConfig(normalizedProviderName);
-		if (!providerConfig) return null;
-		return {
-			...providerConfig,
-			apiKey: state.codexAccessToken ?? providerConfig.apiKey
-		};
-	}
-	const providerConfig = getProviderConfig(normalizedProviderName);
-	if (providerConfig?.name === "xai" && providerConfig.authType === "oauth2") {
-		const accessToken = await getXaiAccessToken();
-		return accessToken ? {
-			...providerConfig,
-			baseUrl: providerConfig.baseUrl === "https://api.x.ai" ? XAI_API_BASE_URL : providerConfig.baseUrl,
-			apiKey: accessToken
-		} : null;
-	}
-	if (providerConfig?.authType !== "azure-entra") return providerConfig;
-	try {
-		return {
-			...providerConfig,
-			apiKey: await getAzureAccessToken()
-		};
-	} catch (error) {
-		throw new Error(`Failed to acquire Azure Entra token for provider '${normalizedProviderName}': ${error instanceof Error ? error.message : String(error)}`, { cause: error });
-	}
+function ensureColumn(db, name, definition) {
+	if (!db.prepare("PRAGMA table_info(token_usage_events)").all().some((row) => row.name === name)) db.exec(`ALTER TABLE token_usage_events ADD COLUMN ${name} ${definition}`);
 }
-/**
-* Returns the parsed "provider/model" alias only when that provider is
-* actually configured, otherwise null so the caller falls through to its
-* default flow. GitHub Copilot enterprise models can ship with namespaced ids
-* such as "org/family/model": without this check the first segment is
-* misrouted to a non-existent provider and surfaced as a 400/404.
-*/
-async function ensureConfiguredProviderModelAlias(alias, resolveConfig = resolveProviderConfig) {
-	if (!alias) return null;
-	return await resolveConfig(alias.provider) ? alias : null;
+function normalizeToken(value) {
+	if (typeof value !== "number" || !Number.isFinite(value)) return 0;
+	return Math.max(0, Math.floor(value));
 }
-/** parseProviderModelAlias + ensureConfiguredProviderModelAlias. */
-async function resolveConfiguredProviderModelAlias(model, resolveConfig = resolveProviderConfig) {
-	return ensureConfiguredProviderModelAlias(parseProviderModelAlias(model), resolveConfig);
+function normalizeOptionalToken(value) {
+	return value === null || value === void 0 ? void 0 : normalizeToken(value);
 }
-//#endregion
-//#region src/lib/event-bus.ts
-var EventBus = class {
-	handlers = /* @__PURE__ */ new Map();
-	publish(name, event) {
-		const handlers = this.handlers.get(name);
-		if (!handlers) return;
-		for (const handler of Array.from(handlers)) handler(event);
-	}
-	subscribe(name, handler) {
-		let handlers = this.handlers.get(name);
-		if (!handlers) {
-			handlers = /* @__PURE__ */ new Set();
-			this.handlers.set(name, handlers);
-		}
-		const registeredHandler = handler;
-		handlers.add(registeredHandler);
-		return () => {
-			handlers.delete(registeredHandler);
-			if (handlers.size === 0) this.handlers.delete(name);
-		};
-	}
-};
-//#endregion
-//#region src/lib/sqlite.ts
-const MINIMUM_NODE_SQLITE_VERSION = "22.13.0";
-const isBunRuntime = () => Boolean(globalThis.Bun);
-function parseNodeVersion(version) {
-	return version.split(".", 3).map((part) => {
-		const parsed = Number.parseInt(part, 10);
-		return Number.isFinite(parsed) && parsed >= 0 ? parsed : 0;
-	});
-}
-function isNodeSqliteSupportedVersion(version) {
-	const current = parseNodeVersion(version);
-	const minimum = parseNodeVersion(MINIMUM_NODE_SQLITE_VERSION);
-	for (const [index, minimumPart] of minimum.entries()) {
-		const currentPart = current[index] ?? 0;
-		if (currentPart > minimumPart) return true;
-		if (currentPart < minimumPart) return false;
-	}
-	return true;
-}
-function isSqliteRuntimeSupported(input = {}) {
-	if (input.isBun ?? isBunRuntime()) return true;
-	return isNodeSqliteSupportedVersion(input.nodeVersion ?? process.versions.node);
-}
-function getUnsupportedNodeSqliteMessage(nodeVersion) {
-	return `SQLite-backed token usage requires Bun or Node.js >= ${MINIMUM_NODE_SQLITE_VERSION}. Detected Node.js ${nodeVersion}. Upgrade Node.js or run the CLI with Bun, for example \`bunx --bun @jeffreycao/copilot-api@latest start\` or \`bun run start start\`.`;
-}
-var UnsupportedNodeSqliteRuntimeError = class extends Error {
-	constructor(nodeVersion, cause) {
-		super(getUnsupportedNodeSqliteMessage(nodeVersion), { cause });
-		this.name = "UnsupportedNodeSqliteRuntimeError";
-	}
-};
-async function openBunDatabase(dbPath) {
-	return new (await (import(["bun", "sqlite"].join(":")))).Database(dbPath);
-}
-async function loadNodeSqliteModule() {
-	const nodeVersion = process.versions.node;
-	if (!isNodeSqliteSupportedVersion(nodeVersion)) throw new UnsupportedNodeSqliteRuntimeError(nodeVersion);
-	const emitWarning = process.emitWarning.bind(process);
-	process.emitWarning = (warning, ...args) => {
-		if (typeof warning === "string" && warning.includes("SQLite is an experimental feature")) return;
-		return emitWarning(warning, ...args);
-	};
-	const specifier = ["node", "sqlite"].join(":");
-	try {
-		return await import(specifier);
-	} catch (error) {
-		throw new UnsupportedNodeSqliteRuntimeError(nodeVersion, error);
-	} finally {
-		process.emitWarning = emitWarning;
-	}
-}
-async function openNodeDatabase(dbPath) {
-	return new (await (loadNodeSqliteModule())).DatabaseSync(dbPath);
-}
-async function openSqliteDatabase(dbPath) {
-	const dir = path.dirname(dbPath);
-	if (dbPath !== ":memory:" && dir !== ".") await fs$1.mkdir(dir, { recursive: true });
-	return isBunRuntime() ? openBunDatabase(dbPath) : openNodeDatabase(dbPath);
-}
-var SqliteDbStore = class {
-	dbPromise = null;
-	options;
-	constructor(options) {
-		this.options = options;
-	}
-	getDb() {
-		this.dbPromise ??= this.open();
-		return this.dbPromise;
-	}
-	async close(input) {
-		const currentDbPromise = this.dbPromise;
-		this.dbPromise = null;
-		if (!currentDbPromise) return;
-		const db = await currentDbPromise;
-		input?.beforeClose?.(db);
-		db.close?.();
-	}
-	async open() {
-		const db = await openSqliteDatabase(this.options.getPath());
-		this.options.initialize?.(db);
-		return db;
-	}
-};
-//#endregion
-//#region src/lib/token-usage/store.ts
-const DB_PATH_ENV = "COPILOT_API_SQLITE_DB_PATH";
-const DEFAULT_DB_FILENAME = "copilot-api.sqlite";
-const COST_NANOS_PER_UNIT$1 = 1e9;
-let writeQueue = Promise.resolve();
-function getDbPath() {
-	return process.env[DB_PATH_ENV] ?? path.join(PATHS.APP_DIR, DEFAULT_DB_FILENAME);
-}
-const tokenUsageDbStore = new SqliteDbStore({
-	getPath: getDbPath,
-	initialize: initializeTokenUsageDb
-});
-function getDb() {
-	return tokenUsageDbStore.getDb();
-}
-function isTokenUsageStorageEnabled() {
-	return isSqliteRuntimeSupported();
-}
-function initializeTokenUsageDb(db) {
-	db.exec("PRAGMA journal_mode = WAL");
-	db.exec("PRAGMA busy_timeout = 5000");
-	db.exec(`
-    CREATE TABLE IF NOT EXISTS token_usage_events (
-      id INTEGER PRIMARY KEY AUTOINCREMENT,
-      created_at_ms INTEGER NOT NULL,
-      created_at_utc TEXT NOT NULL,
-      trace_id TEXT NOT NULL,
-      session_id TEXT NOT NULL DEFAULT '',
-      user_id TEXT NOT NULL DEFAULT '',
-      source TEXT NOT NULL,
