@tekmidian/pai 1.0.0 → 1.0.2

package/dist/{worker-config-BLqa1yiD.mjs → worker-config-BDYNxbYz.mjs}

@@ -1,37 +1,39 @@
 import { _ as warn, c as ok, i as err, n as dim } from "./utils-J2b9y8PC.mjs";
 import { g as paiHomePath, h as paiHomeDir, l as claudeHomeDir, t as aibrokerHomeDir, u as claudeProjectsDir } from "./runtime-paths-WsO9sJFc.mjs";
-import { Et as parseLedger, G as inlineWorkersYamlKeys, J as parseBuiltinAnthropic, K as migrateWorkersToYaml, Q as relocateWorkersYaml, R as isLoopbackUrl, T as parseProvider, W as initWorkersYaml, Y as parseWorkersYamlDocument, c as WorkersConfigError, et as workersYamlLegacyNotice, ft as parseEntry, it as yamlProviderToRaw, j as readWorkersSection, lt as findTier, m as isLocalProvider, o as PROVIDER_TAGS, p as getProviderOrNative, q as needsWorkersYamlRelocation, t as ANTHROPIC_NATIVE, tt as workersYamlPath, wt as appendLedger, y as nativeAnthropicProvider, z as loopbackUrl } from "./config-D5chi2LO.mjs";
-import { D as readJsonStrict, O as writeJsonAtomic, f as readMainConfigRaw, m as deepMergeConfig, n as CONFIG_FILE, o as expandHome, u as paiConfigFilePath, v as shippedDefaultsDir, y as shippedMainDefaults } from "./config-BSXKyeJ1.mjs";
-import { i as ledgerPath, t as WORKER_ID_RE, u as workersLogDir } from "./paths-DoIyRnhx.mjs";
-import { E as worktreesStatsCached, l as formatWorktreesStats } from "./worktree-zmRf556h.mjs";
-import { T as ownsPidCached, b as loadStatusesCached, g as isLive, v as loadStatus } from "./status-CvVi9cDf.mjs";
-import { C as describeMcp, _ as readPoolMetrics, f as poolMetricsPath, h as poolSettings, i as bundleBuildStamp, l as poolHostProbe, o as poolCarry, u as poolHostStale, v as readPoolState, x as warmOrCold } from "./pool-CZ5MZaXg.mjs";
-import { a as ownerSessionName } from "./scope-D5VSQxe4.mjs";
-import { $t as onBattery, At as applyPolicyToRoutes, Bn as reconcileInstances, Bt as routeText, Dt as isTailnetOrLocalName, E as budgetCache, Fn as instanceEngineKey, Gt as setSessionRoute, I as buildSpeed, In as instanceNumber, Jt as POLICY_MODES, L as percentile, Ln as instancesFromProcesses, Lt as resolveSessionArg, Ot as loopbackRefusal, Pt as flowText, Qn as markTimeOrdered, Qt as findLocal, Rn as readInstances, Rt as routeChoices, Vn as writeInstances, Vt as servingChip, W as currentInventory, Wt as setRoute, Yt as clearPolicy, Zn as isTimeOrdered, _n as probeLoopback, at as readInventoryInput, en as parseDuration, gn as pickLocal, hn as localTierOrder, it as readInventory, jt as clearSessionRoute, ln as readProfiles, nn as routeRun, nr as splitBySessionKind, pn as currentLocalStatus, rn as writePolicy, rr as usageLogFile, rt as processTreeRssMb, sn as getProfile, tn as readPolicy, tr as readUsageCached, wt as duplicateModelIds, zn as readLiveInstances, zt as routeRows } from "./routing-CUyiglfu.mjs";
-import { C as stopLines, T as tierButtons, _ as readLocalConfig, a as LocalTierError, c as fetchServedModels, f as listensOnPort, g as planStart, h as parseServers, l as fmtRss, m as parseLocalConfig, u as healthUrl, v as resolveModelPath, x as stopCommandFor } from "./local-tiers-DPZYqv9m.mjs";
-import { b as testSlots, c as listBuildsSync, g as buildSlots, h as waitReason, i as buildsDir, p as ticketKind } from "./caller-identity-CVTuiLSs.mjs";
-import { g as sharedScheduler, i as localStatusLine, m as schedPlain, o as readLocalStatus, u as readSchedFile } from "./local-candidate-jOjTNlOL.mjs";
-import { l as reorderQueue, o as providerCap, s as queueRows, u as reorderQueueTo } from "./capacity-DaXrVj3J.mjs";
-import { t as LOCAL_PROBE_TIMEOUT_MS } from "./timeouts-OwEONmc8.mjs";
-import { n as OPT, t as CliError } from "./define-DGFu-sCf.mjs";
-import { c as resolveFromModule } from "./postgres-CDnd_k5f.mjs";
-import { c as ensureProxyRunning, l as proxyPort, u as stopProxy } from "./launch-DgqQdAjI.mjs";
-import { a as colorEnabled, f as renderTable, n as readBudget, o as makeStyle } from "./health-C6jZWvRG.mjs";
-import { H as testProvider, o as sessionWorkersLine } from "./run-ulAi5dFD.mjs";
-import { U as planRows, V as loadPlanStatuses, i as localIdleLine, n as markControlsHeld, s as readLocalIdle, t as clickrControlsArgv, yt as psTable } from "./controls-BGkK_Bdi.mjs";
-import { n as diskBundleMtime, r as ensurePoolHost } from "./pool-cli-BxNSPzzY.mjs";
