moar fixes

src/shared/users/user-store.ts

@@ -381,24 +381,31 @@ export function hasAvailableQuota({
 
 /**
  * For the given user, refreshes token limits for each model family. The new limit
- * is set to the configured quota value (either from user's tokenRefresh or global config).
- * This replaces the current limits entirely, not adding to them.
+ * is set to the current usage + the refresh amount, ensuring users get their full
+ * refresh allocation regardless of current usage.
  */
 export function refreshQuota(token: string) {
   const user = users.get(token);
   if (!user) return;
-  const { tokenLimits, tokenRefresh } = user;
+  const { tokenCounts, tokenLimits, tokenRefresh } = user;
 
   for (const family of MODEL_FAMILIES) {
-    // Get the quota value to set (from user refresh config or global default)
+    // Get the quota value to add (from user refresh config or global default)
     const userQuota = tokenRefresh[family] ?? 0;
     const globalQuota = config.tokenQuota[family] ?? 0;
-    
-    const quotaToSet = userQuota || globalQuota;
+    const quotaToAdd = userQuota || globalQuota;
 
-    // Only update if we have a valid quota
-    if (quotaToSet > 0) {
-      tokenLimits[family] = quotaToSet;
+    if (quotaToAdd > 0) {
+      // Calculate current usage including legacy
+      const currentUsage = tokenCounts[family] ?? { input: 0, output: 0 };
+      const input = Math.max(0, currentUsage.input ?? 0);
+      const output = Math.max(0, currentUsage.output ?? 0);
+      const legacy = Math.max(0, currentUsage.legacy_total ?? 0);
+      const totalUsage = input + output + legacy;
+      
+      // Set new limit to current usage + refresh amount
+      // This ensures users always get their full refresh allocation
+      tokenLimits[family] = totalUsage + quotaToAdd;
     }
   }
   usersToFlush.add(token);