-      endpoint TEXT NOT NULL,
-      provider_name TEXT,
-      model TEXT NOT NULL,
-      input_tokens INTEGER NOT NULL DEFAULT 0,
-      output_tokens INTEGER NOT NULL DEFAULT 0,
-      cache_read_input_tokens INTEGER NOT NULL DEFAULT 0,
-      cache_creation_input_tokens INTEGER NOT NULL DEFAULT 0,
-      total_tokens INTEGER NOT NULL DEFAULT 0,
-      total_nano_aiu INTEGER,
-      cost_currency TEXT,
-      total_cost_nanos INTEGER,
-      cost_source TEXT
-    )
-  `);
-	ensureColumn(db, "user_id", "TEXT NOT NULL DEFAULT ''");
-	ensureColumn(db, "total_tokens", "INTEGER NOT NULL DEFAULT 0");
-	ensureColumn(db, "total_nano_aiu", "INTEGER");
-	ensureColumn(db, "cost_currency", "TEXT");
-	ensureColumn(db, "total_cost_nanos", "INTEGER");
-	ensureColumn(db, "cost_source", "TEXT");
-	db.exec(`
-    CREATE INDEX IF NOT EXISTS idx_token_usage_events_created_at_ms
-    ON token_usage_events(created_at_ms)
-  `);
-	db.exec(`
-    CREATE INDEX IF NOT EXISTS idx_token_usage_events_model
-    ON token_usage_events(model)
-  `);
-	db.exec(`
-    CREATE INDEX IF NOT EXISTS idx_token_usage_events_trace_id
-    ON token_usage_events(trace_id)
-  `);
-	db.exec(`
-    CREATE INDEX IF NOT EXISTS idx_token_usage_events_session_id
-    ON token_usage_events(session_id)
-  `);
-	db.exec(`
-    CREATE INDEX IF NOT EXISTS idx_token_usage_events_user_id
-    ON token_usage_events(user_id)
-  `);
-}
-function ensureColumn(db, name, definition) {
-	if (!db.prepare("PRAGMA table_info(token_usage_events)").all().some((row) => row.name === name)) db.exec(`ALTER TABLE token_usage_events ADD COLUMN ${name} ${definition}`);
-}
-function normalizeToken(value) {
-	if (typeof value !== "number" || !Number.isFinite(value)) return 0;
-	return Math.max(0, Math.floor(value));
-}
-function normalizeOptionalToken(value) {
-	return value === null || value === void 0 ? void 0 : normalizeToken(value);
-}
-function hasAnyToken(tokens) {
-	return normalizeToken(tokens.input_tokens) > 0 || normalizeToken(tokens.output_tokens) > 0 || normalizeToken(tokens.cache_read_input_tokens) > 0 || normalizeToken(tokens.cache_creation_input_tokens) > 0 || normalizeToken(tokens.total_tokens) > 0 || normalizeToken(tokens.total_nano_aiu) > 0;
+function hasAnyToken(tokens) {
+	return normalizeToken(tokens.input_tokens) > 0 || normalizeToken(tokens.output_tokens) > 0 || normalizeToken(tokens.cache_read_input_tokens) > 0 || normalizeToken(tokens.cache_creation_input_tokens) > 0 || normalizeToken(tokens.total_tokens) > 0 || normalizeToken(tokens.total_nano_aiu) > 0;
 }
 function resolveTotalTokens(input) {
 	const explicitTotal = normalizeOptionalToken(input.total_tokens);
@@ -660,9 +239,14 @@ async function writeTokenUsageEvent(event) {
         total_nano_aiu,
         cost_currency,
         total_cost_nanos,
-        cost_source
-      ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
-    `).run(event.created_at_ms, event.created_at_utc, event.trace_id, event.session_id, event.user_id, event.source, event.endpoint, event.provider_name, event.model, event.input_tokens, event.output_tokens, event.cache_read_input_tokens, event.cache_creation_input_tokens, event.total_tokens, event.total_nano_aiu, event.cost_currency, event.total_cost_nanos, event.cost_source);
+        cost_source,
+        request_id,
+        first_token_ms,
+        duration_ms,
+        request_outcome,
+        user_agent
+      ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+    `).run(event.created_at_ms, event.created_at_utc, event.trace_id, event.session_id, event.user_id, event.source, event.endpoint, event.provider_name, event.model, event.input_tokens, event.output_tokens, event.cache_read_input_tokens, event.cache_creation_input_tokens, event.total_tokens, event.total_nano_aiu, event.cost_currency, event.total_cost_nanos, event.cost_source, event.request_id, event.first_token_ms, event.duration_ms, event.request_outcome, event.user_agent);
 }
 function enqueueTokenUsageWrite(event) {
 	if (!isTokenUsageStorageEnabled()) return;
@@ -670,6 +254,19 @@ function enqueueTokenUsageWrite(event) {
 		consola.warn("Failed to record token usage", error);
 	});
 }
+function enqueueTokenUsageTiming(timing) {
+	if (!isTokenUsageStorageEnabled()) return;
+	const { id, firstTokenMs, durationMs, outcome } = timing;
+	writeQueue = writeQueue.then(async () => {
+		(await getDb()).prepare(`
+      UPDATE token_usage_events
+      SET first_token_ms = ?, duration_ms = ?, request_outcome = ?
+      WHERE request_id = ?
+    `).run(firstTokenMs, durationMs, outcome, id);
+	}).catch((error) => {
+		consola.warn("Failed to record token usage timing", error);
+	});
+}
 async function flushTokenUsageEvents() {
 	let currentQueue = writeQueue;
 	while (true) {
@@ -878,6 +475,8 @@ function totalsFromRow(row, costs = []) {
 function modelSummaryFromRow(row, costs = []) {
 	return {
 		...totalsFromRow(row, costs),
+		average_first_token_ms: nullableNumberFromRow(row, "average_first_token_ms"),
+		output_tokens_per_second: nullableNumberFromRow(row, "output_tokens_per_second"),
 		model: typeof row.model === "string" ? row.model : "unknown"
 	};
 }
@@ -889,8 +488,21 @@ function nullableStringFromRow(row, key) {
 	const value = row[key];
 	return typeof value === "string" ? value : null;
 }
+function outputTokensPerSecond(row) {
+	const first = nullableNumberFromRow(row, "first_token_ms");
+	const duration = nullableNumberFromRow(row, "duration_ms");
+	const output = numberFromRow(row, "output_tokens");
+	if (row.request_outcome !== "completed" || row.usage_event_count !== 1 || first === null || first < 0 || duration === null || duration <= first || output <= 0) return null;
+	const rate = output * 1e3 / (duration - first);
+	return Number.isFinite(rate) ? rate : null;
+}
 function usageEventFromRow(row) {
 	return {
+		request_id: nullableStringFromRow(row, "request_id"),
+		first_token_ms: nullableNumberFromRow(row, "first_token_ms"),
+		duration_ms: nullableNumberFromRow(row, "duration_ms"),
+		request_outcome: nullableStringFromRow(row, "request_outcome"),
+		output_tokens_per_second: outputTokensPerSecond(row),
 		cache_creation_input_tokens: numberFromRow(row, "cache_creation_input_tokens"),
 		cache_read_input_tokens: numberFromRow(row, "cache_read_input_tokens"),
 		cost: eventCostFromRow(row),
@@ -907,7 +519,8 @@ function usageEventFromRow(row) {
 		total_nano_aiu: nullableNumberFromRow(row, "total_nano_aiu"),
 		total_tokens: numberFromRow(row, "total_tokens"),
 		trace_id: stringFromRow(row, "trace_id"),
-		user_id: stringFromRow(row, "user_id")
+		user_id: stringFromRow(row, "user_id"),
+		user_agent: stringFromRow(row, "user_agent")
 	};
 }
 function getTotalsRow(db, range) {
@@ -1034,11 +647,56 @@ function getSessionSummaries(db, range, limit, offset) {
 		};
 	});
 }
+function getModelTimingRows(db, intervals) {
+	if (intervals.length === 0) return [];
+	return db.prepare(`
+    WITH selected_requests AS (
+      SELECT DISTINCT request_id FROM token_usage_events
+      WHERE created_at_ms >= ? AND created_at_ms < ? AND request_id IS NOT NULL
+    ), requests AS (
