adaptive-memory-multi-model-router 2.2.4 → 2.2.6

package/dist/sdk.js

@@ -5,118 +5,127 @@
  * Clean wrapper class providing a better DX than raw exports.
  *
  * Usage:
- *   const { A3MRouter } = require('adaptive-memory-multi-model-router/sdk');
+ *   import { A3MRouter } from 'adaptive-memory-multi-model-router/sdk';
+ *
  *   const router = new A3MRouter();
+ *
+ *   // Route a query (no execution, just model selection)
  *   const decision = router.route("What is 2+2?");
  *   console.log(decision.model, decision.tier, decision.cost);
+ *
+ *   // Start the OpenAI-compatible proxy server
+ *   const proxyURL = await router.serve(8787);
+ *
+ *   // Use with any OpenAI SDK
+ *   import OpenAI from 'openai';
+ *   const client = new OpenAI({ baseURL: router.proxyURL });
+ *   const response = await client.chat.completions.create({
+ *     model: 'auto',
+ *     messages: [{ role: 'user', content: 'Hello' }]
+ *   });
  */
-
-const advancedRouter = require("./routing/advancedRouter");
-const proxyServer = require("./server/proxyServer");
-
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.A3MRouter = void 0;
+exports.createSDK = createSDK;
+const advancedRouter_1 = require("./routing/advancedRouter");
+const proxyServer_1 = require("./server/proxyServer");
 // ============================================================
 // A3MRouter SDK Class
 // ============================================================
-
 class A3MRouter {
-  constructor(config = {}) {
-    this.config = config;
-    this._proxyURL = null;
-  }
-
-  /**
-   * Route a query — returns model selection without executing it.
-   *
-   * @param {string} query - The user prompt to route
-   * @returns {object} Routing decision with model, tier, cost, complexity
-   */
-  route(query) {
-    const features = advancedRouter.extractQueryFeatures(query);
-    const result = advancedRouter.routeQuery(query, this.config.providers);
-
-    return {
-      model: result.primary_model || 'unknown',
-      tier: this.classifyTier(features.complexity),
-      cost: result.estimated_cost || 0,
-      complexity: features.complexity,
-      reasoning: result.reasoning || '',
-      fallbackModels: result.fallback_models || [],
-      isFree: (result.estimated_cost || 0) === 0,
-      isExpert: features.complexity >= 0.65,
-    };
-  }
-
-  /**
-   * Route multiple queries in batch.
-   *
-   * @param {string[]} queries - Array of user prompts
-   * @returns {object[]} Array of routing decisions
-   */
-  routeBatch(queries) {
-    advancedRouter.routeBatch(queries); // warm the internal cache
-    return queries.map((q) => this.route(q));
-  }
-
-  /**
-   * Get model recommendation for a task description.
-   *
-   * @param {string} task - Task description
-   * @returns {object} Routing decision
-   */
-  recommend(task) {
-    advancedRouter.recommendForTask(task);
-    return this.route(task);
-  }
-
-  /**
-   * Start the OpenAI-compatible proxy server.
-   *
-   * @param {number} port - Port to listen on (default: 8787)
-   * @returns {Promise<string>} The proxy base URL
-   */
-  async serve(port = 8787) {
-    proxyServer.createProxyServer(port);
-    this._proxyURL = `http://localhost:${port}/v1`;
-    return this._proxyURL;
-  }
-
-  /**
-   * Get the proxy URL. Available after serve() is called.
-   */
-  get proxyURL() {
-    return this._proxyURL || 'http://localhost:8787/v1';
-  }
-
-  /**
-   * Extract features from a query for debugging or analysis.
-   *
-   * @param {string} query - The user prompt to analyze
-   * @returns {object} Detailed feature breakdown
-   */
-  analyze(query) {
-    return advancedRouter.extractQueryFeatures(query);
-  }
-
-  /**
-   * Classify a complexity score into a named tier.
-   */
-  classifyTier(complexity) {
-    if (complexity < 0.20) return 'free';
-    if (complexity < 0.45) return 'cheap';
-    if (complexity < 0.65) return 'mid';
-    return 'premium';
-  }
+    config;
+    _proxyURL = null;
+    constructor(config = {}) {
+        this.config = config;
+    }
+    /**
+     * Route a query — returns model selection without executing it.
+     *
+     * @param query - The user prompt to route
+     * @returns Routing decision with model, tier, cost, complexity
+     */
+    route(query) {
+        const features = (0, advancedRouter_1.extractQueryFeatures)(query);
+        const result = (0, advancedRouter_1.routeQuery)(query, this.config.providers);
+        return {
+            model: result.primary_model || 'unknown',
+            tier: this.classifyTier(features.complexity),
+            cost: result.estimated_cost || 0,
+            complexity: features.complexity,
+            reasoning: result.reasoning || '',
+            fallbackModels: result.fallback_models || [],
+            isFree: (result.estimated_cost || 0) === 0,
+            isExpert: features.complexity >= 0.65,
+        };
+    }
+    /**
+     * Route multiple queries in batch.
+     *
+     * @param queries - Array of user prompts
+     * @returns Array of routing decisions
+     */
+    routeBatch(queries) {
+        (0, advancedRouter_1.routeBatch)(queries); // warm the internal cache
+        return queries.map((q) => this.route(q));
+    }
+    /**
+     * Get model recommendation for a task description.
+     *
+     * @param task - Task description (e.g. "code generation", "summarization")
+     * @returns Routing decision
+     */
+    recommend(task) {
+        (0, advancedRouter_1.recommendForTask)(task);
+        return this.route(task);
+    }
+    /**
+     * Start the OpenAI-compatible proxy server.
+     *
+     * @param port - Port to listen on (default: 8787)
+     * @returns The proxy base URL (e.g. "http://localhost:8787/v1")
+     */
+    async serve(port = 8787) {
+        (0, proxyServer_1.createProxyServer)(port);
+        this._proxyURL = `http://localhost:${port}/v1`;
+        return this._proxyURL;
+    }
+    /**
+     * Get the proxy URL. Available after serve() is called,
+     * otherwise returns the default.
+     */
+    get proxyURL() {
+        return this._proxyURL || 'http://localhost:8787/v1';
+    }
+    /**
+     * Extract features from a query for debugging or analysis.
+     *
+     * @param query - The user prompt to analyze
+     * @returns Detailed feature breakdown
+     */
+    analyze(query) {
+        return (0, advancedRouter_1.extractQueryFeatures)(query);
+    }
+    /**
+     * Classify a complexity score into a named tier.
+     */
+    classifyTier(complexity) {
+        if (complexity < 0.20)
+            return 'free';
+        if (complexity < 0.45)
+            return 'cheap';
+        if (complexity < 0.65)
+            return 'mid';
+        return 'premium';
+    }
 }
-
+exports.A3MRouter = A3MRouter;
 /**
  * Convenience: create an A3MRouter instance.
  *
- * @param {object} config - Optional configuration
- * @returns {A3MRouter} Configured instance
+ * @param config - Optional configuration
+ * @returns Configured A3MRouter instance
  */
 function createSDK(config) {
-  return new A3MRouter(config);
+    return new A3MRouter(config);
 }
-
-module.exports = { A3MRouter, createSDK };
-module.exports.default = A3MRouter;
+//# sourceMappingURL=sdk.js.map