-import { a as readAudit, d as loadPaneRunStatuses, f as paneRunRows } from "./pane-audit-OGYrOAik.mjs";
-import { b as useProvider, f as setClass, g as setWorkersEnabled, l as removeProvider, m as setProviderEnabled, p as setProviderCap, r as classTargetText, s as describeProviders, t as addProvider, v as unsetClass, y as updateProvider } from "./providers-UsKtLcUR.mjs";
-import { C as embeddingRuntimeName, F as cockpitLayout, I as cockpitProbeSettings, O as liveMemoryDeps, T as cockpitMemoryPayload, _ as runtimeJobProgress, d as readCockpitJob, f as readLabels, g as runningRuntimeJobs, k as memoryJobProgress, m as runCockpitAction, o as defaultDeps, t as COCKPIT_ACTION_EVENT, w as readRuntimeCommands } from "./cockpit-actions-ODmt8yF9.mjs";
-import { i as fallbackStatusText, n as fallbackOn, r as fallbackStatus, t as fallbackOff } from "./fallback-BIyEpPsQ.mjs";
-import { appendFileSync, chmodSync, closeSync, copyFileSync, existsSync, fstatSync, mkdirSync, openSync, readFileSync, readSync, readdirSync, renameSync, rmSync, statSync, unlinkSync, writeFileSync } from "node:fs";
+import { Et as parseLedger, G as inlineWorkersYamlKeys, J as parseBuiltinAnthropic, K as migrateWorkersToYaml, Q as relocateWorkersYaml, R as isLoopbackUrl, T as parseProvider, W as initWorkersYaml, Y as parseWorkersYamlDocument, c as WorkersConfigError, et as workersYamlLegacyNotice, ft as parseEntry, it as yamlProviderToRaw, j as readWorkersSection, lt as findTier, m as isLocalProvider, o as PROVIDER_TAGS, p as getProviderOrNative, q as needsWorkersYamlRelocation, t as ANTHROPIC_NATIVE, tt as workersYamlPath, wt as appendLedger, y as nativeAnthropicProvider, z as loopbackUrl } from "./config-XpZWOYOG.mjs";
+import { D as readJsonStrict, O as writeJsonAtomic, f as readMainConfigRaw, m as deepMergeConfig, n as CONFIG_FILE, o as expandHome, u as paiConfigFilePath, v as shippedDefaultsDir, y as shippedMainDefaults } from "./config-Bu1_vbY8.mjs";
+import { i as ledgerPath, t as WORKER_ID_RE, u as workersLogDir } from "./paths-CHifubDh.mjs";
+import { E as worktreesStatsCached, l as formatWorktreesStats } from "./worktree-CrnOqTBW.mjs";
+import { T as ownsPidCached, b as loadStatusesCached, g as isLive, v as loadStatus } from "./status-ggJ7lgVg.mjs";
+import { C as describeMcp, _ as readPoolMetrics, f as poolMetricsPath, h as poolSettings, i as bundleBuildStamp, l as poolHostProbe, o as poolCarry, u as poolHostStale, v as readPoolState, x as warmOrCold } from "./pool-BRvoOkv6.mjs";
+import { a as ownerSessionName } from "./scope-D-h6F6DN.mjs";
+import { $t as findLocal, Bt as routeRows, E as budgetCache, Ft as flowText, Gn as isTimeOrdered, Gt as setRoute, Ht as servingChip, I as buildSpeed, Kn as markTimeOrdered, Kt as setSessionRoute, L as percentile, Mt as clearSessionRoute, Ot as isTailnetOrLocalName, Rt as resolveSessionArg, Tt as duplicateModelIds, Vt as routeText, W as currentInventory, Xn as usageLogFile, Xt as clearPolicy, Yn as splitBySessionKind, Yt as POLICY_MODES, _n as pickLocal, at as readInventoryInput, cn as getProfile, en as onBattery, gn as localTierOrder, in as writePolicy, it as readInventory, jt as applyPolicyToRoutes, kt as loopbackRefusal, mn as currentLocalStatus, nn as readPolicy, rn as routeRun, rt as processTreeRssMb, tn as parseDuration, un as readProfiles, vn as probeLoopback, zt as routeChoices } from "./routing-BAAST9vT.mjs";
+import { C as stopLines, T as tierButtons, _ as readLocalConfig, a as LocalTierError, c as fetchServedModels, f as listensOnPort, g as planStart, h as parseServers, l as fmtRss, m as parseLocalConfig, u as healthUrl, v as resolveModelPath, x as stopCommandFor } from "./local-tiers-BE4jGaMa.mjs";
+import { _ as buildSlots, a as buildsDir, g as waitReason, l as listBuildsSync, m as ticketKind, n as breakStaleMutex, x as testSlots } from "./caller-identity-yLFb7CKq.mjs";
+import { f as sharedScheduler, o as readSchedFile, u as schedPlain } from "./scheduler-CtUyWJ3O.mjs";
+import { l as reorderQueue, o as providerCap, s as queueRows, u as reorderQueueTo } from "./capacity-DEveQdFV.mjs";
+import { a as instancesFromProcesses, d as reconcileInstances, i as instanceNumber, l as readInstances, p as writeInstances, r as instanceEngineKey, u as readLiveInstances } from "./local-instances-DOxSgT_L.mjs";