+      SELECT e.request_id, MIN(e.model) AS model,
+        MIN(e.created_at_ms) AS first_usage_at_ms,
+        MAX(e.created_at_ms) AS last_usage_at_ms,
+        SUM(e.output_tokens) AS output_tokens,
+        MAX(e.first_token_ms) AS first_token_ms,
+        MAX(e.duration_ms) AS duration_ms
+      FROM selected_requests r JOIN token_usage_events e USING (request_id)
+      GROUP BY e.request_id
+      HAVING COUNT(DISTINCT e.model) = 1
+        AND COUNT(e.first_token_ms) = COUNT(*)
+        AND COUNT(e.duration_ms) = COUNT(*)
+        AND COUNT(e.request_outcome) = COUNT(*)
+        AND MIN(e.request_outcome) = 'completed'
+        AND MAX(e.request_outcome) = 'completed'
+        AND MIN(e.first_token_ms) >= 0
+        AND MIN(e.duration_ms) >= MAX(e.first_token_ms)
+    ), buckets(usage_date, start_ms, end_ms) AS (VALUES ${intervals.map(() => "(?, ?, ?)").join(", ")})
+    SELECT buckets.usage_date, model,
+      AVG(first_token_ms) AS average_first_token_ms,
+      1000.0 * SUM(CASE WHEN duration_ms > first_token_ms AND output_tokens > 0 THEN output_tokens END)
+        / SUM(CASE WHEN duration_ms > first_token_ms AND output_tokens > 0 THEN duration_ms - first_token_ms END)
+        AS output_tokens_per_second
+    FROM requests JOIN buckets
+      ON first_usage_at_ms >= start_ms AND last_usage_at_ms < end_ms
+    GROUP BY buckets.usage_date, model
+  `).all(intervals[0].startMs, intervals[intervals.length - 1].endMs, ...intervals.flatMap(({ date, startMs, endMs }) => [
+		date,
+		startMs,
+		endMs
+	]));
+}
 function getModelSummaries(db, range) {
 	const costMap = getModelCostMap(db, range);
+	const timings = new Map(getModelTimingRows(db, [{
+		date: "",
+		...range
+	}]).map((row) => [stringFromRow(row, "model"), row]));
 	return getModelRows(db, range).map((row) => {
 		const model = stringFromRow(row, "model") || "unknown";
-		return modelSummaryFromRow(row, costMap.get(model) ?? []);
+		return modelSummaryFromRow({
+			...row,
+			...timings.get(model)
+		}, costMap.get(model) ?? []);
 	});
 }
 function getDailyModelSummaries(db, intervals) {
@@ -1093,13 +751,17 @@ function getDailyModelSummaries(db, intervals) {
 		const key = `${date}\u0000${model}`;
 		costsByDateAndModel.set(key, [...costsByDateAndModel.get(key) ?? [], cost]);
 	}
+	const timings = new Map(getModelTimingRows(db, intervals).map((row) => [`${stringFromRow(row, "usage_date")}\u0000${stringFromRow(row, "model")}`, row]));
 	const summariesByDate = /* @__PURE__ */ new Map();
 	for (const row of rows) {
 		const date = stringFromRow(row, "usage_date");
 		if (!date) continue;
 		const model = stringFromRow(row, "model") || "unknown";
 		const summaries = summariesByDate.get(date) ?? [];
-		summaries.push(modelSummaryFromRow(row, costsByDateAndModel.get(`${date}\u0000${model}`) ?? []));
+		summaries.push(modelSummaryFromRow({
+			...row,
+			...timings.get(`${date}\u0000${model}`)
+		}, costsByDateAndModel.get(`${date}\u0000${model}`) ?? []));
 		summariesByDate.set(date, summaries);
 	}
 	return summariesByDate;
@@ -1114,125 +776,622 @@ function createDailyBucket(interval, byModel) {
 		start_ms: interval.startMs,
 		totals
 	};
-}
-async function getTokenUsageSummary(period) {
-	if (!isTokenUsageStorageEnabled()) return createEmptySummary(period);
-	await flushTokenUsageEvents();
-	const db = await getDb();
-	const range = getPeriodRangeFromDb(db, period);
-	const totalsRow = getTotalsRow(db, range);
-	return {
-		byModel: getModelSummaries(db, range),
-		period,
-		range: rangePayload(range),
-		totals: totalsFromRow(totalsRow, getCostRows(db, range))
+}
+async function getTokenUsageSummary(period) {
+	if (!isTokenUsageStorageEnabled()) return createEmptySummary(period);
+	await flushTokenUsageEvents();
+	const db = await getDb();
+	const range = getPeriodRangeFromDb(db, period);
+	const totalsRow = getTotalsRow(db, range);
+	return {
+		byModel: getModelSummaries(db, range),
+		period,
+		range: rangePayload(range),
+		totals: totalsFromRow(totalsRow, getCostRows(db, range))
+	};
+}
+async function getTokenUsageDailySummary(period) {
+	if (!isTokenUsageStorageEnabled()) return createEmptyDailySummary(period);
+	await flushTokenUsageEvents();
+	const db = await getDb();
+	const range = getPeriodRangeFromDb(db, period);
+	const intervals = createDailyIntervals(range);
+	const dailySummaries = getDailyModelSummaries(db, intervals);
+	return {
+		byModel: getModelSummaries(db, range),
+		days: intervals.map((interval) => createDailyBucket(interval, dailySummaries.get(interval.date) ?? [])),
+		period,
+		range: rangePayload(range),
+		totals: totalsFromRow(getTotalsRow(db, range), getCostRows(db, range))
+	};
+}
+async function getTokenUsageEventsPage(input) {
+	if (!isTokenUsageStorageEnabled()) return createEmptyEventsPage(input);
+	await flushTokenUsageEvents();
+	const page = Math.max(1, Math.floor(input.page));
+	const pageSize = Math.min(100, Math.max(1, Math.floor(input.pageSize)));
+	const offset = (page - 1) * pageSize;
+	const db = await getDb();
+	const range = getPeriodRangeFromDb(db, input.period);
+	const totalRow = db.prepare(`
+    SELECT COUNT(*) AS total
+    FROM token_usage_events
+    WHERE created_at_ms >= ? AND created_at_ms < ?
+  `).get(range.startMs, range.endMs);
+	const rows = db.prepare(`
+    SELECT
+      id,
+      created_at_ms,
+      created_at_utc,
+      trace_id,
+      session_id,
+      user_id,
+      source,
+      endpoint,
+      provider_name,
+      model,
+      input_tokens,
+      output_tokens,
+      cache_read_input_tokens,
+      cache_creation_input_tokens,
+      total_nano_aiu,
+      total_tokens,
+      cost_currency,
+      total_cost_nanos,
+      cost_source,
+      request_id,
+      first_token_ms,
+      duration_ms,
+      request_outcome,
+      user_agent,
+      (SELECT COUNT(*) FROM token_usage_events sibling
+        WHERE sibling.request_id = token_usage_events.request_id) AS usage_event_count
+    FROM token_usage_events
+    WHERE created_at_ms >= ? AND created_at_ms < ?
+    ORDER BY created_at_ms DESC, id DESC
+    LIMIT ? OFFSET ?
+  `).all(range.startMs, range.endMs, pageSize, offset);
+	const total = numberFromRow(totalRow, "total");
+	return {
+		items: rows.map((row) => usageEventFromRow(row)),
+		page,
+		page_size: pageSize,
+		period: input.period,
+		range: {
+			end_ms: range.endMs,
+			end_utc: new Date(range.endMs).toISOString(),
+			start_ms: range.startMs,
+			start_utc: new Date(range.startMs).toISOString()
+		},
+		total,
+		total_pages: Math.max(1, Math.ceil(total / pageSize))
+	};
+}
+async function getTokenUsageSessionsPage(input) {
+	const page = Math.max(1, Math.floor(input.page));
+	const pageSize = Math.min(100, Math.max(1, Math.floor(input.pageSize)));
+	if (!isTokenUsageStorageEnabled()) return {
+		...createEmptyEventsPage(input),
+		items: []
+	};
+	await flushTokenUsageEvents();
+	const db = await getDb();
+	const range = getPeriodRangeFromDb(db, input.period);
+	const total = numberFromRow(db.prepare(`
+    SELECT COUNT(DISTINCT session_id) AS total
+    FROM token_usage_events
+    WHERE created_at_ms >= ? AND created_at_ms < ?
