adaptive-memory-multi-model-router 1.9.4 → 2.0.0

package/dist/cache/semanticCache.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"semanticCache.js","sourceRoot":"","sources":["../../src/cache/semanticCache.ts"],"names":[],"mappings":";AAAA;;;;;;;;;GASG;;;AAgCH,+DAA+D;AAC/D,mBAAmB;AACnB,+DAA+D;AAE/D;;GAEG;AACH,SAAS,SAAS,CAAC,IAAY;IAC7B,OAAO,IAAI;SACR,WAAW,EAAE;SACb,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC;SACpB,IAAI,EAAE,CAAC;AACZ,CAAC;AAED;;;GAGG;AACH,SAAS,eAAe,CAAC,IAAY;IACnC,MAAM,UAAU,GAAG,GAAG,GAAG,SAAS,CAAC,IAAI,CAAC,GAAG,GAAG,CAAC;IAC/C,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAU,CAAC;IACnC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAChD,QAAQ,CAAC,GAAG,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;IAC/C,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED;;;GAGG;AACH,SAAS,OAAO,CAAC,CAAc,EAAE,CAAc;IAC7C,IAAI,CAAC,CAAC,IAAI,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,KAAK,CAAC;QAAE,OAAO,GAAG,CAAC;IAC7C,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,KAAK,MAAM,IAAI,IAAI,CAAC,EAAE,CAAC;QACrB,IAAI,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,YAAY,EAAE,CAAC;IAClC,CAAC;IACD,MAAM,KAAK,GAAG,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,GAAG,YAAY,CAAC;IAC7C,OAAO,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,YAAY,GAAG,KAAK,CAAC;AAChD,CAAC;AAED,+DAA+D;AAC/D,iBAAiB;AACjB,+DAA+D;AAE/D,MAAa,aAAa;IAChB,OAAO,GAAiB,EAAE,CAAC;IAC3B,OAAO,CAAS;IAChB,mBAAmB,CAAS;IAC5B,GAAG,CAAS;IACZ,IAAI,GAAG,CAAC,CAAC;IACT,MAAM,GAAG,CAAC,CAAC;IAEnB,YAAY,OAA8B;QACxC,IAAI,CAAC,OAAO,GAAG,OAAO,EAAE,OAAO,IAAI,IAAI,CAAC;QACxC,IAAI,CAAC,mBAAmB,GAAG,OAAO,EAAE,mBAAmB,IAAI,IAAI,CAAC;QAChE,IAAI,CAAC,GAAG,GAAG,OAAO,EAAE,GAAG,IAAI,OAAO,CAAC,CAAC,SAAS;IAC/C,CAAC;IAED;;;OAGG;IACH,KAAK,CAAC,GAAG,CAAC,KAAa;QACrB,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACvB,MAAM,aAAa,GAAG,eAAe,CAAC,KAAK,CAAC,CAAC;QAE7C,IAAI,SAAS,GAAsB,IAAI,CAAC;QACxC,IAAI,SAAS,GAAG,CAAC,CAAC;QAElB,KAAK,MAAM,KAAK,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YACjC,eAAe;YACf,IAAI,GAAG,GAAG,KAAK,CAAC,SAAS;gBAAE,SAAS;YAEpC,MAAM,KAAK,GAAG,OAAO,CAAC,aAAa,EAAE,KAAK,CAAC,QAAQ,CAAC,CAAC;YACrD,IAAI,KAAK,GAAG,SAAS,EAAE,CAAC;gBACtB,SAAS,GAAG,KAAK,CAAC;gBAClB,SAAS,GAAG,KAAK,CAAC;YACpB,CAAC;QACH,CAAC;QAED,IAAI,SAAS,IAAI,SAAS,IAAI,IAAI,CAAC,mBAAmB,EAAE,CAAC;YACvD,IAAI,CAAC,IAAI,EAAE,CAAC;YACZ,SAAS,CAAC,QAAQ,EAAE,CAAC;YACrB,OAAO;gBACL,KAAK,EAAE,SAAS,CAAC,KAAK;gBACtB,QAAQ,EAAE,SAAS,CAAC,QAAQ;gBAC5B,QAAQ,EAAE,SAAS,CAAC,QAAQ;gBAC5B,QAAQ,EAAE,SAAS,CAAC,QAAQ;gBAC5B,QAAQ,EAAE,SAAS,CAAC,QAAQ;aAC7B,CAAC;QACJ,CAAC;QAED,IAAI,CAAC,MAAM,EAAE,CAAC;QACd,OAAO,IAAI,CAAC;IACd,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,GAAG,CAAC,KAAa,EAAE,QAAgB,EAAE,QAAc;QACvD,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAEvB,8BAA8B;QAC9B,IAAI,CAAC,YAAY,EAAE,CAAC;QAEpB,8BAA8B;QAC9B,IAAI,IAAI,CAAC,OAAO,CAAC,MAAM,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YACxC,IAAI,CAAC,WAAW,EAAE,CAAC;QACrB,CAAC;QAED,6DAA6D;QAC7D,MAAM,UAAU,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;QACpC,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,CAChC,CAAC,CAAC,EAAE,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,UAAU,IAAI,GAAG,IAAI,CAAC,CAAC,SAAS,CAC/D,CAAC;QACF,IAAI,QAAQ,EAAE,CAAC;YACb,QAAQ,CAAC,QAAQ,GAAG,QAAQ,CAAC;YAC7B,QAAQ,CAAC,QAAQ,GAAG,QAAQ,CAAC;YAC7B,QAAQ,CAAC,QAAQ,GAAG,GAAG,CAAC;YACxB,QAAQ,CAAC,SAAS,GAAG,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC;YACpC,QAAQ,CAAC,QAAQ,GAAG,eAAe,CAAC,KAAK,CAAC,CAAC;YAC3C,OAAO;QACT,CAAC;QAED,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC;YAChB,KAAK;YACL,QAAQ;YACR,QAAQ;YACR,QAAQ,EAAE,GAAG;YACb,SAAS,EAAE,GAAG,GAAG,IAAI,CAAC,GAAG;YACzB,QAAQ,EAAE,CAAC;YACX,QAAQ,EAAE,eAAe,CAAC,KAAK,CAAC;SACjC,CAAC,CAAC;IACL,CAAC;IAED;;OAEG;IACH,KAAK;QACH,IAAI,CAAC,OAAO,GAAG,EAAE,CAAC;IACpB,CAAC;IAED;;OAEG;IACH,QAAQ;QACN,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,MAAM,CAAC;QACtC,OAAO;YACL,IAAI,EAAE,IAAI,CAAC,IAAI;YACf,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,OAAO,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;YAC1C,IAAI,EAAE,IAAI,CAAC,OAAO,CAAC,MAAM;SAC1B,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,YAAY;QAClB,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACvB,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC,SAAS,CAAC,CAAC;IAChE,CAAC;IAED;;OAEG;IACK,WAAW;QACjB,IAAI,IAAI,CAAC,OAAO,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO;QACtC,IAAI,SAAS,GAAG,CAAC,CAAC;QAClB,IAAI,UAAU,GAAG,QAAQ,CAAC;QAC1B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC7C,IAAI,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,QAAQ,GAAG,UAAU,EAAE,CAAC;gBAC1C,UAAU,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC;gBACtC,SAAS,GAAG,CAAC,CAAC;YAChB,CAAC;QACH,CAAC;QACD,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;IACpC,CAAC;CACF;AAtID,sCAsIC"}

package/dist/cli.js

@@ -3,6 +3,7 @@
  * A3M Router CLI - Adaptive Memory Multi-Model Router
  * 
  * Commands:
+ *   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 providers               List all configured providers
@@ -24,6 +25,13 @@ const {
   getAvailableProviders, providerConfig, registerProvider, loadProviders,
 } = require('./index.js');
 
+let createProxyServer;
+try {
+  createProxyServer = require('./server/proxyServer.js').createProxyServer;
+} catch (e) {
+  // Server module not yet compiled
+}
+
 const args = process.argv.slice(2);
 const command = args[0];
 
@@ -394,11 +402,36 @@ async function main() {
       break;
     }
 
+    case 'serve': {
+      if (!createProxyServer) {
+        console.error('\n  Error: Server module not available.');
+        console.error('  Build first: npm run build');
+        console.error('  Or use directly: node -e "require(\'./server/proxyServer.js\').createProxyServer()"\n');
+        process.exit(1);
+      }
+
+      // Parse --port argument
+      var portArg = args.indexOf('--port');
+      var port = undefined;
+      if (portArg !== -1 && args[portArg + 1]) {
+        port = parseInt(args[portArg + 1], 10);
+        if (isNaN(port)) {
+          console.error('  Error: --port must be a number');
+          process.exit(1);
+        }
+      }
+
+      console.log('\n  Starting A3M Router Proxy Server...');
+      createProxyServer(port);
+      break;
+    }
+
     default:
       console.log('\n🔀 A3M Router — Adaptive Memory Multi-Model Router');
       console.log('━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━');
       console.log('');
       console.log('  Commands:');
+      console.log('    serve [--port 8787]    Start OpenAI-compatible proxy server');
       console.log('    route <query>          Route query to best provider');
       console.log('    batch <q1> <q2>..      Route multiple queries');
       console.log('    compare <query>        Compare providers side by side');
@@ -420,6 +453,8 @@ async function main() {
       console.log('  Env:    GROQ_API_KEY, CEREBRAS_API_KEY, MISTRAL_API_KEY, etc.');
       console.log('');
       console.log('  Examples:');
+      console.log('    npx a3m-router serve                        # Start proxy on :8787');