+import { i as localStatusLine, s as readLocalStatus } from "./local-candidate-DLdwQt0s.mjs";
+import { t as LOCAL_PROBE_TIMEOUT_MS } from "./timeouts-1MLooEBB.mjs";
+import { n as OPT, t as CliError } from "./define-CjJsmq-6.mjs";
+import { c as resolveFromModule } from "./postgres-C6ti4haA.mjs";
+import { c as ensureProxyRunning, l as proxyPort, u as stopProxy } from "./launch-Jh5NVeNe.mjs";
+import { a as colorEnabled, f as renderTable, n as readBudget, o as makeStyle } from "./health-Dnym-qcT.mjs";
+import { H as testProvider, o as sessionWorkersLine } from "./run-6Q9kCPem.mjs";
+import { U as planRows, V as loadPlanStatuses, i as localIdleLine, n as markControlsHeld, s as readLocalIdle, t as clickrControlsArgv, yt as psTable } from "./controls-BcKFHeG1.mjs";
+import { n as diskBundleMtime, r as ensurePoolHost } from "./pool-cli-B_YR8_Gl.mjs";
+import { a as readAudit, d as loadPaneRunStatuses, f as paneRunRows } from "./pane-audit-BQytYa60.mjs";
+import { b as useProvider, f as setClass, g as setWorkersEnabled, l as removeProvider, m as setProviderEnabled, p as setProviderCap, r as classTargetText, s as describeProviders, t as addProvider, v as unsetClass, y as updateProvider } from "./providers-CblTFrhz.mjs";
+import { A as cockpitMemoryPayload, B as cockpitLayout, N as liveMemoryDeps, O as embeddingRuntimeName, P as memoryJobProgress, S as runtimeJobProgress, V as cockpitProbeSettings, c as defaultDeps, g as readLabels, h as readCockpitJob, k as readRuntimeCommands, t as COCKPIT_ACTION_EVENT, v as runCockpitAction, x as runningRuntimeJobs } from "./cockpit-actions-C8GlTzts.mjs";
+import { i as fallbackStatusText, n as fallbackOn, r as fallbackStatus, t as fallbackOff } from "./fallback-DFlItw1Z.mjs";
+import { appendFileSync, chmodSync, closeSync, copyFileSync, existsSync, fstatSync, mkdirSync, openSync, readFileSync, readSync, readdirSync, renameSync, rmSync, rmdirSync, statSync, unlinkSync, writeFileSync } from "node:fs";
+import { createHash } from "node:crypto";
 import { basename, dirname, join, resolve } from "node:path";
 import chalk from "chalk";
 import { homedir, loadavg } from "node:os";
 import { execFile, execFileSync, spawn } from "node:child_process";
 import { fileURLToPath } from "node:url";
 import { LineCounter, isMap, isScalar, parse, parseDocument } from "yaml";
-import { createHash } from "node:crypto";
 import { connect, createServer } from "node:net";
 
 //#region src/cli/commands/providers.ts
@@ -5871,7 +5873,6 @@ function cockpitHtml() {
 * UI that shows the project separately. Unknown ids map to null.
 */
 function defaultSources() {
-	homedir();
 	return {
 		workerDirs: [paiHomePath("logs", "workers"), join(claudeHomeDir(), "logs", "workers")],
 		agentSessionsFile: paiHomePath("agent-sessions.json"),
@@ -5897,6 +5898,11 @@ function jobSessionName(id, jobsDir = defaultSources().jobsDir, now = Date.now()
 	if (hit && now - hit.at < JOB_TTL_MS) return hit.name;
 	const raw = readJson(join(jobsDir, id8, "state.json"))?.name;
 	const name = typeof raw === "string" && raw ? raw : null;
+	jobCache.delete(key);
+	for (const [k, v] of jobCache) {
+		if (now - v.at < JOB_TTL_MS) break;
+		jobCache.delete(k);
+	}
 	jobCache.set(key, {
 		at: now,
 		name
@@ -6035,20 +6041,26 @@ function loadSessionMap(src = defaultSources()) {
 /** The loader below as steps: a `yield` after each directory or file, so a driver can hand the event loop back between them. */
 function* loadSessionMapSteps(src = defaultSources()) {
 	const map = /* @__PURE__ */ new Map();
-	if (src.projectsDir && existsSync(src.projectsDir)) for (const d of readdirSync(src.projectsDir)) {
-		const name = d.split("-").filter(Boolean).pop();
-		if (!name) continue;
-		let files = [];
+	if (src.projectsDir && existsSync(src.projectsDir)) {
+		let projectDirs = [];
 		try {
-			files = readdirSync(join(src.projectsDir, d));
-		} catch {
-			continue;
-		}
-		for (const f of files) {
-			const id = f.endsWith(".jsonl") ? f.slice(0, -6) : "";
-			if (id) map.set(id, { name });
+			projectDirs = readdirSync(src.projectsDir);
+		} catch {}
+		for (const d of projectDirs) {
+			const name = d.split("-").filter(Boolean).pop();
+			if (!name) continue;
+			let files = [];
+			try {
+				files = readdirSync(join(src.projectsDir, d));
+			} catch {
+				continue;
+			}
+			for (const f of files) {
