adaptive-memory-multi-model-router 2.11.0 → 2.12.1

package/demo.tape

@@ -0,0 +1,259 @@
+# A3M Router VHS Demo - 90 Second Terminal Demo
+# Target: 60-90 seconds
+# Output: a3m-demo.gif
+
+# =============================================================================
+# SETUP
+# =============================================================================
+
+Output demo.gif
+Set FontSize 16
+Set Width 1200
+Set Height 675
+Set TypingSpeed 30ms
+Set Shell bash
+Set FontFamily "JetBrains Mono"
+
+# Theme - dark terminal
+Set ThemeFlavor catppuccin-mocha
+
+# =============================================================================
+# SCENE 1: THE PROBLEM (0-15s)
+# =============================================================================
+
+Type "#!/bin/bash"
+Enter
+
+Sleep 500ms
+
+Type "# Your LLM app today — costs $0.015 per request, no fallback"
+Enter
+
+Type "cat << 'EOF'"
+Enter
+Sleep 100ms
+Type "// Without A3M: you manage everything manually"
+Enter
+Type "const response = await openai.chat.completions.create({"
+Enter
+Type "  model: \"gpt-4o\",  // Always expensive"
+Enter
+Type "  messages: [{ role: \"user\", content: \"Explain quantum entanglement\" }]"
+Enter
+Type "});"
+Enter
+Type "// If OpenAI goes down → your app crashes"
+Enter
+Type "// You pay $0.015 per request forever"
+Enter
+Type "EOF"
+Enter
+
+Sleep 1s
+
+# Red X marks
+Type "echo \"❌ Problem: $0.015/req, no fallback, manual management\""
+Enter
+
+Sleep 2s
+
+# =============================================================================
+# SCENE 2: WITH A3M - ONE LINE CHANGE (15-35s)
+# =============================================================================
+
+Ctrl+L
+Sleep 500ms
+
+Type "# With A3M: one line change, infinite benefits"
+Enter
+
+Type "cat << 'EOF'"
+Enter
+Sleep 100ms
+Type "// Same request with A3M router"
+Enter
+Type "const response = await fetch(\"http://localhost:8787/v1/chat/completions\", {"
+Enter
+Type "  method: \"POST\","
+Enter
+Type "  headers: { \"Content-Type\": \"application/json\" },"
+Enter
+Type "  body: JSON.stringify({"
+Enter
+Type "    model: \"auto\",  // ← A3M picks the best model"
+Enter
+Type "    messages: [{ role: \"user\", content: \"Explain quantum entanglement\" }]"
+Enter
+Type "  })"
+Enter
+Type "});"
+Enter
+Type "EOF"
+Enter
+
+Sleep 1s
+
+# Show routing decision
+Enter
+Type "# A3M analyzes and decides:"
+Enter
+Sleep 200ms
+Type "echo \"📊 Task: explanation → Complexity: simple → Budget: minimize\""
+Enter
+Sleep 500ms
+Type "echo \"\""
+Enter
+Sleep 200ms
+Type "echo \"✅ Routed to: Groq (FREE, 847ms)\""
+Enter
+Sleep 500ms
+Type "echo \"   Cost: $0.00 | Quality: 94%\""
+Enter
+
+Sleep 2s
+
+# =============================================================================
+# SCENE 3: FAILURE RECOVERY (35-55s)
+# =============================================================================
+
+Ctrl+L
+Sleep 500ms
+
+Type "# But what happens when a provider fails?"
+Enter
+Sleep 1s
+