+      console.log('    npx a3m-router serve --port 3000            # Start proxy on :3000');
       console.log('    npx a3m-router route "Write a Python function to sort"');
       console.log('    npx a3m-router compare "What is 2+2?"');
       console.log('    npx a3m-router providers');

package/dist/geo/generativeEngineOptimization.js

@@ -0,0 +1,321 @@
+"use strict";
+/**
+ * A3M Router - Generative Engine Optimization (GEO)
+ *
+ * Features to make the package discoverable by AI agents and LLMs:
+ * - Structured metadata for AI consumption
+ * - Intent-to-code mapping
+ * - AI-friendly documentation generation
+ * - LLM-optimized examples
+ */
+
+// Intent-to-code mapping for AI agents
+const INTENT_MAP = {
+  // Routing intents
+  'route llm query': {
+    code: `const { routeQuery } = require('adaptive-memory-multi-model-router');
+const result = routeQuery("Your query here");
+console.log(result.primary_model);`,
+    description: 'Route a query to the optimal LLM provider',
+  },
+  'route to cheapest provider': {
+    code: `const { routeQuery } = require('adaptive-memory-multi-model-router');
+const result = routeQuery("Your query", { budget_multiplier: 0.3 });`,
+    description: 'Route to cheapest capable provider',
+  },
+  'route to best quality': {
+    code: `const { routeQuery } = require('adaptive-memory-multi-model-router');
+const result = routeQuery("Your query", { quality_priority: true });`,
+    description: 'Route to highest quality provider',
+  },
+
+  // Batch processing intents
+  'batch process prompts': {
+    code: `const { routeBatch } = require('adaptive-memory-multi-model-router');
+const queries = ["Q1", "Q2", "Q3"];
+const results = routeBatch(queries);`,
+    description: 'Process multiple prompts with automatic routing',
+  },
+  'parallel llm calls': {
+    code: `const { routeBatch } = require('adaptive-memory-multi-model-router');
+const results = routeBatch(queries, { concurrency: 5 });`,
+    description: 'Execute LLM calls in parallel',
+  },
+
+  // Cost tracking intents
+  'track llm costs': {
+    code: `const { createA3MRouter } = require('adaptive-memory-multi-model-router');
+const router = createA3MRouter();
+const summary = router.costTracker.getSummary();
+console.log(\`Total: \$\${summary.totalSpent}\`);`,
+    description: 'Track API costs across all providers',
+  },
+  'estimate api cost': {
+    code: `const { estimateCost } = require('adaptive-memory-multi-model-router');
+const cost = estimateCost(1000, 500, 'gpt-4o');
+console.log(\`Cost: \$\${cost.toFixed(6)}\`);`,
+    description: 'Estimate cost before making API call',
+  },
+
+  // Provider management intents
+  'list llm providers': {
+    code: `const { getAvailableProviders } = require('adaptive-memory-multi-model-router');
+const providers = getAvailableProviders();
+console.log(Object.keys(providers));`,
+    description: 'List all configured LLM providers',
+  },
+  'add custom provider': {
+    code: `const { registerProvider } = require('adaptive-memory-multi-model-router');
+registerProvider('my-provider', {
+  baseUrl: 'https://api.myprovider.com',
+  models: ['my-model'],
+  type: 'api'
+});`,
+    description: 'Register a custom LLM provider',
+  },
+  'check provider health': {
+    code: `const { providerConfig } = require('adaptive-memory-multi-model-router');
+const health = await providerConfig.healthCheck('groq');
+console.log(health.healthy ? '✅' : '❌');`,
+    description: 'Check if a provider is healthy',
+  },
+