+				const id = f.endsWith(".jsonl") ? f.slice(0, -6) : "";
+				if (id) map.set(id, { name });
+			}
+			yield;
 		}
-		yield;
 	}
 	const agents = existsSync(src.agentSessionsFile) ? readJson(src.agentSessionsFile) : void 0;
 	if (agents) for (const [id, project] of Object.entries(agents)) {
@@ -6087,7 +6099,13 @@ function* loadSessionMapSteps(src = defaultSources()) {
 	}
 	for (const dir of src.workerDirs) {
 		if (!existsSync(dir)) continue;
-		for (const f of readdirSync(dir)) {
+		let dirFiles = [];
+		try {
+			dirFiles = readdirSync(dir);
+		} catch {
+			continue;
+		}
+		for (const f of dirFiles) {
 			if (!f.endsWith(".status")) continue;
 			yield;
 			const s = readJson(join(dir, f));
@@ -6202,7 +6220,7 @@ function* withWorkerAttributionSteps(map, rows, windows, now = Date.now()) {
 			hits.set(w.worker, w);
 		}
 		if (!hits.size) continue;
-		const first = Math.min(...reqs.map((r) => r.t));
+		const first = reqs.reduce((m, r) => Math.min(m, r.t), Infinity);
 		const best = [...hits.values()].sort((a, b) => b.startMs - a.startMs)[0];
 		if (best.startMs > first) continue;
 		out.set(id, {
@@ -6725,26 +6743,86 @@ function tierProviderList() {
 }
 const WHOLE_LOG_TTL_MS = 3e4;
 const wholeLogMemo = /* @__PURE__ */ new WeakMap();
+/** Whole-log figures, folded row by row as the log is read so the rows themselves can be dropped once they are older than any window. */
+var LogAgg = class {
+	/** per provider+model: requests, prompt tokens, first ts (a fold by model owner is applied on top at read time) */
+	pairs = /* @__PURE__ */ new Map();
+	/** tokens per UTC day per model, with the sequence number of the model's first row that day (keeps the model order of a row scan) */
+	daily = /* @__PURE__ */ new Map();
+	seq = 0;
+	add(r) {
+		const model = typeof r.model === "string" ? canonicalModel(r.model) : "";
+		const k = `${r.provider}\u0000${model}`;
+		const p = this.pairs.get(k) ?? {
+			provider: r.provider,
+			model,
+			n: 0,
+			prompt: 0,
+			since: r.ts
+		};
+		p.n++;
+		p.prompt += (r.input_tokens ?? 0) + (r.cache_read_input_tokens ?? 0) + (r.cache_creation_input_tokens ?? 0);
+		if (r.ts < p.since) p.since = r.ts;
+		this.pairs.set(k, p);
+		const day = bucketStart(Date.parse(r.ts), DAY$1);
+		const d = this.daily.get(day) ?? /* @__PURE__ */ new Map();
+		const e = d.get(model) ?? {
+			sum: 0,
+			seq: this.seq
+		};
+		e.sum += (r.input_tokens ?? 0) + (r.output_tokens ?? 0) + (r.cache_creation_input_tokens ?? 0) + (r.cache_read_input_tokens ?? 0);
+		d.set(model, e);
+		this.daily.set(day, d);
+		this.seq++;
+	}
+	/** Forget the day buckets before `cut`: no window looks back further than the 14 days the daily chart shows. */
+	pruneDaily(cut) {
+		for (const day of this.daily.keys()) if (day < cut) this.daily.delete(day);
+	}
+	/** The pairs as the fold of the providers would have rewritten their rows. */
+	folded(providers) {
+		const owners = modelOwners(providers);
+		const all = [...this.pairs.values()];
+		return owners.size ? foldProviders(all, providers, owners) : all;
+	}
+};
+const aggs = /* @__PURE__ */ new WeakMap();
+/** The aggregates a cached log carries; a plain array (tests, callers that bring their own rows) is scanned. */
+function aggOf(rows) {
+	const have = aggs.get(rows);
+	if (have) return have;
+	const a = new LogAgg();
+	for (const r of rows) a.add(r);
+	return a;
+}
 /** The parts of the cockpit payload that read the whole log whatever the window is. Costly on a big log, so they are kept per log array until it grows or 30 s pass (the speed table moves with the clock). */
-function wholeLog(allRows, now, providers) {
+function wholeLog(allRows, agg, now, providers) {
 	const sig = JSON.stringify(Object.entries(providers).map(([k, v]) => [k, v.baseUrl]));
 	const m = wholeLogMemo.get(allRows);
 	if (m && m.n === allRows.length && m.sig === sig && Math.abs(now - m.at) < WHOLE_LOG_TTL_MS) return m.out;
 	const dayFrom = bucketStart(now, DAY$1) - 13 * DAY$1;
 	const days = [];
 	for (let t = dayFrom; t <= now; t += DAY$1) days.push(t);
-	const daily = {};
-	for (const r of allRows) {
-		const t = Date.parse(r.ts);
-		if (t < dayFrom) continue;
-		const arr = daily[canonicalModel(r.model)] ??= days.map(() => 0);