+Type "# A3M gracefully handles failures..."
+Enter
+
+Sleep 1s
+
+# Simulate the failure scenario
+Type "echo \"━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\""
+Enter
+Sleep 300ms
+Type "echo \"  [Groq] — Attempting connection...\""
+Enter
+Sleep 1s
+Type "echo \"  [Groq] — ✗ FAILED — 503 Service Unavailable\""
+Enter
+Sleep 500ms
+Type "echo \"  [Circuit Breaker] — Tripped after 3 failures\""
+Enter
+Sleep 500ms
+Type "echo \"  [DeepSeek] — HEALTHY — Switching...\""
+Enter
+Sleep 1s
+Type "echo \"\""
+Enter
+Sleep 300ms
+Type "echo \"✅ Response delivered via DeepSeek fallback\""
+Enter
+Sleep 500ms
+Type "echo \"   Your app: never knew there was a problem\""
+Enter
+Sleep 500ms
+Type "echo \"━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\""
+Enter
+
+Sleep 2s
+
+# =============================================================================
+# SCENE 4: COST SAVINGS (55-70s)
+# =============================================================================
+
+Ctrl+L
+Sleep 500ms
+
+Type "# Same query. Dramatically different costs."
+Enter
+Sleep 500ms
+
+Type "echo \"\""
+Enter
+Sleep 200ms
+Type "echo \"  WITHOUT A3M          WITH A3M\""
+Enter
+Sleep 200ms
+Type "echo \"  ─────────────────    ────────────────\""
+Enter
+Sleep 200ms
+Type "echo \"  GPT-4o @ $3.00/1K     Groq @ $0.00/1K\""
+Enter
+Sleep 200ms
+Type "echo \"  500 tokens = $0.0015  500 tokens = $0.00\""
+Enter
+Sleep 1s
+Type "echo \"\""
+Enter
+Sleep 200ms
+Type "echo \"💰 SAVINGS: 100% per query\""
+Enter
+Sleep 300ms
+Type "echo \"   At 1000 queries/day → $1.50 saved daily\""
+Enter
+Sleep 300ms
+Type "echo \"   At 1000 queries/day → $547 saved yearly\""
+Enter
+
+Sleep 2s
+
+# =============================================================================
+# SCENE 5: ONE COMMAND (70-90s)
+# =============================================================================
+
+Ctrl+L
+Sleep 500ms
+
+Type "echo \"\""
+Enter
+Sleep 200ms
+Type "echo \"╔═══════════════════════════════════════════════════╗\""
+Enter
+Sleep 100ms
+Type "echo \"║         Get started in 10 seconds:                ║\""
+Enter
+Sleep 100ms
+Type "echo \"╚═══════════════════════════════════════════════════╝\""
+Enter
+Sleep 500ms
+Type "echo \"\""
+Enter
+Sleep 200ms
+Type "echo \"  npm install -g adaptive-memory-multi-model-router\""
+Enter
+Sleep 300ms
+Type "echo \"\""
+Enter
+Sleep 200ms
+Type "echo \"  # Auto-detects your API keys, zero config\""
+Enter
+Sleep 300ms
+Type "echo \"  npx a3m-router serve\""
+Enter
+Sleep 500ms
+Type "echo \"\""
+Enter
+Sleep 200ms
+Type "echo \"  # Change: model: 'gpt-4o' → model: 'auto'\""
+Enter
+Sleep 500ms
+Type "echo \"\""
+Enter
+Sleep 300ms
+Type "echo \"  GitHub: github.com/Das-rebel/adaptive-memory-multi-model-router\""
+Enter
+Sleep 200ms
+Type "echo \"  npm:    npmjs.com/package/adaptive-memory-multi-model-router\""
+Enter
+
+Sleep 2s
+
+# Final pulse effect
+Type "echo \"\""
+Enter
+Sleep 200ms
+Type "echo \"✨ One prompt in. The right model out. ✨\""
+Enter
+
+Sleep 1s