package/dist/sdk.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sdk.js","sourceRoot":"","sources":["../src/sdk.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;;;AAqKH,8BAEC;AArKD,6DAKkC;AAClC,sDAAyD;AAoDzD,+DAA+D;AAC/D,sBAAsB;AACtB,+DAA+D;AAE/D,MAAa,SAAS;IACZ,MAAM,CAAkB;IACxB,SAAS,GAAkB,IAAI,CAAC;IAExC,YAAY,SAA0B,EAAE;QACtC,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;IACvB,CAAC;IAED;;;;;OAKG;IACH,KAAK,CAAC,KAAa;QACjB,MAAM,QAAQ,GAAG,IAAA,qCAAoB,EAAC,KAAK,CAAC,CAAC;QAC7C,MAAM,MAAM,GAAG,IAAA,2BAAU,EAAC,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;QAExD,OAAO;YACL,KAAK,EAAE,MAAM,CAAC,aAAa,IAAI,SAAS;YACxC,IAAI,EAAE,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,UAAU,CAAC;YAC5C,IAAI,EAAE,MAAM,CAAC,cAAc,IAAI,CAAC;YAChC,UAAU,EAAE,QAAQ,CAAC,UAAU;YAC/B,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,EAAE;YACjC,cAAc,EAAE,MAAM,CAAC,eAAe,IAAI,EAAE;YAC5C,MAAM,EAAE,CAAC,MAAM,CAAC,cAAc,IAAI,CAAC,CAAC,KAAK,CAAC;YAC1C,QAAQ,EAAE,QAAQ,CAAC,UAAU,IAAI,IAAI;SACtC,CAAC;IACJ,CAAC;IAED;;;;;OAKG;IACH,UAAU,CAAC,OAAiB;QAC1B,IAAA,2BAAU,EAAC,OAAO,CAAC,CAAC,CAAC,0BAA0B;QAC/C,OAAO,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3C,CAAC;IAED;;;;;OAKG;IACH,SAAS,CAAC,IAAY;QACpB,IAAA,iCAAgB,EAAC,IAAI,CAAC,CAAC;QACvB,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC1B,CAAC;IAED;;;;;OAKG;IACH,KAAK,CAAC,KAAK,CAAC,OAAe,IAAI;QAC7B,IAAA,+BAAiB,EAAC,IAAI,CAAC,CAAC;QACxB,IAAI,CAAC,SAAS,GAAG,oBAAoB,IAAI,KAAK,CAAC;QAC/C,OAAO,IAAI,CAAC,SAAS,CAAC;IACxB,CAAC;IAED;;;OAGG;IACH,IAAI,QAAQ;QACV,OAAO,IAAI,CAAC,SAAS,IAAI,0BAA0B,CAAC;IACtD,CAAC;IAED;;;;;OAKG;IACH,OAAO,CAAC,KAAa;QACnB,OAAO,IAAA,qCAAoB,EAAC,KAAK,CAAC,CAAC;IACrC,CAAC;IAED;;OAEG;IACK,YAAY,CAClB,UAAkB;QAElB,IAAI,UAAU,GAAG,IAAI;YAAE,OAAO,MAAM,CAAC;QACrC,IAAI,UAAU,GAAG,IAAI;YAAE,OAAO,OAAO,CAAC;QACtC,IAAI,UAAU,GAAG,IAAI;YAAE,OAAO,KAAK,CAAC;QACpC,OAAO,SAAS,CAAC;IACnB,CAAC;CACF;AA7FD,8BA6FC;AAED;;;;;GAKG;AACH,SAAgB,SAAS,CAAC,MAAwB;IAChD,OAAO,IAAI,SAAS,CAAC,MAAM,CAAC,CAAC;AAC/B,CAAC"}