+  // Caching intents
+  'cache llm responses': {
+    code: `const { createA3MRouter } = require('adaptive-memory-multi-model-router');
+const router = createA3MRouter({ cache: { ttl_seconds: 3600 } });`,
+    description: 'Enable response caching',
+  },
+
+  // Fallback intents
+  'setup llm fallback': {
+    code: `const { createA3MRouter } = require('adaptive-memory-multi-model-router');
+const router = createA3MRouter();
+// Fallback is automatic - just use router.route()`,
+    description: 'Automatic fallback when providers fail',
+  },
+
+  // CLI intents
+  'route from command line': {
+    code: `npx a3m-router route "Your query here"`,
+    description: 'Route queries via CLI',
+  },
+  'benchmark providers': {
+    code: `npx a3m-router benchmark`,
+    description: 'Compare all configured providers',
+  },
+};
+
+// AI-optimized package metadata
+const PACKAGE_METADATA = {
+  name: 'adaptive-memory-multi-model-router',
+  shortName: 'A3M Router',
+  category: 'LLM Routing & Cost Optimization',
+  primaryUseCases: [
+    'Route LLM queries to optimal providers',
+    'Reduce API costs through intelligent selection',
+    'Handle provider failures with automatic fallback',
+    'Process prompts in parallel batches',
+    'Track spending across multiple LLM APIs',
+  ],
+  keyFeatures: [
+    'Learned routing based on query analysis',
+    'Cost optimization (50-80% savings)',
+    'Automatic provider fallback',
+    'Batch processing with rate limiting',
+    'Response caching',
+    'Real-time cost tracking',
+    'CLI tools',
+  ],
+  supportedProviders: [
+    'Groq', 'Cerebras', 'Mistral', 'OpenAI', 'Anthropic',
+    'Google', 'DeepSeek', 'CommandCode', 'OpenCode',
+    'Ollama', 'vLLM', 'LM Studio',
+  ],
+  alternatives: ['LiteLLM', 'OpenRouter', 'Helicone'],
+  differentiators: [
+    'Learned routing (not just proxying)',
+    'Cost-quality tradeoff optimization',
+    'Generic provider system (user-configurable)',
+    'Built-in fallback and retry',
+    'Comprehensive CLI',
+  ],
+};
+
+// Keywords for AI indexing
+const AI_INDEXING_KEYWORDS = [
+  // Primary
+  'llm routing', 'cost optimization', 'multi-provider', 'adaptive routing',
+  // Features
+  'batch processing', 'parallel execution', 'fallback', 'retry', 'load balancing',
+  // Providers
+  'openai', 'anthropic', 'claude', 'groq', 'cerebras', 'mistral',
+  // Concepts
+  'routellm', 'radix-attention', 'medusa', 'token compression',
+  // Use cases
+  'api gateway', 'llm proxy', 'model router', 'cost tracking',
+  // Integrations
+  'github', 'slack', 'telegram', 'notion', 'discord',
+  // Technical
+  'typescript', 'javascript', 'nodejs', 'cli', 'sdk',
+];
+
+class GenerativeEngineOptimizer {
+  constructor() {
+    this.intentMap = INTENT_MAP;
+    this.metadata = PACKAGE_METADATA;
+    this.keywords = AI_INDEXING_KEYWORDS;
+  }
+
+  /**
+   * Get code example for a specific intent
+   */
+  getCodeForIntent(intent) {
+    const normalized = intent.toLowerCase().trim();
+
+    // Exact match
+    if (this.intentMap[normalized]) {
+      return this.intentMap[normalized];
+    }
+
+    // Partial match
+    for (const [key, value] of Object.entries(this.intentMap)) {
+      if (normalized.includes(key) || key.includes(normalized)) {
+        return value;
+      }
+    }
+
+    // Fuzzy match - find best similarity
+    let bestMatch = null;
+    let bestScore = 0;
+
+    for (const key of Object.keys(this.intentMap)) {
+      const score = this.calculateSimilarity(normalized, key);