+  `).get(range.startMs, range.endMs), "total");
+	return {
+		items: getSessionSummaries(db, range, pageSize, (page - 1) * pageSize),
+		page,
+		page_size: pageSize,
+		period: input.period,
+		range: rangePayload(range),
+		total,
+		total_pages: Math.max(1, Math.ceil(total / pageSize))
+	};
+}
+async function closeUsageStore() {
+	await flushTokenUsageEvents();
+	await tokenUsageDbStore.close({ beforeClose: (db) => {
+		try {
+			db.exec("PRAGMA wal_checkpoint(TRUNCATE)");
+		} catch {}
+	} });
+	writeQueue = Promise.resolve();
+}
+registerProcessCleanup(closeUsageStore);
+//#endregion
+//#region src/lib/request-events.ts
+const TRACKED = /\/(chat\/completions|responses|messages|alpha\/search)$/;
+function setPluginRequestModel(model) {
+	requestContext.getStore()?.pluginRequest?.setModel(model);
+}
+function requestEventsMiddleware(plugins, port) {
+	return async (c, next) => {
+		if (c.req.method !== "POST" || !TRACKED.test(c.req.path)) return await next();
+		const store = requestContext.getStore();
+		if (!store) return await next();
+		const started = performance.now();
+		const info = {
+			id: randomUUID(),
+			traceId: store.traceId,
+			userAgent: store.userAgent,
+			sessionId: store.sessionAffinity?.trim() || null,
+			path: c.req.path,
+			port,
+			startedAt: Date.now(),
+			model: null,
+			actualModel: null
+		};
+		let finished = false;
+		const signal = c.req.raw.signal;
+		const timing = {
+			id: info.id,
+			firstTokenMs: null,
+			durationMs: null,
+			outcome: null,
+			hasUsage: false,
+			recordFirstToken: () => {
+				if (finished || timing.firstTokenMs !== null) return;
+				timing.firstTokenMs = performance.now() - started;
+				plugins?.publish("request.first_token", {
+					...info,
+					firstTokenMs: timing.firstTokenMs
+				});
+			}
+		};
+		store.requestTiming = timing;
+		let status = 0;
+		const finish = (outcome) => {
+			if (finished) return;
+			finished = true;
+			timing.durationMs = performance.now() - started;
+			timing.outcome = outcome;
+			if (timing.hasUsage) enqueueTokenUsageTiming(timing);
+			plugins?.publish("request.end", {
+				...info,
+				durationMs: timing.durationMs,
+				firstTokenMs: timing.firstTokenMs,
+				status,
+				outcome
+			});
+			signal.removeEventListener("abort", abort);
+		};
+		const abort = () => finish("cancelled");
+		if (plugins?.enabled) store.pluginRequest = {
+			id: info.id,
+			setModel: (model) => {
+				if (finished || typeof model !== "string" || !model || info.model === model) return;
+				info.model = model;
+				plugins.publish("request.model", info);
+			},
+			setActualModel: (model) => {
+				if (finished || typeof model !== "string" || !model || info.actualModel === model) return;
+				info.actualModel = model;
+				plugins.publish("request.actual_model", info);
+			},
+			setSessionId: (sessionId) => {
+				if (!sessionId || info.sessionId === sessionId) return;
+				info.sessionId = sessionId;
+				if (!finished) plugins.publish("request.session", info);
+			},
+			recordFirstToken: timing.recordFirstToken,
+			recordUsage: (event) => plugins.publish("usage.recorded", {
+				...event,
+				requestId: info.id,
+				userAgent: info.userAgent,
+				sessionId: event.session_id || null,
+				actualModel: event.model
+			})
+		};
+		plugins?.publish("request.start", info);
+		signal.addEventListener("abort", abort, { once: true });
+		if (signal.aborted) abort();
+		try {
+			await next();
+			const response = c.res;
+			status = response.status;
+			if (!response.body) {
+				finish(status >= 400 ? "error" : "completed");
+				return;
+			}
+			const reader = response.body.getReader();
+			const body = new ReadableStream({
+				async pull(controller) {
+					try {
+						const result = await reader.read();
+						if (result.done) {
+							finish(status >= 400 ? "error" : "completed");
+							controller.close();
+							reader.releaseLock();
+							return;
+						}
+						controller.enqueue(result.value);
+					} catch (error) {
+						finish("error");
+						controller.error(error);
+						reader.releaseLock();
+					}
+				},
+				async cancel(reason) {
+					finish("cancelled");
+					try {
+						await reader.cancel(reason);
+					} finally {
+						reader.releaseLock();
+					}
+				}
+			});
+			c.res = new Response(body, {
+				status: response.status,
+				statusText: response.statusText,
+				headers: response.headers
+			});
+		} catch (error) {
+			finish("error");
+			throw error;
+		}
+	};
+}
+//#endregion
+//#region src/lib/logger.ts
+const LOG_RETENTION_MS = 6048e5;
+const CLEANUP_INTERVAL_MS = 864e5;
+const LOG_DIR_ENV = "COPILOT_API_LOG_DIR";
+const DEFAULT_LOG_DIR = path.join(PATHS.APP_DIR, "logs");
+const FLUSH_INTERVAL_MS = 1e3;
+const MAX_BUFFER_SIZE = 100;
+const getLogDir = () => process.env[LOG_DIR_ENV]?.trim() || DEFAULT_LOG_DIR;
+const logStreams = /* @__PURE__ */ new Map();
+const logBuffers = /* @__PURE__ */ new Map();
+let runtimeInitialized = false;
+let flushInterval;
+let cleanupInterval;
+let currentLogDateKey;
+const ensureLogDirectory = () => {
+	const logDir = getLogDir();
+	if (!fs.existsSync(logDir)) fs.mkdirSync(logDir, { recursive: true });
+};
+const formatArgs = (args) => args.map((arg) => typeof arg === "string" ? arg : util.inspect(arg, {
+	depth: null,
+	colors: false
+})).join(" ");
+const sanitizeName = (name) => {
+	const normalized = name.toLowerCase().replaceAll(/[^a-z0-9]+/g, "-").replaceAll(/^-+|-+$/g, "");
+	return normalized === "" ? "handler" : normalized;
+};
+const maybeUnref = (timer) => {
+	timer.unref();
+};
+const flushBuffer = (filePath) => {
+	const buffer = logBuffers.get(filePath);
+	if (!buffer) return;
+	logBuffers.delete(filePath);
+	if (buffer.length === 0) return;
+	const stream = getLogStream(filePath);
+	const content = buffer.join("\n") + "\n";
+	stream.write(content, (error) => {
+		if (error) console.warn("Failed to write handler log", error);
+	});
+};
+const flushAllBuffers = () => {
+	for (const filePath of logBuffers.keys()) flushBuffer(filePath);
+};
+const closeLogFile = async (filePath) => {
+	flushBuffer(filePath);
+	const stream = logStreams.get(filePath);
+	logStreams.delete(filePath);
+	logBuffers.delete(filePath);
+	if (!stream || stream.closed) return;
+	await new Promise((resolve) => {
+		stream.once("close", resolve);
+		stream.end();
+	});
+};
+const closePreviousDateLogFiles = (dateKey) => {
+	if (currentLogDateKey === void 0) {
+		currentLogDateKey = dateKey;
+		return;
+	}
+	if (currentLogDateKey === dateKey) return;
+	const filePaths = /* @__PURE__ */ new Set([...logBuffers.keys(), ...logStreams.keys()]);
+	currentLogDateKey = dateKey;
+	for (const filePath of filePaths) closeLogFile(filePath).catch((error) => {
+		console.warn("Failed to rotate handler log", error);
+	});
+};
+const cleanupOldLogs = async () => {
+	const logDir = getLogDir();
+	if (!fs.existsSync(logDir)) return;
+	const now = Date.now();
+	for (const entry of fs.readdirSync(logDir)) {
+		const filePath = path.join(logDir, entry);
+		let stats;
+		try {
+			stats = fs.statSync(filePath);
+		} catch {
+			continue;
+		}
+		if (!stats.isFile() || now - stats.mtimeMs <= LOG_RETENTION_MS) continue;
+		try {
+			await closeLogFile(filePath);
+		} catch (error) {
+			console.warn(`Failed to close old handler log: ${filePath}`, error);
+			continue;
+		}
+		try {
+			fs.rmSync(filePath);
+		} catch (error) {
+			console.warn(`Failed to remove old handler log: ${filePath}`, error);
+		}
+	}
+};
+const scheduleLogCleanup = () => {
+	cleanupOldLogs().catch((error) => {
+		console.warn("Failed to clean up handler logs", error);
+	});
+};
+const flushAndCloseLogStreams = () => {
+	flushAllBuffers();
+	for (const stream of logStreams.values()) stream.end();
+	logStreams.clear();
+	logBuffers.clear();
+	currentLogDateKey = void 0;
+};
+const shutdownLoggerRuntime = () => {
+	if (flushInterval) {
+		clearInterval(flushInterval);
+		flushInterval = void 0;
+	}
+	if (cleanupInterval) {
+		clearInterval(cleanupInterval);
+		cleanupInterval = void 0;
+	}
+	flushAndCloseLogStreams();
+	runtimeInitialized = false;
+};
+const initializeLoggerRuntime = () => {
+	if (runtimeInitialized) return;
+	runtimeInitialized = true;
+	ensureLogDirectory();
+	scheduleLogCleanup();
+	flushInterval = setInterval(flushAllBuffers, FLUSH_INTERVAL_MS);
+	maybeUnref(flushInterval);
+	cleanupInterval = setInterval(scheduleLogCleanup, CLEANUP_INTERVAL_MS);
+	maybeUnref(cleanupInterval);
+	registerProcessCleanup(shutdownLoggerRuntime);
+};
+const getLogStream = (filePath) => {
+	initializeLoggerRuntime();
+	let stream = logStreams.get(filePath);
+	if (!stream || stream.destroyed) {
+		const createdStream = fs.createWriteStream(filePath, { flags: "a" });
+		stream = createdStream;
+		logStreams.set(filePath, createdStream);
+		createdStream.on("error", (error) => {
+			console.warn("Log stream error", error);
+			if (logStreams.get(filePath) === createdStream) logStreams.delete(filePath);
+		});
+	}
+	return stream;
+};
+const appendLine = (filePath, line) => {
+	let buffer = logBuffers.get(filePath);
+	if (!buffer) {
+		buffer = [];
+		logBuffers.set(filePath, buffer);
+	}
+	buffer.push(line);
+	if (buffer.length >= MAX_BUFFER_SIZE) flushBuffer(filePath);
+};
+const debugLazy = (logger, factory) => {
+	if (!state.verbose) return;
+	logger.debug(...factory());
+};
+const debugJson = (logger, label, value) => {
+	debugLazy(logger, () => [label, JSON.stringify(value)]);
+};
+const debugJsonAsync = async (logger, label, factory) => {
+	if (!state.verbose) return;
+	logger.debug(label, JSON.stringify(await factory()));
+};
+const debugJsonTail = (logger, label, { value, tailLength = 400 }) => {
+	debugLazy(logger, () => [label, JSON.stringify(value).slice(-tailLength)]);
+};
+const createHandlerLogger = (name) => {
+	const sanitizedName = sanitizeName(name);
+	const instance = consola.withTag(name);
+	if (state.verbose) instance.level = 5;
+	instance.setReporters([]);
+	instance.addReporter({ log(logObj) {
+		initializeLoggerRuntime();
+		const traceId = requestContext.getStore()?.traceId;
+		const date = logObj.date;
+		const dateKey = date.toLocaleDateString("sv-SE");
+		const timestamp = date.toLocaleString("sv-SE", { hour12: false });
+		const filePath = path.join(getLogDir(), `${sanitizedName}-${dateKey}.log`);
+		const message = formatArgs(logObj.args);
+		const traceIdStr = traceId ? ` [${traceId}]` : "";
+		const line = `[${timestamp}] [${logObj.type}] [${logObj.tag || name}]${traceIdStr}${message ? ` ${message}` : ""}`;
+		closePreviousDateLogFiles(dateKey);
+		appendLine(filePath, line);
+	} });
+	return instance;
+};
+//#endregion
+//#region src/lib/models.ts
+/**
+* Converts a Copilot upstream model ID to a client-friendly ID that Claude Code
+* and Claude Desktop recognize (dots in version replaced with hyphens).
