feat(rate-limiting): implement per-user rate limiting for assistant m…

packages/app/api/lib/route-message.ts

@@ -19,6 +19,33 @@ import type { AgentLlmConfig } from './clawscale-agent.js';
 import { parseCommand, resolveTarget, resolveAddRemoveArg, formatCommandHelp } from './slash-commands.js';
 import type { AiBackendType, AiBackendProviderConfig } from '../../shared/index.js';
 
+// ── Per-user rate limiter (in-memory) ────────────────────────────────────────
+
+interface RateBucket {
+  count: number;
+  resetAt: number;
+}
+
+/** Key: `${tenantId}:${endUserId}` */
+const rateBuckets = new Map<string, RateBucket>();
+
+function isRateLimited(
+  tenantId: string,
+  endUserId: string,
+  limit: { maxMessages: number; windowSeconds: number },
+): boolean {
+  if (limit.maxMessages <= 0) return false; // 0 = unlimited
+  const key = `${tenantId}:${endUserId}`;
+  const now = Date.now();
+  const bucket = rateBuckets.get(key);
+  if (!bucket || now >= bucket.resetAt) {
+    rateBuckets.set(key, { count: 1, resetAt: now + limit.windowSeconds * 1000 });
+    return false;
+  }
+  bucket.count++;
+  return bucket.count > limit.maxMessages;
+}
+
 export interface Attachment {
   url: string;
   filename: string;
@@ -68,7 +95,7 @@ export async function routeInboundMessage(input: InboundMessage): Promise<RouteR
     personaName?: string;
     endUserAccess?: 'anonymous' | 'whitelist' | 'blacklist';
     allowList?: string[];
-    clawscale?: { name?: string; answerStyle?: string; isActive?: boolean; llm?: AgentLlmConfig };
+    clawscale?: { name?: string; answerStyle?: string; isActive?: boolean; rateLimit?: { maxMessages: number; windowSeconds: number }; llm?: AgentLlmConfig };
     blockList?: string[];
   };
 
@@ -130,6 +157,7 @@ export async function routeInboundMessage(input: InboundMessage): Promise<RouteR
   const clawscaleStyle = clawscaleCfg.answerStyle;
   const clawscaleActive = clawscaleCfg.isActive !== false;
   const clawscaleLlm = clawscaleCfg.llm ?? { model: 'openai:gpt-5.4-mini' };
+  const clawscaleRateLimit = clawscaleCfg.rateLimit;
 
   const allBackends = await db.aiBackend.findMany({
     where: { tenantId, isActive: true },
@@ -145,6 +173,14 @@ export async function routeInboundMessage(input: InboundMessage): Promise<RouteR
    * Slash commands are executed via a callback that re-enters routeInboundMessage.
    */
   async function runAgent(userText: string, mode: 'select' | 'direct'): Promise<RouteResult> {
+    // Rate limit check
+    if (clawscaleRateLimit && isRateLimited(tenantId, endUser!.id, clawscaleRateLimit)) {
+      const mins = Math.ceil(clawscaleRateLimit.windowSeconds / 60);
+      return reply(
+        `⏳ You've reached the message limit (${clawscaleRateLimit.maxMessages} per ${mins === 1 ? 'minute' : `${mins} minutes`}). Please wait a moment before sending another message.`,
+      );
+    }
+
     // If attachments are present but multimodal is not enabled, nudge the admin
     if (attachments?.length && !clawscaleLlm?.multimodal) {
       return reply(

packages/app/api/routes/tenant.ts

@@ -16,6 +16,10 @@ const updateSettingsSchema = z.object({
         name:        z.string().min(1).max(80).optional(),
         answerStyle: z.string().max(500).optional(),
         isActive:    z.boolean().optional(),
+        rateLimit: z.object({
+          maxMessages: z.number().int().min(0).max(10000),
+          windowSeconds: z.number().int().min(1).max(86400),
+        }).nullable().optional(),
         llm: z.object({
           model: z.string().min(1).max(100),
           apiKey: z.string().max(500).optional(),