+      if (score > bestScore && score > 0.3) {
+        bestScore = score;
+        bestMatch = key;
+      }
+    }
+
+    return bestMatch ? this.intentMap[bestMatch] : null;
+  }
+
+  /**
+   * Simple similarity calculation
+   */
+  calculateSimilarity(a, b) {
+    const aWords = a.split(/\s+/);
+    const bWords = b.split(/\s+/);
+
+    const intersection = aWords.filter(word => bWords.includes(word));
+    return intersection.length / Math.max(aWords.length, bWords.length);
+  }
+
+  /**
+   * Get all intents matching a category
+   */
+  getIntentsByCategory(category) {
+    const categories = {
+      routing: ['route llm query', 'route to cheapest provider', 'route to best quality'],
+      batch: ['batch process prompts', 'parallel llm calls'],
+      cost: ['track llm costs', 'estimate api cost'],
+      providers: ['list llm providers', 'add custom provider', 'check provider health'],
+      caching: ['cache llm responses'],
+      fallback: ['setup llm fallback'],
+      cli: ['route from command line', 'benchmark providers'],
+    };
+
+    const keys = categories[category] || [];
+    return keys.map(key => this.intentMap[key]).filter(Boolean);
+  }
+
+  /**
+   * Generate AI-optimized documentation
+   */
+  generateAIDocumentation() {
+    return {
+      metadata: this.metadata,
+      quickStart: this.intentMap['route llm query'],
+      commonIntents: Object.entries(this.intentMap).slice(0, 5).map(([intent, data]) => ({
+        intent,
+        ...data,
+      })),
+      keywords: this.keywords,
+    };
+  }
+
+  /**
+   * Get package metadata for AI consumption
+   */
+  getPackageMetadata() {
+    return {
+      ...this.metadata,
+      installation: 'npm install adaptive-memory-multi-model-router',
+      github: 'https://github.com/Das-rebel/adaptive-memory-multi-model-router',
+      npm: 'https://www.npmjs.com/package/adaptive-memory-multi-model-router',
+      weeklyDownloads: 872,
+      testCount: 33,
+      keywordCount: 139,
+    };
+  }
+
+  /**
+   * Search intents by keyword
+   */
+  searchIntents(keyword) {
+    const results = [];
+    const normalized = keyword.toLowerCase();
+
+    for (const [intent, data] of Object.entries(this.intentMap)) {
+      if (intent.includes(normalized) || data.description.toLowerCase().includes(normalized)) {
+        results.push({ intent, ...data });
+      }
+    }
+
+    return results;
+  }
+
+  /**
+   * Get comparison with alternatives
+   */
+  getComparison() {
+    return {
+      package: this.metadata.name,
+      alternatives: this.metadata.alternatives,
+      differentiators: this.metadata.differentiators,
+      recommendation: 'Use A3M Router for learned routing with cost optimization. Use LiteLLM for simple proxying. Use OpenRouter for hosted routing.',
+    };
+  }
+}
+
+// Convenience functions
+function getCodeForIntent(intent) {
+  const geo = new GenerativeEngineOptimizer();
+  return geo.getCodeForIntent(intent);
+}
+
+function searchIntents(keyword) {
+  const geo = new GenerativeEngineOptimizer();
+  return geo.searchIntents(keyword);
+}
+
+function getPackageMetadata() {
+  const geo = new GenerativeEngineOptimizer();
+  return geo.getPackageMetadata();
+}
+
+function generateAIDocumentation() {
+  const geo = new GenerativeEngineOptimizer();
+  return geo.generateAIDocumentation();
+}
+
+module.exports = {
+  GenerativeEngineOptimizer,
+  getCodeForIntent,
+  searchIntents,
+  getPackageMetadata,
+  generateAIDocumentation,
+  INTENT_MAP,
+  PACKAGE_METADATA,
+  AI_INDEXING_KEYWORDS,
+};