-		arr[Math.floor((bucketStart(t, DAY$1) - dayFrom) / DAY$1)] += (r.input_tokens ?? 0) + (r.output_tokens ?? 0) + (r.cache_creation_input_tokens ?? 0) + (r.cache_read_input_tokens ?? 0);
+	const daily = dict();
+	const inRange = [];
+	for (const [day, ms] of agg.daily) {
+		if (day < dayFrom) continue;
+		for (const [model, e] of ms) inRange.push({
+			day,
+			model,
+			...e
+		});
+	}
+	inRange.sort((a, b) => a.seq - b.seq);
+	for (const e of inRange) {
+		const arr = daily[e.model] ??= days.map(() => 0);
+		arr[Math.floor((e.day - dayFrom) / DAY$1)] += e.sum;
 	}
-	const promptByProvider = {};
-	const promptSince = {};
-	for (const r of allRows) {
-		if (!promptSince[r.provider] || r.ts < promptSince[r.provider]) promptSince[r.provider] = r.ts;
-		promptByProvider[r.provider] = (promptByProvider[r.provider] ?? 0) + (r.input_tokens ?? 0) + (r.cache_read_input_tokens ?? 0) + (r.cache_creation_input_tokens ?? 0);
+	const promptByProvider = dict();
+	const promptSince = dict();
+	for (const p of agg.folded(providers)) {
+		if (!promptSince[p.provider] || p.since < promptSince[p.provider]) promptSince[p.provider] = p.since;
+		promptByProvider[p.provider] = (promptByProvider[p.provider] ?? 0) + p.prompt;
 	}
 	const out = {
 		days,
@@ -6794,23 +6872,23 @@ function windowRows(allRows, from) {
 function buildCockpitData(allRows, win, now = Date.now(), sessions = /* @__PURE__ */ new Map(), pool = null, providers = {}) {
 	const { span, bucket } = WINDOWS[win];
 	const from = bucketStart(now - span, bucket);
+	const agg = aggOf(allRows);
 	allRows = foldedRows(allRows, providers);
 	const rows = windowRows(allRows, from);
 	const retries = retryRows(rows);
 	const ecoCount = /* @__PURE__ */ new Map();
-	for (const r of allRows) {
-		const m = canonicalModel(r.model);
-		const c = ecoCount.get(m) ?? /* @__PURE__ */ new Map();
+	for (const r of agg.folded(providers)) {
+		const c = ecoCount.get(r.model) ?? /* @__PURE__ */ new Map();
 		const e = providerEcosystem(r.provider, providers);
-		c.set(e, (c.get(e) ?? 0) + 1);
-		ecoCount.set(m, c);
+		c.set(e, (c.get(e) ?? 0) + r.n);
+		ecoCount.set(r.model, c);
 	}
-	const ecosystems = {};
+	const ecosystems = dict();
 	for (const [m, c] of ecoCount) ecosystems[m] = [...c.entries()].sort((a, b) => b[1] - a[1] || a[0].localeCompare(b[0]))[0][0];
 	const series = [];
 	for (let t = from; t <= now; t += bucket) series.push({
 		t,
-		models: {}
+		models: dict()
 	});
 	const total = tok();
 	let errors = 0;
@@ -6821,7 +6899,7 @@ function buildCockpitData(allRows, win, now = Date.now(), sessions = /* @__PURE_
 	const byInstance = /* @__PURE__ */ new Map();
 	const bySession = /* @__PURE__ */ new Map();
 	const pause = /* @__PURE__ */ new Map();
-	const bustParts = {};
+	const bustParts = dict();
 	for (const r of rows) {
 		const model = canonicalModel(r.model);
 		const t = Date.parse(r.ts);
@@ -6938,7 +7016,7 @@ function buildCockpitData(allRows, win, now = Date.now(), sessions = /* @__PURE_
 			workerEnd: info?.endMs ?? null
 		};
 	}).sort((a, b) => b.input + b.output + b.cacheWrite + b.cacheRead - (a.input + a.output + a.cacheWrite + a.cacheRead));
-	const { daily, days, promptByProvider, promptSince, speed } = wholeLog(allRows, now, providers);
+	const { daily, days, promptByProvider, promptSince, speed } = wholeLog(allRows, agg, now, providers);
 	const withRatio = (x) => ({
 		...x,
 		hitRatio: hitRatio(x)
@@ -6995,45 +7073,93 @@ function buildCockpitData(allRows, win, now = Date.now(), sessions = /* @__PURE_
 		pool: pool && pool.tasks > 0 ? pool : null
 	};
 }
+const USAGE_HEAD_BYTES = 4096;
+/** A map keyed by logged names (model, provider): no prototype, so a key such as "__proto__" is an ordinary entry. */
+const dict = () => Object.create(null);
 /**
 * The usage log as rows, kept in memory: a log that only grew is read from where the last read
 * stopped (the router appends a row per request, so a full reread of tens of MB on every append
 * would block the event loop); one that shrank or was replaced is read again from the start.
+* The whole-log figures (prompt tokens and first ts per provider, requests per provider+model,
+* tokens per day) are folded into running aggregates as rows are read; the rows themselves are
+* kept for ROWS_KEPT_MS only (the 7 d window reaches back 8 d at most, one day of slack on top)
+* and dropped once a day, so memory follows the window, not the log.