+* e.g. "claude-sonnet-4.6" -> "claude-sonnet-4-6"
+* Non-Claude models are returned unchanged.
+*/
+const toClientModelId = (modelId) => {
+	const normalized = normalizeSdkModelId(modelId);
+	if (!normalized) return modelId;
+	const versionHyphenated = normalized.version.replaceAll(".", "-");
+	return `claude-${normalized.family}-${versionHyphenated}`;
+};
+const findEndpointModel = (sdkModelId) => {
+	const models = state.models?.data ?? [];
+	const exactMatch = models.find((m) => m.id === sdkModelId);
+	if (exactMatch) return exactMatch;
+	const normalized = normalizeSdkModelId(sdkModelId);
+	if (!normalized) return;
+	const modelName = `claude-${normalized.family}-${normalized.version}`;
+	const model = models.find((m) => m.id === modelName);
+	if (model) return model;
+};
+/**
+* Normalizes an SDK model ID to extract the model family and version.
+* this method from github copilot extension
+* Examples:
+* - "claude-opus-4-5-20251101" -> { family: "opus", version: "4.5" }
+* - "claude-3-5-sonnet-20241022" -> { family: "sonnet", version: "3.5" }
+* - "claude-sonnet-4-20250514" -> { family: "sonnet", version: "4" }
+* - "claude-haiku-3-5-20250514" -> { family: "haiku", version: "3.5" }
+* - "claude-haiku-4.5" -> { family: "haiku", version: "4.5" }
+*/
+const normalizeSdkModelId = (sdkModelId) => {
+	const withoutDate = sdkModelId.toLowerCase().replace(/-\d{8}$/, "");
+	const pattern1 = withoutDate.match(/^claude-(\w+)-(\d+)\.(\d+)$/);
+	if (pattern1) return {
+		family: pattern1[1],
+		version: `${pattern1[2]}.${pattern1[3]}`
+	};
+	const pattern2 = withoutDate.match(/^claude-(\w+)-(\d+)-(\d+)$/);
+	if (pattern2) return {
+		family: pattern2[1],
+		version: `${pattern2[2]}.${pattern2[3]}`
 	};
-}
-async function getTokenUsageDailySummary(period) {
-	if (!isTokenUsageStorageEnabled()) return createEmptyDailySummary(period);
-	await flushTokenUsageEvents();
-	const db = await getDb();
-	const range = getPeriodRangeFromDb(db, period);
-	const intervals = createDailyIntervals(range);
-	const dailySummaries = getDailyModelSummaries(db, intervals);
-	return {
-		byModel: getModelSummaries(db, range),
-		days: intervals.map((interval) => createDailyBucket(interval, dailySummaries.get(interval.date) ?? [])),
-		period,
-		range: rangePayload(range),
-		totals: totalsFromRow(getTotalsRow(db, range), getCostRows(db, range))
+	const pattern3 = withoutDate.match(/^claude-(\d+)-(\d+)-(\w+)$/);
+	if (pattern3) return {
+		family: pattern3[3],
+		version: `${pattern3[1]}.${pattern3[2]}`
 	};
-}
-async function getTokenUsageEventsPage(input) {
-	if (!isTokenUsageStorageEnabled()) return createEmptyEventsPage(input);
-	await flushTokenUsageEvents();
-	const page = Math.max(1, Math.floor(input.page));
-	const pageSize = Math.min(100, Math.max(1, Math.floor(input.pageSize)));
-	const offset = (page - 1) * pageSize;
-	const db = await getDb();
-	const range = getPeriodRangeFromDb(db, input.period);
-	const totalRow = db.prepare(`
-    SELECT COUNT(*) AS total
-    FROM token_usage_events
-    WHERE created_at_ms >= ? AND created_at_ms < ?
-  `).get(range.startMs, range.endMs);
-	const rows = db.prepare(`
-    SELECT
-      id,
-      created_at_ms,
-      created_at_utc,
-      trace_id,
-      session_id,
-      user_id,
-      source,
-      endpoint,
-      provider_name,
-      model,
-      input_tokens,
-      output_tokens,
-      cache_read_input_tokens,
-      cache_creation_input_tokens,
-      total_nano_aiu,
-      total_tokens,
-      cost_currency,
-      total_cost_nanos,
-      cost_source
-    FROM token_usage_events
-    WHERE created_at_ms >= ? AND created_at_ms < ?
-    ORDER BY created_at_ms DESC, id DESC
-    LIMIT ? OFFSET ?
-  `).all(range.startMs, range.endMs, pageSize, offset);
-	const total = numberFromRow(totalRow, "total");
-	return {
-		items: rows.map((row) => usageEventFromRow(row)),
-		page,
-		page_size: pageSize,
-		period: input.period,
-		range: {
-			end_ms: range.endMs,
-			end_utc: new Date(range.endMs).toISOString(),
-			start_ms: range.startMs,
-			start_utc: new Date(range.startMs).toISOString()
-		},
-		total,
-		total_pages: Math.max(1, Math.ceil(total / pageSize))
+	const pattern4 = withoutDate.match(/^claude-(\w+)-(\d+)$/);
+	if (pattern4) return {
+		family: pattern4[1],
+		version: pattern4[2]
 	};
-}
-async function getTokenUsageSessionsPage(input) {
-	const page = Math.max(1, Math.floor(input.page));
-	const pageSize = Math.min(100, Math.max(1, Math.floor(input.pageSize)));
-	if (!isTokenUsageStorageEnabled()) return {
-		...createEmptyEventsPage(input),
-		items: []
+	const pattern5 = withoutDate.match(/^claude-(\d+)-(\w+)$/);
+	if (pattern5) return {
+		family: pattern5[2],
+		version: pattern5[1]
 	};
-	await flushTokenUsageEvents();
-	const db = await getDb();
-	const range = getPeriodRangeFromDb(db, input.period);
-	const total = numberFromRow(db.prepare(`
-    SELECT COUNT(DISTINCT session_id) AS total
-    FROM token_usage_events
-    WHERE created_at_ms >= ? AND created_at_ms < ?