package/dist/cli/setupWizard.js

@@ -0,0 +1,194 @@
+/**
+ * A3M Router Setup Wizard
+ * Interactive configuration wizard
+ */
+
+const fs = require('fs');
+const path = require('path');
+const readline = require('readline');
+
+const CONFIG_DIR = path.join(process.env.HOME || '/tmp', '.config', 'a3m-router');
+const CONFIG_FILE = path.join(CONFIG_DIR, 'providers.json');
+
+// API key environment variable mappings
+const API_KEY_ENV_MAP = {
+  'GROQ_API_KEY': 'groq',
+  'OPENAI_API_KEY': 'openai',
+  'ANTHROPIC_API_KEY': 'anthropic',
+  'DEEPSEEK_API_KEY': 'deepseek',
+  'MISTRAL_API_KEY': 'mistral',
+  'GOOGLE_API_KEY': 'google',
+  'CEREBRAS_API_KEY': 'cerebras',
+  'TOGETHER_API_KEY': 'together',
+  'AI21_API_KEY': 'ai21',
+  'COHERE_API_KEY': 'cohere',
+  'MINIMAX_API_KEY': 'minimax',
+  'KIMI_API_KEY': 'kimi',
+  'MOONSHOT_API_KEY': 'moonshot',
+  'QWEN_API_KEY': 'qwen',
+  'ZHIPU_API_KEY': 'zhipu',
+  'YI_API_KEY': 'yi',
+  'BAICHUAN_API_KEY': 'baichuan',
+};
+
+// Provider metadata
+const PROVIDER_INFO = {
+  groq: { name: 'Groq', models: 'llama-3.3-70b-versatile', tier: 'free', strength: 'Fast, free tier' },
+  openai: { name: 'OpenAI', models: 'gpt-4o-mini', tier: 'paid', strength: 'GPT-4, most capable' },
+  anthropic: { name: 'Anthropic', models: 'claude-3.5-haiku', tier: 'paid', strength: 'Claude, best reasoning' },
+  deepseek: { name: 'DeepSeek', models: 'deepseek-chat-v3', tier: 'cheap', strength: 'Cheap, good code' },
+  mistral: { name: 'Mistral', models: 'mistral-small-latest', tier: 'cheap', strength: 'European, balanced' },
+  google: { name: 'Google AI', models: 'gemini-1.5-flash', tier: 'free', strength: 'Gemini, multimodal' },
+  cerebras: { name: 'Cerebras', models: 'llama-3.3-70b', tier: 'free', strength: 'Fastest inference' },
+  together: { name: 'Together AI', models: 'Llama-3.3-70B-Instruct', tier: 'cheap', strength: 'Managed, reliable' },
+  ai21: { name: 'AI21', models: 'jamba-1.5-medium', tier: 'paid', strength: 'Jamba, long context' },
+  cohere: { name: 'Cohere', models: 'command-r7b', tier: 'cheap', strength: 'Command series, fast' },
+  minimax: { name: 'MiniMax', models: 'abab6.5s-chat', tier: 'cheap', strength: 'Chinese, cheap' },
+  kimi: { name: 'Kimi/Moonshot', models: 'moonshot-v1-8k', tier: 'cheap', strength: 'Chinese, 128k context' },
+  moonshot: { name: 'Moonshot', models: 'moonshot-v1-8k', tier: 'cheap', strength: 'Chinese, good' },
+  qwen: { name: 'Qwen', models: 'qwen-turbo', tier: 'cheap', strength: 'Alibaba, multilingual' },
+  zhipu: { name: 'Zhipu GLM', models: 'glm-4', tier: 'cheap', strength: 'Chinese, smart' },
+  yi: { name: 'Yi', models: 'yi-large', tier: 'cheap', strength: 'Chinese, good reasoning' },
+  baichuan: { name: 'Baichuan', models: 'baichuan-4', tier: 'cheap', strength: 'Chinese, balanced' },
+};
+
+function createInterface() {
+  return readline.createInterface({
+    input: process.stdin,
+    output: process.stdout
+  });
+}
+
+function question(rl, text) {
+  return new Promise((resolve) => {
+    rl.question(text, (answer) => resolve(answer));
+  });
+}
+
+async function detectApiKeys() {
+  const detected = [];
+  for (const [envVar, providerId] of Object.entries(API_KEY_ENV_MAP)) {
+    if (process.env[envVar]) {
+      detected.push({ envVar, providerId, info: PROVIDER_INFO[providerId] });
+    }
+  }
+  return detected;
+}
+
+async function runWizard() {
+  console.log('\n🔧 A3M Router Setup Wizard');
+  console.log('━━━━━━━━━━━━━━━━━━━━━━━━━\n');
+  
+  const rl = createInterface();
+  
+  // Ensure config directory exists
+  if (!fs.existsSync(CONFIG_DIR)) {
+    fs.mkdirSync(CONFIG_DIR, { recursive: true });
+  }
+  
+  // Check for existing config
+  let existingConfig = {};