 */
-function cachedRows(file = usageLogFile()) {
+const ROWS_KEPT_MS = 9 * DAY$1;
+const DAILY_KEPT_MS = 15 * DAY$1;
+function cachedRows(file = usageLogFile(), clock = Date.now) {
 	let rows = [];
+	let agg = new LogAgg();
+	let lastTrim = 0;
+	let ordered = true;
 	let offset = 0;
 	let mtimeMs = 0;
+	let ino = -1;
+	let head = "";
+	let lastMs = 0;
+	const fresh = (kept) => {
+		rows = kept;
+		if (ordered) markTimeOrdered(rows);
+		aggs.set(rows, agg);
+	};
+	const reset = () => {
+		ordered = true;
+		agg = new LogAgg();
+		lastTrim = clock();
+		fresh([]);
+		offset = 0;
+		ino = -1;
+		head = "";
+		lastMs = 0;
+		mtimeMs = 0;
+	};
+	reset();
 	return () => {
 		try {
-			const st = statSync(file);
-			if (st.size === offset && st.mtimeMs === mtimeMs) return rows;
-			if (st.size < offset) {
-				rows = [];
-				offset = 0;
-			}
 			const fd = openSync(file, "r");
 			try {
+				const st = fstatSync(fd);
+				const headLen = Math.min(USAGE_HEAD_BYTES, st.size);
+				const headBuf = Buffer.alloc(headLen);
+				readSync(fd, headBuf, 0, headLen, 0);
+				const nowHead = headBuf.toString("latin1");
+				if (offset > 0 && (st.ino !== ino || st.size < offset || !nowHead.startsWith(head))) reset();
+				if (st.size === offset && st.mtimeMs === mtimeMs) return rows;
 				const buf = Buffer.alloc(st.size - offset);
 				const n = readSync(fd, buf, 0, buf.length, offset);
-				const text = buf.subarray(0, n).toString("utf8");
-				const end = text.lastIndexOf("\n") + 1;
-				const added = [];
-				for (const l of text.slice(0, end).split("\n")) {
+				let chunk = buf.subarray(0, n);
+				const nl = chunk.lastIndexOf(10);
+				chunk = nl < 0 ? chunk.subarray(0, 0) : chunk.subarray(0, nl + 1);
+				const from = rows.length;
+				for (const l of chunk.toString("utf8").split("\n")) {
 					if (!l) continue;
 					try {
-						added.push(JSON.parse(l));
+						const r = JSON.parse(l);
+						agg.add(r);
+						rows.push(r);
 					} catch {}
 				}
-				if (added.length) rows = rows.concat(added);
-				offset += Buffer.byteLength(text.slice(0, end), "utf8");
+				for (let i = from; i < rows.length; i++) {
+					const t = Date.parse(rows[i].ts);
+					if (t < lastMs) ordered = false;
+					else if (t >= 0) lastMs = t;
+				}
+				if (!ordered) fresh([...rows]);
+				const at = clock();
+				if (at - lastTrim >= DAY$1) {
+					lastTrim = at;
+					agg.pruneDaily(bucketStart(at, DAY$1) - DAILY_KEPT_MS);
+					fresh(rows.filter((r) => !(Date.parse(r.ts) < at - ROWS_KEPT_MS)));
+				}
+				offset += chunk.length;
+				ino = st.ino;
+				if (head.length < USAGE_HEAD_BYTES) head = nowHead.slice(0, Math.min(offset, USAGE_HEAD_BYTES));
+				mtimeMs = st.mtimeMs;
 			} finally {
 				closeSync(fd);
 			}
-			mtimeMs = st.mtimeMs;
 		} catch {
-			rows = [];
-			offset = 0;
+			reset();
 		}
 		return rows;
 	};
@@ -7533,15 +7659,18 @@ function cachedPoolCarry() {
 /** The whole request body as text. */
 function readBody(req) {
 	return new Promise((resolve, reject) => {
-		let data = "";
+		const chunks = [];
+		let size = 0;
 		req.on("data", (c) => {
-			data += c;
-			if (data.length > 64e3) {
+			const b = typeof c === "string" ? Buffer.from(c) : c;
+			chunks.push(b);
+			size += b.length;
+			if (size > 64e3) {
 				reject(/* @__PURE__ */ new Error("body too large"));
 				req.destroy();
 			}
 		});
-		req.on("end", () => resolve(data));
+		req.on("end", () => resolve(Buffer.concat(chunks).toString("utf8")));
 		req.on("error", reject);
 	});
 }
@@ -7724,7 +7853,7 @@ function cockpitBuildId() {
 * unless router.bind opens a second listener (then every cockpit route,
 * writes included, is token-gated — see src/router/cockpit-auth.ts).
 */
-function handleCockpit(req, res, limits, readRows = readUsageCached, sessionMap = loadSessionMap, system, workerWindows = loadWorkerWindows, readPool = cachedPoolCarry(), readPoolPanel = sharedPoolPanel, queueLogDir = workersLogDir) {
+function handleCockpit(req, res, limits, readRows = cachedRows(), sessionMap = loadSessionMap, system, workerWindows = loadWorkerWindows, readPool = cachedPoolCarry(), readPoolPanel = sharedPoolPanel, queueLogDir = workersLogDir) {
 	const url = new URL(req.url ?? "/", "http://localhost");
 	if (url.pathname !== "/cockpit" && !url.pathname.startsWith("/cockpit/")) return false;
 	const send = (status, type, body) => {
@@ -8599,6 +8728,12 @@ function writeResults(suite, rows) {
 	writeFileSync(file, `# A/B results\n\nGenerated by \`pai worker bench\`; void rows excluded from pass/total.\n\n${aggregate(rows)}\n`);
 	return file;
 }
+/** A throttle setting must be a positive number: NaN would disable the gate, 0 would wait forever. */