-  `).get(range.startMs, range.endMs), "total");
+};
+//#endregion
+//#region src/lib/provider-model.ts
+const parseProviderModelAlias = (model) => {
+	const separatorIndex = model.indexOf("/");
+	if (separatorIndex <= 0 || separatorIndex === model.length - 1) return null;
+	const provider = model.slice(0, separatorIndex).trim();
+	const providerModel = model.slice(separatorIndex + 1).trim();
+	if (!provider || !providerModel) return null;
 	return {
-		items: getSessionSummaries(db, range, pageSize, (page - 1) * pageSize),
-		page,
-		page_size: pageSize,
-		period: input.period,
-		range: rangePayload(range),
-		total,
-		total_pages: Math.max(1, Math.ceil(total / pageSize))
+		model: providerModel,
+		provider
 	};
+};
+const createFallbackModel = (modelId) => ({
+	capabilities: {
+		family: "provider",
+		limits: {},
+		object: "model_capabilities",
+		supports: {},
+		tokenizer: "o200k_base",
+		type: "chat"
+	},
+	id: modelId,
+	model_picker_enabled: false,
+	name: modelId,
+	object: "model",
+	preview: false,
+	vendor: "provider",
+	version: "unknown"
+});
+//#endregion
+//#region src/lib/provider-resolver.ts
+const AZURE_OPENAI_SCOPE = "https://cognitiveservices.azure.com/.default";
+let cachedAzureAccessTokenProvider = null;
+/**
+* Lazily creates the Entra token provider on first use and reuses it
+* afterwards. Importing @azure/identity eagerly adds ~0.5s to every process
+* startup, and constructing a new DefaultAzureCredential per call would drop
+* the token cache, so both the import and the credential are singletons.
+*/
+async function getAzureAccessTokenProvider() {
+	cachedAzureAccessTokenProvider ??= (async () => {
+		try {
+			const { DefaultAzureCredential, getBearerTokenProvider } = await import("@azure/identity");
+			return getBearerTokenProvider(new DefaultAzureCredential(), AZURE_OPENAI_SCOPE);
+		} catch (error) {
+			cachedAzureAccessTokenProvider = null;
+			throw error;
+		}
+	})();
+	return cachedAzureAccessTokenProvider;
 }
-async function closeUsageStore() {
-	await flushTokenUsageEvents();
-	await tokenUsageDbStore.close({ beforeClose: (db) => {
+async function getDefaultAzureAccessToken() {
+	return (await getAzureAccessTokenProvider())();
+}
+function isMissingCodexCredentialsError(error) {
+	return error instanceof Error && error.message === "Codex credentials not found. Run `copilot-api auth login --provider codex` first.";
+}
+async function resolveProviderConfig(providerName, getAzureAccessToken = getDefaultAzureAccessToken) {
+	const normalizedProviderName = providerName.trim();
+	if (!normalizedProviderName) return null;
+	if (normalizedProviderName === "codex") {
+		if (getRawProviderConfig(normalizedProviderName)?.enabled === false) return null;
 		try {
-			db.exec("PRAGMA wal_checkpoint(TRUNCATE)");
-		} catch {}
-	} });
-	writeQueue = Promise.resolve();
+			await setupCodexToken();
+		} catch (error) {
+			if (isMissingCodexCredentialsError(error)) return null;
+			throw error;
+		}
+		const providerConfig = getProviderConfig(normalizedProviderName);
+		if (!providerConfig) return null;
+		return {
+			...providerConfig,
+			apiKey: state.codexAccessToken ?? providerConfig.apiKey
+		};
+	}
+	const providerConfig = getProviderConfig(normalizedProviderName);
+	if (providerConfig?.name === "xai" && providerConfig.authType === "oauth2") {
+		const accessToken = await getXaiAccessToken();
+		return accessToken ? {
+			...providerConfig,
+			baseUrl: providerConfig.baseUrl === "https://api.x.ai" ? XAI_API_BASE_URL : providerConfig.baseUrl,
+			apiKey: accessToken
+		} : null;
+	}
+	if (providerConfig?.authType !== "azure-entra") return providerConfig;
+	try {
+		return {
+			...providerConfig,
+			apiKey: await getAzureAccessToken()
+		};
+	} catch (error) {
+		throw new Error(`Failed to acquire Azure Entra token for provider '${normalizedProviderName}': ${error instanceof Error ? error.message : String(error)}`, { cause: error });
+	}
+}
+/**
+* Returns the parsed "provider/model" alias only when that provider is
+* actually configured, otherwise null so the caller falls through to its
+* default flow. GitHub Copilot enterprise models can ship with namespaced ids
+* such as "org/family/model": without this check the first segment is
+* misrouted to a non-existent provider and surfaced as a 400/404.
+*/
+async function ensureConfiguredProviderModelAlias(alias, resolveConfig = resolveProviderConfig) {
+	if (!alias) return null;
+	return await resolveConfig(alias.provider) ? alias : null;
+}
+/** parseProviderModelAlias + ensureConfiguredProviderModelAlias. */
+async function resolveConfiguredProviderModelAlias(model, resolveConfig = resolveProviderConfig) {
+	return ensureConfiguredProviderModelAlias(parseProviderModelAlias(model), resolveConfig);
 }
-registerProcessCleanup(closeUsageStore);
 //#endregion
 //#region src/lib/token-usage/pricing.ts
 const COST_NANOS_PER_UNIT = 1e9;
@@ -1385,6 +1544,7 @@ function costNanosForTokens(tokens, pricePerMillionTokens) {
 //#endregion
 //#region src/lib/token-usage/index.ts
 const tokenUsageEventBus = new EventBus();
+const USER_AGENT_MAX_LENGTH = 64;
 function resolveTraceId(traceId) {
 	return traceId?.trim() || requestContext.getStore()?.traceId || generateTraceId();
 }
@@ -1398,11 +1558,16 @@ function resolveUserId(input) {
 function toPersistedEvent(input) {
 	if (!hasAnyToken(input)) return null;
 	const now = /* @__PURE__ */ new Date();
+	const timing = requestContext.getStore()?.requestTiming;
 	const cost = resolveTokenUsageCost({
 		...input,
 		at: now
 	});
 	return {
+		request_id: timing?.id ?? null,
+		first_token_ms: timing?.firstTokenMs ?? null,
+		duration_ms: timing?.durationMs ?? null,
+		request_outcome: timing?.outcome ?? null,
 		cache_creation_input_tokens: normalizeToken(input.cache_creation_input_tokens),
 		cache_read_input_tokens: normalizeToken(input.cache_read_input_tokens),
 		cost_currency: cost?.currency ?? null,
@@ -1420,6 +1585,7 @@ function toPersistedEvent(input) {
 		total_cost_nanos: cost?.total_cost_nanos ?? null,
 		total_tokens: resolveTotalTokens(input),
 		trace_id: resolveTraceId(input.traceId),
+		user_agent: (requestContext.getStore()?.userAgent ?? "").slice(0, USER_AGENT_MAX_LENGTH),
 		user_id: resolveUserId(input)
 	};
 }
@@ -1427,9 +1593,15 @@ tokenUsageEventBus.subscribe("token_usage.recorded", enqueueTokenUsageWrite);
 function recordTokenUsageEvent(input) {
 	const event = toPersistedEvent(input);
 	if (!event) return;
+	const observer = requestContext.getStore()?.pluginRequest;
+	const timing = requestContext.getStore()?.requestTiming;
+	if (timing) timing.hasUsage = true;
+	observer?.setSessionId(event.session_id);
+	observer?.recordUsage(event);
 	tokenUsageEventBus.publish("token_usage.recorded", event);
 }
 function createTokenUsageRecorder(options) {
+	requestContext.getStore()?.pluginRequest?.setSessionId(resolveTokenUsageSessionId(options.sessionId, options.fallbackSessionId));
 	return (usage) => {
 		recordTokenUsageEvent({
 			...usage,
@@ -1805,6 +1977,7 @@ const createResponses = async (payload, { vision, initiator, subagentMarker, req
 	payload.service_tier = void 0;
 	consola.log(`<-- model: ${payload.model}`);
 	const effectiveTransport = compactType === 1 ? "http" : transport;
+	setPluginRequestActualModel(payload.model);
 	if (payload.stream === true && effectiveTransport === "websocket") {
 		clientSignal?.throwIfAborted();
 		const websocketRequest = prepareResponsesWebSocketRequest(payload, headers, {
@@ -2058,6 +2231,7 @@ async function forwardProviderMessages(providerConfig, payload, requestHeaders,
 	const headers = buildProviderUpstreamHeaders(providerConfig, requestHeaders);
 	applyOpencodeSessionHeader(providerConfig, headers, resolveOpencodeMessagesSession(payload));
 	const transportConfig = getUpstreamTransportConfig();
+	setPluginRequestActualModel(payload.model);
 	return await fetchUpstreamWithLifecycle(resolveProviderEndpointUrl(providerConfig, "messages"), {
 		method: "POST",
 		headers,
@@ -2073,6 +2247,7 @@ async function forwardProviderChatCompletions(providerConfig, payload, requestHe
