@juspay/neurolink 10.12.0 → 10.12.2

package/dist/server/routes/claudeProxyRoutes.js

@@ -110,6 +110,9 @@ function fetchAnthropicUpstream(url, init) {
     });
 }
 const accountRuntimeState = new Map();
+/** Persisted quota is advisory after a restart. Older snapshots cannot reject
+ * an account or create a new cooldown because the provider may have reset it. */
+const QUOTA_SNAPSHOT_FRESHNESS_MS = 15 * 60 * 1000;
 /** Shared across requests so a concurrent burst gets at most two retries for
  *  the account/window, rather than every request starting its own retry chain. */
 const transientRateLimitRetryBudgets = new Map();
@@ -603,8 +606,9 @@ function reconcileCooldownFromQuota(state, quota, now) {
  * proxy restart: all accounts tie, selection falls back to token-store
  * enumeration order, and the first account served becomes self-reinforcing
  * (it alone has data) — starving the others regardless of their resets.
- * Never overwrites fresher in-memory quota; stale disk snapshots degrade
- * gracefully because past reset timestamps are ignored by resetEpochToMs.
+ * Never overwrites fresher in-memory quota. Persisted quota cannot create or
+ * clear a cooldown: only an existing cooldown or fresh upstream response
+ * headers can change admission state.
  */
 async function seedRuntimeQuotasFromDisk(accounts) {
     try {
@@ -625,15 +629,6 @@ async function seedRuntimeQuotasFromDisk(accounts) {
                 state.coolingUntil = persistedCooldown.coolingUntil;
                 state.coolingReason = persistedCooldown.reason;
             }
-            if (state.quota) {
-                const cooldownUpdate = reconcileCooldownFromQuota(state, state.quota, now);
-                if (cooldownUpdate?.kind === "cooled") {
-                    await saveAccountCooldown(account.key, cooldownUpdate.coolingUntil, cooldownUpdate.coolingReason);
-                }
-                else if (cooldownUpdate?.kind === "cleared") {
-                    await clearAccountCooldown(account.key, cooldownUpdate.coolingUntil);
-                }
-            }
         }
     }
     catch {
@@ -802,47 +797,59 @@ function getSessionResetToleranceMs() {
 function accountSortMetrics(accountKey, now, sessionSoftLimit, sessionResetToleranceMs) {
     const st = accountRuntimeState.get(accountKey);
     const q = st?.quota;
+    const quotaLastUpdated = q && Number.isFinite(q.lastUpdated) ? q.lastUpdated : null;
+    const quotaAgeMs = quotaLastUpdated === null ? null : Math.max(0, now - quotaLastUpdated);
+    const quotaStale = quotaAgeMs !== null && quotaAgeMs > QUOTA_SNAPSHOT_FRESHNESS_MS;
+    const routingQuota = quotaStale ? undefined : q;
     const coolingActive = !!st?.coolingUntil && now < st.coolingUntil;
     // resetEpochToMs returns undefined for absent OR passed resets, so a
     // ticking window is exactly "reset !== undefined".
-    const weeklyReset = resetEpochToMs(q?.weeklyResetAt, now);
-    const sessionReset = resetEpochToMs(q?.sessionResetAt, now);
+    const weeklyReset = resetEpochToMs(routingQuota?.weeklyResetAt, now);
+    const sessionReset = resetEpochToMs(routingQuota?.sessionResetAt, now);
     const sessionTicking = sessionReset !== undefined;
     const weeklyTicking = weeklyReset !== undefined;
-    const sessionUsed = q ? (sessionTicking ? (q.sessionUsed ?? 0) : 0) : null;
-    const weeklyUsed = q ? (weeklyTicking ? (q.weeklyUsed ?? null) : 0) : null;
-    const sessionStatus = q
+    const sessionUsed = routingQuota
         ? sessionTicking
-            ? (q.sessionStatus ?? "unknown")
+            ? (routingQuota.sessionUsed ?? 0)
+            : 0
+        : null;
+    const weeklyUsed = routingQuota
+        ? weeklyTicking
+            ? (routingQuota.weeklyUsed ?? null)
+            : 0
+        : null;
+    const sessionStatus = routingQuota
+        ? sessionTicking
+            ? (routingQuota.sessionStatus ?? "unknown")
             : "allowed"
         : null;
-    const weeklyStatus = q
+    const weeklyStatus = routingQuota
         ? weeklyTicking
-            ? (q.weeklyStatus ?? "unknown")
+            ? (routingQuota.weeklyStatus ?? "unknown")
             : "allowed"
         : null;
-    const overageEligible = isQuotaOverageAvailable(q);
+    const overageEligible = isQuotaOverageAvailable(routingQuota);
     const saturated = sessionStatus === "throttled" ||
         (sessionTicking && (sessionUsed ?? 0) >= sessionSoftLimit);
-    const quotaLastUpdated = q && Number.isFinite(q.lastUpdated) ? q.lastUpdated : null;
     return {
         usable: !coolingActive &&
             weeklyStatus !== "rejected" &&
             (sessionStatus !== "rejected" || overageEligible) &&
-            (q?.unifiedStatus?.trim().toLowerCase() !== "rejected" ||
+            (routingQuota?.unifiedStatus?.trim().toLowerCase() !== "rejected" ||
                 overageEligible),
-        saturated,
-        hasQuota: !!q,
+        saturated: !quotaStale && saturated,
+        hasQuota: !!routingQuota,
+        quotaStale,
         quotaLastUpdated,
-        quotaAgeMs: quotaLastUpdated === null ? null : Math.max(0, now - quotaLastUpdated),
+        quotaAgeMs,
         coolingActive,
         coolingReason: st?.coolingReason ?? null,
         coolingUntil: st?.coolingUntil ?? 0,
-        unifiedStatus: q?.unifiedStatus ?? null,
-        fallbackStatus: q?.fallbackStatus ?? null,
-        upgradePaths: q?.upgradePaths ?? null,
+        unifiedStatus: routingQuota?.unifiedStatus ?? null,
+        fallbackStatus: routingQuota?.fallbackStatus ?? null,
+        upgradePaths: routingQuota?.upgradePaths ?? null,
         overageEligible,
-        overageStatus: q?.overageStatus ?? null,
+        overageStatus: routingQuota?.overageStatus ?? null,
         sessionStatus,
         sessionUsed,
         sessionResetBucket: sessionTicking
@@ -960,6 +967,7 @@ function buildRoutingDecision(args) {
             usable: metrics.usable,
             saturated: metrics.saturated,
             quotaObserved: metrics.hasQuota,
+            quotaStale: metrics.quotaStale,
             quotaLastUpdated: metrics.quotaLastUpdated,
             quotaAgeMs: metrics.quotaAgeMs,
             coolingActive: metrics.coolingActive,