package/dist/security/guardrails.d.ts

@@ -74,3 +74,4 @@ export declare class GuardrailEngine {
 }
 export declare function createGuardrails(config?: Partial<GuardrailConfig>): GuardrailEngine;
 export {};
+//# sourceMappingURL=guardrails.d.ts.map

package/dist/server/dashboard.d.ts

@@ -56,3 +56,4 @@ export declare function registerProvider(id: string, name: string): void;
  */
 export declare function handleDashboardRequest(req: http.IncomingMessage, res: http.ServerResponse): boolean;
 export declare function getDashboardHTML(): string;
+//# sourceMappingURL=dashboard.d.ts.map

package/dist/server/modelMapper.d.ts

@@ -41,3 +41,4 @@ export declare function listAvailableModels(): Array<{
     created: number;
     owned_by: string;
 }>;
+//# sourceMappingURL=modelMapper.d.ts.map

package/dist/server/proxyServer.d.ts

@@ -39,3 +39,4 @@ declare const costTracker: any;
 export declare function createProxyServer(port?: number): http.Server;
 export { CostTracker, costTracker, requestLogs };
 export default createProxyServer;
+//# sourceMappingURL=proxyServer.d.ts.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "adaptive-memory-multi-model-router",
-  "version": "2.2.4",
+  "version": "2.2.6",
   "shortName": "A3M Router",
   "displayName": "A3M Router - Adaptive Memory Multi-Model Router",
   "description": "LLM router & AI gateway with 99.5% routing accuracy — supports 47 providers including DeepSeek, Kimi (Moonshot), Qwen, Zhipu GLM, Yi, Baichuan, MiniMax, StepFun. Zero ML, 19.5KB. Multi-signal routing, semantic cache, guardrails, cost analytics. MIT. TypeScript SDK + Python SDK + OpenAI proxy.",
@@ -354,7 +354,108 @@
     "pi-coding-agent",
     "pi-agent",
     "agent-discoverable",
-    "ai-native"
+    "ai-native",
+    "01-ai",
+    "01ai",
+    "128k上下文",
+    "ai-gateway-cn",
+    "ai路由",
+    "aliyun",
+    "baichuan-ai",
+    "baichuan2",
+    "baichuan2-flash",
+    "baiducloud",
+    "chinese-ai-proxy",
+    "chinese-api-gateway",
+    "chinese-chatgpt",
+    "chinese-language-model",
+    "chinese-llm-proxy",
+    "chinese-models",
+    "chinese-nlp",
+    "chinesellm",
+    "chinesellm-routing",
+    "deepseek-ai",
+    "deepseek-api",
+    "deepseek-chat",
+    "deepseek-coder",
+    "deepseek-reasoner",
+    "embedding-model",
+    "glm-4",
+    "glm-4-air",
+    "glm-4-flash",
+    "glm-4-long",
+    "glm-4-plus",
+    "huawei-cloud",
+    "langchain-cn",
+    "lingyi",
+    "lingyi-wanwu",
+    "llamaindex-cn",
+    "llm-gateway-cn",
+    "long-context-model",
+    "minimax-ai",
+    "minimax-api",
+    "minimax-chat",
+    "minimax-turbo",
+    "moonshot-ai",
+    "moonshot-v1",
+    "moonshot-v1-128k",
+    "moonshot-v1-32k",
+    "moonshot-v1-8k",
+    "multimodal",
+    "qianwen",
+    "qwen-long",
+    "qwen-max",
+    "qwen-plus",
+    "qwen-turbo",
+    "qwen2",
+    "qwen3",
+    "step-1",
+    "step-1v",
+    "step-2",
+    "tencentcloud",
+    "tongji",
+    "tongyi",
+    "tongyi-qianwen",
+    "vision-model",
+    "volcengine",
+    "wu-yuan",
+    "wuyuan",