 	const headers = buildProviderUpstreamHeaders(providerConfig, requestHeaders);
 	applyOpencodeSessionHeader(providerConfig, headers, payload.prompt_cache_key?.trim() || void 0);
 	const transportConfig = getUpstreamTransportConfig();
+	setPluginRequestActualModel(payload.model);
 	return await fetchUpstreamWithLifecycle(resolveProviderEndpointUrl(providerConfig, "chat/completions"), {
 		method: "POST",
 		headers,
@@ -2089,6 +2264,7 @@ async function forwardProviderResponses(providerConfig, payload, requestHeaders,
 	const headers = buildProviderUpstreamHeaders(providerConfig, requestHeaders);
 	applyOpencodeSessionHeader(providerConfig, headers, payload.prompt_cache_key?.trim() || void 0);
 	await applyXaiGrokBuildHeaders(providerConfig, headers, payload);
+	setPluginRequestActualModel(payload.model);
 	return await fetchUpstreamWithLifecycle(resolveProviderEndpointUrl(providerConfig, "responses"), {
 		method: "POST",
 		headers,
@@ -2713,8 +2889,9 @@ function parseDebugBody(body) {
 		return body;
 	}
 }
-async function forwardCodexAlphaSearchRequest(request) {
+async function forwardCodexAlphaSearchRequest(request, model) {
 	await debugJsonAsync(logger$18, "alpha_search.codex.request", async () => ({ body: parseDebugBody(await request.clone().text()) }));
+	if (model) setPluginRequestActualModel(model);
 	const upstreamResponse = await alphaSearchRouteDependencies.forwardCodexAlphaSearch(request);
 	await debugJsonAsync(logger$18, "alpha_search.codex.response", async () => ({
 		body: parseDebugBody(await upstreamResponse.clone().text()),
@@ -2730,12 +2907,12 @@ function createAlphaSearchRequest(request, payload) {
 		signal: request.signal
 	});
 }
-async function handleCodexRequest(c, request, resolvedProviderConfig) {
+async function handleCodexRequest(c, request, resolvedProviderConfig, model) {
 	if (!(resolvedProviderConfig ?? await alphaSearchRouteDependencies.resolveProviderConfig("codex"))) return c.json({ error: {
 		message: "Provider 'codex' not found or disabled",
 		type: "invalid_request_error"
 	} }, 404);
-	return await forwardCodexAlphaSearchRequest(request);
+	return await forwardCodexAlphaSearchRequest(request, model);
 }
 function invalidRequest$1(c, message) {
 	return c.json({ error: {
@@ -2750,7 +2927,9 @@ async function parseAlphaSearchBody(c) {
 	} catch {
 		return invalidRequest$1(c, "Invalid alpha search request: expected JSON body");
 	}
-	if (typeof body?.model !== "string") return invalidRequest$1(c, "Invalid alpha search request: model must be a string");
+	const model = body?.model;
+	if (typeof model !== "string") return invalidRequest$1(c, "Invalid alpha search request: model must be a string");
+	setPluginRequestModel(model);
 	return body;
 }
 /**
@@ -2769,12 +2948,12 @@ async function handleAlphaSearchRequest(c, resolvedProviderConfig) {
 	let providerModelAlias = parseProviderModelAlias(payload.model);
 	if (providerModelAlias) {
 		payload.model = providerModelAlias.model;
-		if (providerModelAlias.provider === "codex") return await handleCodexRequest(c, createAlphaSearchRequest(c.req.raw, payload));
+		if (providerModelAlias.provider === "codex") return await handleCodexRequest(c, createAlphaSearchRequest(c.req.raw, payload), void 0, payload.model);
 	}
 	if (alphaSearchRouteDependencies.isAlphaSearchCodexPriorityEnabled()) {
 		if (await alphaSearchRouteDependencies.resolveProviderConfig("codex")) {
 			if (!payload.model.startsWith("gpt")) payload.model = "gpt-5.6-luna";
-			return await forwardCodexAlphaSearchRequest(createAlphaSearchRequest(c.req.raw, payload));
+			return await forwardCodexAlphaSearchRequest(createAlphaSearchRequest(c.req.raw, payload), payload.model);
 		}
 	}
 	const messagesBackedModel = await isMessagesBackedModel(resolvedRequestedModel);
@@ -2788,7 +2967,7 @@ async function handleAlphaSearchRequest(c, resolvedProviderConfig) {
 		providerModelAlias = parseProviderModelAlias(payload.model);
 		if (providerModelAlias) {
 			payload.model = providerModelAlias.model;
-			if (providerModelAlias.provider === "codex") return await handleCodexRequest(c, createAlphaSearchRequest(c.req.raw, payload));
+			if (providerModelAlias.provider === "codex") return await handleCodexRequest(c, createAlphaSearchRequest(c.req.raw, payload), void 0, payload.model);
 		}
 	}
 	const fallbackRequestedModel = messagesBackedModel ? payload.model : requestedModel;
@@ -2832,6 +3011,46 @@ alphaSearchRoutes.post("/", async (c) => {
 	}
 });
 //#endregion
+//#region src/lib/first-token.ts
+const nonempty = (value) => typeof value === "string" && value.length > 0;
+function recordChatCompletionFirstToken(chunk) {
+	const observer = requestContext.getStore()?.requestTiming ?? requestContext.getStore()?.pluginRequest;
+	if (!observer || !chunk) return;
+	if (chunk.choices?.some(({ delta }) => delta && (nonempty(delta.content) || nonempty(delta.reasoning_text) || nonempty(delta.reasoning_content) || nonempty(delta.reasoning) || delta.tool_calls?.some((tool) => nonempty(tool.function?.arguments))))) observer.recordFirstToken();
+}
+function recordResponsesFirstToken(event) {
+	const observer = requestContext.getStore()?.requestTiming ?? requestContext.getStore()?.pluginRequest;
+	if (!observer || !event) return;
+	switch (event.type) {
+		case "response.output_item.added":
+			if (event.item?.type === "reasoning") observer.recordFirstToken();
+			break;
+		case "response.reasoning_summary_part.added":
+			observer.recordFirstToken();
+			break;
+		case "response.output_text.delta":
+		case "response.reasoning_summary_text.delta":
+		case "response.function_call_arguments.delta":
+		case "response.custom_tool_call_input.delta": if (nonempty(event.delta)) observer.recordFirstToken();
+	}
+}
+function recordAnthropicFirstToken(event) {
+	const observer = requestContext.getStore()?.requestTiming ?? requestContext.getStore()?.pluginRequest;
+	if (!observer || !event) return;
+	if (event.type === "content_block_delta") {
+		const { delta } = event;
+		if (delta.type === "text_delta" && nonempty(delta.text) || delta.type === "thinking_delta" && nonempty(delta.thinking) || delta.type === "input_json_delta" && nonempty(delta.partial_json)) observer.recordFirstToken();
+	} else if (event.type === "content_block_start") {
+		const block = event.content_block;
+		if (block.type === "text" && nonempty(block.text) || block.type === "thinking" || (block.type === "tool_use" || block.type === "server_tool_use") && block.input && Object.keys(block.input).length > 0) observer.recordFirstToken();
+	}
+}
+function recordGoogleFirstToken(response) {
+	const observer = requestContext.getStore()?.requestTiming ?? requestContext.getStore()?.pluginRequest;
+	if (!observer) return;
+	if (response.candidates?.some((candidate) => candidate.content?.parts.some((part) => part.thought === true || nonempty(part.text) || nonempty(part.functionCall?.name)))) observer.recordFirstToken();
+}
+//#endregion
 //#region src/lib/sse.ts
 async function writeSSEIfConnected(stream, message) {
 	if (stream.aborted) return false;
@@ -3039,6 +3258,7 @@ const streamProviderChatCompletions = (c, upstreamResponse, options) => {
 				debugJson(logger$17, "provider.chat_completions.stream_chunk", chunk);
 				if (chunk.data && chunk.data !== "[DONE]") {
 					const parsedChunk = parseChatCompletionChunkData(chunk.data);
+					recordChatCompletionFirstToken(parsedChunk);
 					if (parsedChunk?.usage) usage = normalizeOpenAIUsage(parsedChunk.usage);
 				}
 				await writeSSEIfConnected(stream, {
@@ -3076,6 +3296,7 @@ const createChatCompletions = async (payload, options) => {
 	prepareForCompact(headers, options.compactType);
 	consola.log(`<-- model: ${payload.model}`);
 	const transportConfig = getUpstreamTransportConfig();
+	setPluginRequestActualModel(payload.model);
 	const response = await fetchUpstreamWithLifecycle(`${copilotBaseUrl(state)}/chat/completions`, {
 		method: "POST",
 		headers,
@@ -3098,6 +3319,7 @@ const createChatCompletions = async (payload, options) => {
 const logger$16 = createHandlerLogger("chat-completions-handler");
 async function handleCompletion$1(c) {