+  if (fs.existsSync(CONFIG_FILE)) {
+    try {
+      existingConfig = JSON.parse(fs.readFileSync(CONFIG_FILE, 'utf8'));
+      console.log('✓ Found existing config at', CONFIG_FILE);
+      console.log('  Providers:', Object.keys(existingConfig.providers || {}).join(', '));
+      console.log('');
+    } catch (e) {
+      console.log('⚠ Could not read existing config, starting fresh\n');
+    }
+  }
+  
+  // Auto-detect API keys
+  console.log('🔍 Scanning for API keys in environment...');
+  const detected = await detectApiKeys();
+  
+  if (detected.length === 0) {
+    console.log('⚠ No API keys detected in environment.');
+    console.log('  Set any of: GROQ_API_KEY, OPENAI_API_KEY, DEEPSEEK_API_KEY, etc.\n');
+  } else {
+    console.log('✓ Found', detected.length, 'API key(s):');
+    detected.forEach(({ envVar, providerId, info }) => {
+      console.log('  ✓', envVar, '→', info?.name || providerId);
+    });
+    console.log('');
+  }
+  
+  // Provider selection
+  const allProviders = Object.keys(PROVIDER_INFO);
+  const selected = new Set();
+  
+  // Pre-select providers with detected keys
+  detected.forEach(({ providerId }) => selected.add(providerId));
+  
+  console.log('📡 Select providers to configure (comma-separated numbers, or "all"):');
+  console.log('');
+  
+  const numbered = allProviders.map((id, i) => ({ id, i }));
+  numbered.forEach(({ id, i }) => {
+    const info = PROVIDER_INFO[id];
+    const selected_mark = selected.has(id) ? '[x]' : '[ ]';
+    const tier_mark = info?.tier === 'free' ? '(FREE)' : info?.tier === 'cheap' ? '(cheap)' : '(paid)';
+    console.log(`  ${String(i + 1).padStart(2)}. ${selected_mark} ${id.padEnd(12)} ${tier_mark} - ${info?.strength || ''}`);
+  });
+  
+  console.log('');
+  const answer = await question(rl, '  Enter numbers or "all" [all with keys detected]: ');
+  
+  if (answer.toLowerCase().trim() === 'all') {
+    allProviders.forEach(id => selected.add(id));
+  } else if (answer.trim()) {
+    const nums = answer.split(',').map(s => parseInt(s.trim())).filter(n => !isNaN(n));
+    nums.forEach(n => {
+      const idx = n - 1;
+      if (idx >= 0 && idx < allProviders.length) {
+        selected.add(allProviders[idx]);
+      }
+    });
+  }
+  
+  console.log('\n✓ Selected providers:', Array.from(selected).join(', '));
+  
+  // Build config
+  const config = {
+    version: '1.0',
+    providers: {}
+  };
+  
+  selected.forEach(providerId => {
+    const info = PROVIDER_INFO[providerId];
+    const envKey = Object.entries(API_KEY_ENV_MAP).find(([k, v]) => v === providerId)?.[0];
+    
+    config.providers[providerId] = {
+      name: info?.name || providerId,
+      apiKey: envKey ? process.env[envKey] : '',
+      models: [info?.models || 'default'],
+      type: 'api',
+      enabled: true
+    };
+  });
+  
+  // Save config
+  fs.writeFileSync(CONFIG_FILE, JSON.stringify(config, null, 2));
+  console.log('\n✓ Config saved to', CONFIG_FILE);
+  
+  // Test connections
+  console.log('\n🧪 Testing connections...');
+  console.log('  (Skipped in wizard mode - run "npx a3m-router test" to verify)\n');
+  
+  // Ready message
+  console.log('━━━━━━━━━━━━━━━━━━━━━━━━━');
+  console.log('✅ A3M Router is ready!');
+  console.log('');
+  console.log('  Next steps:');
+  console.log('  1. npx a3m-router serve     # Start proxy server');
+  console.log('  2. npx a3m-router test       # Test provider connections');
+  console.log('  3. npx a3m-router route "hi" # Try routing a query');
+  console.log('');
+  console.log('  Docs: https://github.com/Das-rebel/adaptive-memory-multi-model-router');
+  console.log('');
+  
+  rl.close();
+}
+
+module.exports = { runWizard };