+function positiveNumber(flag, v) {
+	const n = typeof v === "number" ? v : typeof v === "string" && v.trim() !== "" ? Number(v) : NaN;
+	if (!Number.isFinite(n) || n <= 0) throw new Error(`${flag} must be a positive number, got "${String(v)}"`);
+	return n;
+}
 
 //#endregion
 //#region src/cli/lib/settings-manager.ts
@@ -9040,7 +9175,7 @@ function collectEnv(pairs) {
 	return o;
 }
 function parseCostTier(v) {
-	const n = parseInt(v, 10);
+	const n = /^\s*\d+\s*$/.test(v) ? Number(v) : NaN;
 	if (!Number.isInteger(n) || n < 1 || n > 5) throw new WorkersConfigError(`cost tier must be an integer 1 (cheapest) … 5 (most expensive), got "${v}"`);
 	return n;
 }
@@ -9138,7 +9273,7 @@ const providersAddCommand = {
 			name: "key",
 			type: "string",
 			value: "TOKEN",
-			summary: "API token, written into workers.yaml; use instead of --key-file"
+			summary: "API token, written into workers.yaml; visible in ps and shell history, prefer --key-file"
 		},
 		{
 			name: "key-file",
@@ -9218,6 +9353,7 @@ const providersAddCommand = {
 			const baseUrl = o["base-url"] ?? o.url;
 			if (protocol !== "openai" && !baseUrl) throw new WorkersConfigError("--base-url (or --url) is required (only --protocol openai goes without it)");
 			if (o.key && o["key-file"]) throw new WorkersConfigError("--key and --key-file are alternatives — pass one, not both");
+			if (o.key) out(warn("--key puts the token in the process list and shell history; prefer --key-file"));
 			const costTier = o["cost-tier"] !== void 0 ? parseCostTier(String(o["cost-tier"])) : void 0;
 			const tags = collectTags(ctx.multi?.tags);
 			addProvider({
@@ -9859,6 +9995,16 @@ function launchDetached(plan) {
 	closeSync(fd);
 	return Object.assign(() => child.exitCode, child.pid !== void 0 ? { pid: child.pid } : {});
 }
+/** SIGTERM a launched server's process group (it is a group leader), else the pid. */
+function killGroup(pid) {
+	try {
+		process.kill(-pid, "SIGTERM");
+	} catch {
+		try {
+			process.kill(pid, "SIGTERM");
+		} catch {}
+	}
+}
 /** Live servers of catalogue models, one per (model, port). */
 function liveServers(cfg) {
 	return instancesFromProcesses(psTable("pid=,rss=,command="), cfg, scriptListeners(cfg));
@@ -9875,6 +10021,36 @@ function currentInstances(cfg, deps = {}) {
 * instances keep running; a short memory or a full model (maxInstances) refuses.
 */
 async function startInstance(cfg, model, deps = {}) {
+	return withStartLock(deps.startDir ?? logDir(), model, (release) => startInstanceLocked(cfg, model, deps, release));
+}
+/** One start per model at a time: the lock covers read-check-allocate-write and is released once the row is written. */
+async function withStartLock(dir, model, fn) {
+	mkdirSync(dir, { recursive: true });
+	const m = join(dir, `start-${model.replace(/[^\w.-]/g, "_")}.lock`);
+	const t0 = Date.now();
+	for (;;) try {
+		mkdirSync(m);
+		break;
+	} catch {
+		breakStaleMutex(m, 6e4);
+		if (Date.now() - t0 > 12e4) throw new LocalTierError(`${model}: another start is in progress`);
+		await sleep(10 + Math.random() * 20);
+	}
+	let released = false;
+	const release = () => {
+		if (released) return;
+		released = true;
+		try {
+			rmdirSync(m);
+		} catch {}
+	};
+	try {
+		return await fn(release);
+	} finally {
+		release();
+	}
+}
+async function startInstanceLocked(cfg, model, deps, release) {
 	const tier = cfg.tiers[model];
 	if (!tier) throw new LocalTierError(`no model "${model}" (configured: ${Object.keys(cfg.tiers).join(", ") || "none"})`);
 	const dir = deps.startDir ?? logDir();
@@ -9925,10 +10101,12 @@ async function startInstance(cfg, model, deps = {}) {
 	});
 	writeInstances(dir, [...rows.filter((r) => r.id !== id), instance]);
 	const exited = (deps.launch ?? launchDetached)(plan);
+	release();
 	out(dim(`starting ${tier.label} #${n} (${tier.provider}) on :${port}, log ${plan.logPath}`));
 	const up = deps.up ?? ((p) => healthOk(tier, p));
 	const deadline = Date.now() + plan.readyMs;
 	const failWith = (error, msg) => {
+		if (exited.pid !== void 0 && exited() === null) killGroup(exited.pid);
 		failLocalStart(dir, error, id);
 		writeInstances(dir, readInstances(dir).map((r) => r.id === id ? {
 			...r,
@@ -10086,6 +10264,25 @@ async function stopInstance(cfg, target, opts = {}, deps = {}) {
 	}
 	return picked;
 }
+/**
+* Requests in flight per instance id, from the snapshot the router mirrors to its log dir (the scheduler lives in
+* the router process, not here). A missing or unreadable snapshot cannot be checked: the lookup throws so
+* the stop is refused unless confirmed.