 	let payload = await c.req.json();
+	setPluginRequestModel(payload.model);
 	const requestedModel = payload.model;
 	payload.model = resolveMappedModel(payload.model);
 	if (payload.model !== requestedModel) consola.debug(`Resolved model mapping: ${requestedModel} -> ${payload.model}`);
@@ -3151,6 +3373,7 @@ async function handleCompletion$1(c) {
 		for await (const chunk of response) {
 			debugJson(logger$16, "Streaming chunk:", chunk);
 			const parsedChunk = parseChatCompletionChunk(chunk);
+			recordChatCompletionFirstToken(parsedChunk);
 			if (parsedChunk?.usage || parsedChunk?.copilot_usage) usage = {
 				...normalizeOpenAIUsage(parsedChunk.usage),
 				total_nano_aiu: normalizeOptionalToken(parsedChunk.copilot_usage?.total_nano_aiu)
@@ -8627,6 +8850,7 @@ async function forwardGoogleGenerateContent(provider, model, payload, requestHea
 	applyOpencodeSessionHeader(provider, headers, options.sessionId);
 	headers.accept = options.stream ? "text/event-stream" : "application/json";
 	const transport = getUpstreamTransportConfig();
+	setPluginRequestActualModel(model.replace(/^models\//u, ""));
 	return await fetchUpstreamWithLifecycle(url, {
 		method: "POST",
 		headers,
@@ -8689,6 +8913,7 @@ async function handleGoogleProviderMessages(c, payload, provider, usageEndpoint)
 			if (response.headers.get("content-type")?.includes("text/event-stream")) for await (const event of events(response, signal)) {
 				if (!event.data || event.data === "[DONE]") continue;
 				const response = JSON.parse(event.data);
+				recordGoogleFirstToken(response);
 				debugJson(logger$13, "provider.messages.google.raw_stream_event:", response);
 				await writeEvents(translator.push(response));
 			}
@@ -8718,6 +8943,7 @@ const providerMessagesHandlerDependencies = { resolveProviderConfig };
 async function handleProviderMessages(c) {
 	const provider = c.req.param("provider");
 	const payload = await c.req.json();
+	setPluginRequestModel(payload.model);
 	payload.model = fromClaudeDiscoveryModelId(payload.model);
 	applyClaudeNoToolsEffort(payload, c.req.header("user-agent"));
 	const claudeAutoModel = getClaudeAutoModel();
@@ -9160,6 +9386,7 @@ const streamOpenAICompatibleProviderMessages = ({ c, modelConfig, payload, prici
 					continue;
 				}
 				const parsed = parseOpenAICompatibleStreamChunk(chunk.data);
+				recordChatCompletionFirstToken(parsed);
 				if (!parsed) continue;
 				if (parsed.usage) usage = normalizeOpenAIUsage(parsed.usage);
 				const events = translateChunkToAnthropicEvents(parsed, streamState);
@@ -9217,6 +9444,7 @@ const streamResponsesProviderMessages = ({ c, modelConfig, payload, pricingCurre
 				continue;
 			}
 			const parsed = parseResponsesProviderStreamChunk(chunk.data, providerConfig);
+			recordResponsesFirstToken(parsed);
 			if (!parsed) continue;
 			if (parsed.type === "response.completed" || parsed.type === "response.failed" || parsed.type === "response.incomplete") usage = normalizeResponsesUsage(parsed.response.usage);
 			const events = translateResponsesStreamEvent(parsed, streamState);
@@ -9267,6 +9495,7 @@ const parseResponsesProviderStreamChunk = (data, providerConfig) => {
 const parseProviderStreamEvent = (data) => {
 	try {
 		const parsed = JSON.parse(data);
+		recordAnthropicFirstToken(parsed);
 		if (parsed.type === "message_start") return {
 			data: JSON.stringify(parsed),
 			model: parsed.message.model,
@@ -9433,6 +9662,7 @@ const createMessages = async (payload, anthropicBetaHeader, options) => {
 	if (anthropicBeta) headers["anthropic-beta"] = anthropicBeta;
 	consola.log(`<-- model: ${payload.model}`);
 	const transportConfig = getUpstreamTransportConfig();
+	setPluginRequestActualModel(payload.model);
 	const response = await fetchUpstreamWithLifecycle(`${copilotBaseUrl(state)}/v1/messages`, {
 		method: "POST",
 		headers,
@@ -9512,6 +9742,7 @@ const handleWithChatCompletions = async (c, anthropicPayload, options) => {
 				if (rawEvent.data === "[DONE]") break;
 				if (!rawEvent.data) continue;
 				const chunk = JSON.parse(rawEvent.data);
+				recordChatCompletionFirstToken(chunk);
 				if (chunk.usage || chunk.copilot_usage) usage = {
 					...normalizeOpenAIUsage(chunk.usage),
 					total_nano_aiu: normalizeOptionalToken(chunk.copilot_usage?.total_nano_aiu)
@@ -9585,6 +9816,7 @@ const handleWithResponsesApi = async (c, anthropicPayload, options) => {
 				if (!data) continue;
 				debugLazy(logger, () => ["Responses raw stream event:", data]);
 				const responseEvent = JSON.parse(data);
+				recordResponsesFirstToken(responseEvent);
 				if (responseEvent.type === "response.completed" || responseEvent.type === "response.failed" || responseEvent.type === "response.incomplete") usage = {
 					...normalizeResponsesUsage(responseEvent.response.usage),
 					total_nano_aiu: normalizeOptionalToken(responseEvent.copilot_usage?.total_nano_aiu)
@@ -9656,6 +9888,7 @@ const handleWithMessagesApi = async (c, anthropicPayload, options) => {
 					if (!data) continue;
 					debugLazy(logger, () => ["Messages raw stream event:", data]);
 					const parsedEvent = parseAnthropicStreamEvent(data);
+					recordAnthropicFirstToken(parsedEvent);
 					if (parsedEvent?.type === "message_start") usage = mergeAnthropicUsage(usage, {
 						...normalizeAnthropicUsage(parsedEvent.message.usage),
 						...normalizeCopilotUsage(parsedEvent.message.copilot_usage)
@@ -9787,6 +10020,7 @@ const messagesFlowHandlers = {
 };
 async function handleCompletion(c) {
 	const anthropicPayload = await c.req.json();
+	setPluginRequestModel(anthropicPayload.model);
 	anthropicPayload.model = fromClaudeDiscoveryModelId(anthropicPayload.model);
 	applyClaudeNoToolsEffort(anthropicPayload, c.req.header("user-agent"));
 	return await handleCompletionPayload(c, anthropicPayload);
@@ -13452,6 +13686,7 @@ const streamProviderResponses = async (c, upstreamResponse, options) => {
 					event: event.type
 				};
 			}
+			recordResponsesFirstToken(event);
 			if (event) {
 				const nextUsage = getResponsesStreamEventUsage(event);
 				if (nextUsage) usage = nextUsage;
@@ -13496,8 +13731,10 @@ providerResponsesRoutes.use(zstdDecompressionMiddleware);
 providerResponsesRoutes.post("/", async (c) => {
 	try {
 		const provider = c.req.param("provider") ?? "";
+		const payload = await c.req.json();
+		setPluginRequestModel(payload.model);
 		return await handleProviderResponsesForProvider(c, {
-			payload: await c.req.json(),
+			payload,
 			provider
 		});
 	} catch (error) {
@@ -13551,6 +13788,7 @@ const responsesHandlerDependencies = {
 };
 const handleResponses = async (c) => {
 	const payload = await c.req.json();
+	setPluginRequestModel(payload.model);
 	const requestedModel = payload.model;
 	payload.model = responsesHandlerDependencies.resolveMappedModel(payload.model);
 	if (payload.model !== requestedModel) consola.debug(`Resolved model mapping: ${requestedModel} -> ${payload.model}`);
@@ -13644,6 +13882,7 @@ const handleResponses = async (c) => {
 				for await (const chunk of { [Symbol.asyncIterator]: () => iterator }) {
 					debugJson(logger$2, "Responses stream chunk:", chunk);
 					const parsedEvent = parseResponsesStreamEvent(chunk);
+					recordResponsesFirstToken(parsedEvent);
 					if (parsedEvent?.type === "response.completed" || parsedEvent?.type === "response.failed" || parsedEvent?.type === "response.incomplete") usage = {
 						...normalizeResponsesUsage(parsedEvent.response.usage),
 						total_nano_aiu: normalizeOptionalToken(parsedEvent.copilot_usage?.total_nano_aiu)
@@ -13927,6 +14166,7 @@ function createServer(options = {}) {
 		allowUnauthenticatedPaths: [],
 		allowWhenNoApiKeys: false
 	}));
+	server.use("*", requestEventsMiddleware(options.plugins, options.port ?? 4141));
 	server.get("/", (c) => c.text("Server running"));
 	server.get("/usage-viewer", (c) => {
 		const usageViewerFileUrl = new URL("../pages/index.html", import.meta.url);
@@ -13966,4 +14206,4 @@ createServer();
 //#endregion
 export { createServer };
 
-//# sourceMappingURL=server-BogSoFE5.js.map
+//# sourceMappingURL=server-CEF4Sh-g.js.map