package/dist/cli/tui.js

@@ -0,0 +1,152 @@
+#!/usr/bin/env node
+/**
+ * A3M Router Quick TUI
+ * Zero-config experience - no API key needed for free tier
+ */
+
+const args = process.argv.slice(2);
+
+if (args.length === 0) {
+  console.log(`
+╔══════════════════════════════════════════════╗
+║         🔀 A3M Router - Zero Config          ║
+╠══════════════════════════════════════════════╣
+║                                              ║
+║  One prompt in. The right model out.         ║
+║                                              ║
+║  Usage:                                      ║
+║    a3m hello world      # Route a query       ║
+║    a3m --serve         # Start proxy          ║
+║    a3m --setup        # Configure API keys   ║
+║    a3m --demo         # Try without keys     ║
+║                                              ║
+╚══════════════════════════════════════════════╝
+`);
+  process.exit(0);
+}
+
+const command = args[0];
+
+// Demo mode - works without API key
+if (command === '--demo') {
+  console.log(`
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🎮  A3M Router Demo Mode
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Try these queries:
+  • "Write Python hello world"
+  • "Explain quantum physics"
+  • "Translate to Japanese: Hello"
+  • "Write a haiku about coding"
+`);
+  
+  const readline = require('readline');
+  const rl = readline.createInterface({ input: process.stdin, output: process.stdout });
+  
+  const demoResults = [
+    { provider: 'groq', model: 'llama-3.3-70b-versatile', tier: 'free', cost: '$0.00' },
+    { provider: 'deepseek', model: 'deepseek-chat-v3', tier: 'cheap', cost: '$0.00' },
+    { provider: 'kimi', model: 'moonshot-v1-8k', tier: 'cheap', cost: '$0.00' },
+    { provider: 'openai', model: 'gpt-4o-mini', tier: 'paid', cost: '$0.00' },
+  ];
+  
+  let count = 0;
+  function ask() {
+    rl.question('\n🔀 Query: ', (q) => {
+      if (!q || q.toLowerCase() === 'exit') {
+        console.log('\n👋 Ready for the real thing? Run: npm install adaptive-memory-multi-model-router\n');
+        rl.close();
+        return;
+      }
+      const r = demoResults[count++ % demoResults.length];
+      console.log(`\n  🏆 Best: ${r.provider} / ${r.model}`);
+      console.log(`  💰 Cost: ${r.cost} (${r.tier})`);
+      ask();
+    });
+  }
+  ask();
+  return;
+}
+
+// Serve mode
+if (command === '--serve') {
+  console.log(`
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+🚀  A3M Router Proxy Server
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Server: http://localhost:8787/v1/chat/completions
+API: OpenAI-compatible
+
+No API keys configured? Add free tier:
+  GROQ_API_KEY=your_key npx a3m-router serve
+
+Or run setup: a3m --setup
+`);
+  
+  try {
+    const server = require('./server/proxyServer.js');
+    const port = parseInt(args[1]) || 8787;
+    server.createProxyServer({ port });
+    console.log(`\n✅ Server running at http://localhost:${port}`);
+  } catch (e) {
+    console.log('\n⚠️  Run "npm install" first, then "npm run build"\n');
+  }
+  return;
+}
+
+// Setup mode
+if (command === '--setup') {
+  const wizard = require('./cli/setupWizard.js');
+  wizard.runWizard();
+  return;
+}
+
+// Zero-config mode - route with free tier defaults
+const query = args.join(' ');
+
+// Check for GROQ_API_KEY first (free tier)
+if (process.env.GROQ_API_KEY) {
+  console.log('\n🔀 Routing: "' + query + '"');
+  console.log('📡 Using free tier (Groq)...\n');
+  
+  // Quick route via groq
+  const { routeQuery } = require('./index.js');
+  routeQuery(query).then(r => {
+    console.log(`  🏆 Provider: ${r.provider || 'groq'}`);
+    console.log(`  🤖 Model: ${r.model || 'llama-3.3-70b-versatile'}`);
+    console.log(`  💰 Est: $${(r.estimated_cost || 0.0001).toFixed(6)}`);
+  }).catch(e => {
+    console.log('  ⚠️  Error:', e.message);
+    console.log('  Tip: Get free key at https://console.groq.com\n');
+  });
+  return;
+}
+
+// No API key - show helpful message
+console.log(`
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+⚠️  No API key detected
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+A3M Router needs at least one API key to route queries.
+
+🔥 FREE TIER OPTIONS:
+  1. Groq (fastest, 1000+ req/day free)
+     → https://console.groq.com/apikeys
+     → Set: export GROQ_API_KEY=your_key
+
+  2. Cerebras (fastest inference, free)
+     → https://cerebras.ai/labs
+     → Set: export CEREBRAS_API_KEY=your_key
+
+  3. Google AI (Gemini, free tier)
+     → https://aistudio.google.com/app/apikey
+     → Set: export GOOGLE_API_KEY=your_key
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Quick setup: npx a3m --setup
+Quick start: GROQ_API_KEY=your_key npx a3m-router serve
+`);