packages/app/shared/index.ts

@@ -19,6 +19,13 @@ export interface ClawScaleAgentSettings {
   answerStyle?: string;
   /** Whether the orchestrator responds at all (default: true) */
   isActive?: boolean;
+  /** Per-user rate limiting for the assistant */
+  rateLimit?: {
+    /** Maximum messages allowed per window (0 = unlimited) */
+    maxMessages: number;
+    /** Window duration in seconds (default: 60) */
+    windowSeconds: number;
+  };
   /** LLM configuration for the ClawScale agent */
   llm?: {
     /** LangChain model string, e.g. "openai:gpt-5.4-mini", "anthropic:claude-haiku-4-5-20251001" */

packages/app/web/app/(dashboard)/settings/page.tsx

@@ -23,6 +23,9 @@ export default function Settings() {
   const [clawscaleApiKey, setClawscaleApiKey] = useState('');
   const [apiKeySet, setApiKeySet] = useState(false);
   const [clawscaleMultimodal, setClawscaleMultimodal] = useState(false);
+  const [rateLimitEnabled, setRateLimitEnabled] = useState(false);
+  const [rateLimitMax, setRateLimitMax] = useState(20);
+  const [rateLimitWindow, setRateLimitWindow] = useState(60);
 
   useEffect(() => {
     api.get<ApiResponse<Tenant>>('/api/tenant').then((res) => {
@@ -34,6 +37,8 @@ export default function Settings() {
         setClawscaleModel(s.clawscale?.llm?.model ?? 'openai:gpt-5.4-mini');
         setApiKeySet(!!s.clawscale?.llm?.apiKey && s.clawscale.llm.apiKey !== '');
         setClawscaleMultimodal(s.clawscale?.llm?.multimodal ?? false);
+        const rl = s.clawscale?.rateLimit;
+        if (rl) { setRateLimitEnabled(true); setRateLimitMax(rl.maxMessages); setRateLimitWindow(rl.windowSeconds); }
       }
       setLoading(false);
     });
@@ -52,6 +57,7 @@ export default function Settings() {
               ...(clawscaleApiKey ? { apiKey: clawscaleApiKey } : {}),
               multimodal: clawscaleMultimodal,
             },
+            rateLimit: rateLimitEnabled ? { maxMessages: rateLimitMax, windowSeconds: rateLimitWindow } : null,
           },
         },
       });
@@ -108,6 +114,25 @@ export default function Settings() {
                 <span className="text-xs text-gray-500 block">Allow the assistant to process images, files, and audio sent by users. Requires a vision-capable model (e.g. GPT-4o, Claude Sonnet).</span>
               </span>
             </label>
+            <label className="flex items-start gap-3 cursor-pointer">
+              <input type="checkbox" checked={rateLimitEnabled} onChange={(e) => setRateLimitEnabled(e.target.checked)} disabled={!isAdmin} className="mt-0.5" />
+              <span>
+                <span className="text-sm font-medium text-gray-900">Rate limit</span>
+                <span className="text-xs text-gray-500 block">Limit how many messages each end-user can send within a time window.</span>
+              </span>
+            </label>
+            {rateLimitEnabled && (
+              <div className="flex items-center gap-3 pl-7">
+                <div className="flex-1">
+                  <label className="label">Max messages</label>
+                  <input type="number" className="input" min={1} max={10000} value={rateLimitMax} onChange={(e) => setRateLimitMax(Number(e.target.value))} disabled={!isAdmin} />
+                </div>
+                <div className="flex-1">
+                  <label className="label">Window (seconds)</label>
+                  <input type="number" className="input" min={1} max={86400} value={rateLimitWindow} onChange={(e) => setRateLimitWindow(Number(e.target.value))} disabled={!isAdmin} />
+                </div>
+              </div>
+            )}
           </div>
         </div>