package/dist/types/proxy.d.ts

@@ -439,6 +439,7 @@ export type ProxyAccountRoutingCandidate = {
     usable: boolean;
     saturated: boolean;
     quotaObserved: boolean;
+    quotaStale: boolean;
     quotaLastUpdated: number | null;
     quotaAgeMs: number | null;
     coolingActive: boolean;
@@ -477,6 +478,7 @@ export type ProxyAccountSortMetrics = {
     usable: boolean;
     saturated: boolean;
     hasQuota: boolean;
+    quotaStale: boolean;
     quotaLastUpdated: number | null;
     quotaAgeMs: number | null;
     coolingActive: boolean;
@@ -941,6 +943,8 @@ export type AccountQuota = {
     upgradePaths?: string;
     /** "allowed" | "rejected" */
     overageStatus: string;
+    /** Whether Anthropic reports that paid overage is actively serving traffic. */
+    overageInUse?: boolean;
     /** Epoch ms when we last captured this data */
     lastUpdated: number;
     /** Dynamic per-plan limit buckets from the usage API `limits[]` array

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@juspay/neurolink",
-  "version": "10.12.0",
+  "version": "10.12.2",
   "packageManager": "pnpm@10.15.1",
   "description": "TypeScript AI SDK with 24+ LLM providers behind one consistent API. MCP-native (connect any MCP server), voice TTS/STT/realtime, RAG, agents, memory, context compaction. OpenAI · Anthropic · Gemini · Bedrock · Azure · Ollama · DeepSeek · NVIDIA NIM and more.",
   "author": {