package/dist/cli.js

@@ -3,9 +3,10 @@
  * A3M Router CLI - Adaptive Memory Multi-Model Router
  * 
  * Commands:
- *   npx a3m-router serve [--port 8787]     Start OpenAI-compatible proxy server
+ *   npx a3m-router serve [--port 8787]    Start OpenAI-compatible proxy server
  *   npx a3m-router route <query>           Route query to best provider
- *   npx a3m-router batch <q1> <q2>..       Route multiple queries
+ *   npx a3m-router setup                  Interactive setup wizard (auto-detect API keys)
+ *   npx a3m-router batch <q1> <q2>..      Route multiple queries
  *   npx a3m-router providers               List all configured providers
  *   npx a3m-router test                    Test all providers
  *   npx a3m-router compare <query>         Compare providers side by side
@@ -155,6 +156,12 @@ async function main() {
   const router = createA3MRouter({ memory: { maxSize: 1000 } });
 
   switch (command) {
+    case 'setup': {
+      const { runWizard } = require('./cli/setupWizard.js');
+      runWizard();
+      break;
+    }
+
     case 'providers': {
       const providers = providerConfig.getAvailableProviders();
       const allProviders = providerConfig._providers;

package/dist/routing/providerRetry.d.ts

@@ -72,6 +72,11 @@ export declare class ProviderRetryHandler {
      * Check if an error should trigger a retry
      */
     isRetryableError(error: any): boolean;
+    /**
+     * Detect hard non-retryable account/policy states.
+     * These should fail fast instead of wasting retries.
+     */
+    private isPermanentProviderStateError;
     /**
      * Check if error is a rate limit (429)
      */

package/dist/routing/providerRetry.js

@@ -283,6 +283,10 @@ class ProviderRetryHandler {
     isRetryableError(error) {
         if (!error)
             return false;
+        // Known non-retryable account/quota/policy states seen in Chinese provider APIs.
+        if (this.isPermanentProviderStateError(error)) {
+            return false;
+        }
         const config = this.configs.get('default').retry;
         const retryableErrors = config.retryableErrors || DEFAULT_RETRY_CONFIG.retryableErrors;
         // Check error code/message
@@ -304,6 +308,39 @@ class ProviderRetryHandler {
             return true;
         return false;
     }
+    /**
+     * Detect hard non-retryable account/policy states.
+     * These should fail fast instead of wasting retries.
+     */
+    isPermanentProviderStateError(error) {
+        const status = Number(error?.status ?? error?.statusCode ?? 0);
+        const message = String(error?.message || '').toLowerCase();
+        const code = String(error?.code || '').toLowerCase();
+        // Payment/balance/account states are generally non-retryable without human action.
+        if (status === 402)
+            return true;
+        if (status === 401 || status === 403)
+            return true;
+        const hardPatterns = [
+            'insufficient',
+            'balance',
+            'invalid api key',
+            'incorrect api key',
+            'access terminated',
+            'account abnormal',
+            '违规',
+            'quota exhausted',
+            'hour allocated quota exceeded',
+            'week allocated quota exceeded',
+            'month allocated quota exceeded',
+            'tpd rate limit'
+        ];
+        for (const p of hardPatterns) {
+            if (message.includes(p) || code.includes(p))
+                return true;
+        }
+        return false;
+    }
     /**
      * Check if error is a rate limit (429)
      */