+*/
+async function cliInflight(cfg, deps = {}) {
+	const [{ engineInflight }, { readSchedFile }, { instanceEngineKey }] = await Promise.all([
+		import("./cockpit-actions-BWunx3rY.mjs"),
+		import("./scheduler-dmHXoRvm.mjs"),
+		import("./local-instances-ZeRkkuS5.mjs")
+	]);
+	const now = deps.now ?? Date.now();
+	const snap = readSchedFile(deps.startDir ?? logDir(), now);
+	const keys = new Map(currentInstances(cfg, deps).map((r) => [r.id, instanceEngineKey(r)]));
+	return (id) => {
+		if (!snap) throw new LocalTierError(`${id}: in-flight requests could not be checked (no router snapshot); confirm to stop anyway (--yes)`);
+		return engineInflight(snap, keys.get(id) ?? id);
+	};
+}
 /** Stop every local server PAI knows (the instances, then anything left on a catalogue port), then wait until the ports are free. */
 async function stopAll(cfg) {
 	for (const m of new Set(currentInstances(cfg).map((r) => r.model))) await stopInstance(cfg, m, { all: true }, { confirm: true });
@@ -10194,7 +10391,10 @@ const workerLocalCommand = {
 				out(ok(`stopped ${(await stopInstance(cfg, name, {
 					count,
 					all: !!ctx.opts.all
-				}, { confirm: !!ctx.opts.yes })).map((d) => d.id).join(", ")}`));
+				}, {
+					confirm: !!ctx.opts.yes,
+					inflight: await cliInflight(cfg)
+				})).map((d) => d.id).join(", ")}`));
 				return;
 			}
 			const tier = cfg.tiers[name];
@@ -10293,6 +10493,8 @@ const workerBenchCommand = {
 			const tasks = o.tasks ? String(o.tasks).split(",") : all;
 			const unknown = tasks.filter((t) => !all.includes(t));
 			if (unknown.length) throw new Error(`unknown task(s): ${unknown.join(", ")}`);
+			const maxParallel = positiveNumber("--max-parallel", o["max-parallel"]);
+			const maxLoad = positiveNumber("--max-load", o["max-load"]);
 			const arms = planArms(tasks, String(o.providers).split(","), prompt, Math.max(1, Number(o.repeat) || 1));
 			out(`${arms.length} arm(s), prompt ${prompt}, max ${o["max-parallel"]} parallel, load gate ${o["max-load"]}`);
 			if (ctx.dryRun) {
@@ -10302,8 +10504,8 @@ const workerBenchCommand = {
 			const rows = await runBench({
 				suite,
 				arms,
-				maxParallel: Number(o["max-parallel"]),
-				maxLoad: Number(o["max-load"]),
+				maxParallel,
+				maxLoad,
 				log: (s) => out(s)
 			});
 			out(aggregate(rows));
@@ -10489,7 +10691,7 @@ const workerPoolCommand = {
 	async run(ctx) {
 		try {
 			const action = ctx.args.ACTION;
-			const { poolOn, poolOff, poolStatusText } = await import("./pool-cli-CbJfl37y.mjs");
+			const { poolOn, poolOff, poolStatusText } = await import("./pool-cli-DkZTAlSW.mjs");
 			if (action === void 0 || action === "status") return void out(await poolStatusText(logDir()));
 			if (action === "on") return void (await poolOn(logDir())).forEach((l) => out(l));
 			if (action === "off") return void (await poolOff(logDir())).forEach((l) => out(l));
@@ -10665,5 +10867,5 @@ const WORKER_CONFIG_COMMANDS = [
 ];
 
 //#endregion
-export { createEndpointHealth as $, workerOffCommand as A, allocatePort as B, workerClassesCommand as C, workerInstallCommand as D, workerFallbackCommand as E, workerProxyCommand as F, cachedSessionMap as G, sizeOnDiskGb as H, workerRolesCommand as I, setActionHooks as J, cachedWorkerWindows as K, workerRouteCommand as L, workerPolicyCommand as M, workerPoolCommand as N, workerLocalCommand as O, workerProvidersCommand as P, exportDefaultLayout as Q, installWorkers as R, workerBenchCommand as S, workerControlsCommand as T, SystemMonitor as U, canBind as V, cachedRows as W, setRuntimeSampler as X, setEndpointHealth as Y, warmCockpit as Z, serverLine as _, collectTags as a, hashText as at, stopAll as b, fail as c, killInstance as d, loadSessionMap as et, launchDetached as f, portOpen as g, portListeners as h, collectEnv as i, defaultRegistryCtx as it, workerOnCommand as j, workerMcpCommand as k, healthOk as l, parseCostTier as m, WORKER_CONFIG_COMMANDS as n, sessionDisplayName as nt, controlsArgv as o, saveRegistry as ot, listServers as p, handleCockpit as q, buildClassTarget as r, withWorkerAttribution as rt, currentInstances as s, providersCommand as st, PROVIDER_EDIT_COMMANDS as t, loadWorkerWindows as tt, isUp as u, startInstance as v, workerConfigCommand as w, stopInstance as x, startTier as y, mergeSettings as z };
-//# sourceMappingURL=worker-config-BLqa1yiD.mjs.map
+export { exportDefaultLayout as $, workerMcpCommand as A, mergeSettings as B, workerBenchCommand as C, workerFallbackCommand as D, workerControlsCommand as E, workerProvidersCommand as F, cachedRows as G, canBind as H, workerProxyCommand as I, handleCockpit as J, cachedSessionMap as K, workerRolesCommand as L, workerOnCommand as M, workerPolicyCommand as N, workerInstallCommand as O, workerPoolCommand as P, warmCockpit as Q, workerRouteCommand as R, stopInstance as S, workerConfigCommand as T, sizeOnDiskGb as U, allocatePort as V, SystemMonitor as W, setEndpointHealth as X, setActionHooks as Y, setRuntimeSampler as Z, portOpen as _, collectEnv as a, defaultRegistryCtx as at, startTier as b, currentInstances as c, providersCommand as ct, isUp as d, createEndpointHealth as et, killInstance as f, portListeners as g, parseCostTier as h, cliInflight as i, withWorkerAttribution as it, workerOffCommand as j, workerLocalCommand as k, fail as l, listServers as m, WORKER_CONFIG_COMMANDS as n, loadWorkerWindows as nt, collectTags as o, hashText as ot, launchDetached as p, cachedWorkerWindows as q, buildClassTarget as r, sessionDisplayName as rt, controlsArgv as s, saveRegistry as st, PROVIDER_EDIT_COMMANDS as t, loadSessionMap as tt, healthOk as u, serverLine as v, workerClassesCommand as w, stopAll as x, startInstance as y, installWorkers as z };
+//# sourceMappingURL=worker-config-BDYNxbYz.mjs.map