+    "yi-ai",
+    "yi-large",
+    "yi-lightning",
+    "yi-medium",
+    "zai-glm",
+    "zai-glm-4",
+    "zhipu-ai",
+    "中国ai",
+    "中国大模型",
+    "中文chatgpt",
+    "中文embedding",
+    "中文langchain",
+    "中文llamaindex",
+    "中文nlp",
+    "中文seo",
+    "中文多模态",
+    "中文大模型",
+    "中文搜索引擎优化",
+    "中文模型",
+    "华为云",
+    "向量化",
+    "国产ai",
+    "国产大模型",
+    "多模型路由",
+    "多模态",
+    "大模型路由",
+    "字节ai",
+    "智能路由",
+    "火山引擎",
+    "百度ai",
+    "百度云",
+    "腾讯ai",
+    "腾讯云",
+    "长上下文",
+    "阿里ai",
+    "阿里云"
   ],
   "author": "Das-rebel <subho@example.com>",
   "license": "MIT",
@@ -370,7 +471,8 @@
     "test": "node test.js && node test/provider-test.js",
     "test:providers": "node test/provider-test.js",
     "benchmark": "node test/benchmark.js",
-    "benchmark:verbose": "node test/benchmark.js --verbose"
+    "benchmark:verbose": "node test/benchmark.js --verbose",
+    "build": "tsc"
   },
   "engines": {
     "node": ">=18.0.0"
@@ -388,6 +490,7 @@
   },
   "devDependencies": {
     "@types/node": "^25.8.0",
+    "tsx": "^4.22.3",
     "typescript": "^6.0.3"
   }
 }

package/src/cache/cacheKeyGenerator.ts

@@ -0,0 +1,242 @@
+/**
+ * A3M Router - Cross-Provider Cache Key Generator
+ *
+ * Normalizes prompts so same semantic content maps to same cache key
+ * regardless of provider-specific formatting, system prompts, etc.
+ *
+ * Usage:
+ *   const cacheKey = generateCacheKey("What is Python?", { provider: "openai" });
+ *   const key2 = generateCacheKey("What is Python?", { provider: "anthropic" });
+ *   // key === key2 (same semantic content = same key)
+ */
+
+import * as crypto from 'crypto';
+
+// ============================================================
+// Types
+// ============================================================
+
+export interface CacheKeyOptions {
+  /** Target provider (affects normalization rules) */
+  provider?: string;
+  /** Target model (for model-specific normalization) */
+  model?: string;
+  /** Whether to include system prompt in normalization */
+  includeSystemPrompt?: boolean;
+  /** Custom normalization rules */
+  customRules?: NormalizationRule[];
+}
+
+export interface NormalizationRule {
+  pattern: RegExp;
+  replacement: string;
+}
+
+export interface CacheKeyResult {
+  /** The normalized cache key string */
+  key: string;
+  /** Hash of the normalized content */
+  hash: string;
+  /** Metadata about what was normalized */
+  metadata: {
+    originalLength: number;
+    normalizedLength: number;
+    rulesApplied: number;
+    provider?: string;
+  };
+}
+
+// ============================================================
+// Provider-specific system prompt patterns
+// ============================================================
+
+const PROVIDER_SYSTEM_PATTERNS: Record<string, RegExp[]> = {
+  anthropic: [
+    /<anthropic_thinking>[\s\S]*?<\/anthropic_thinking>/gi,
+    /<thinking>[\s\S]*?<\/thinking>/gi,
+    /Human:/gi,
+    /Assistant:/gi,
+  ],
+  openai: [
+    /<|im_start|>/gi,
+    /<|im_end|>/gi,
+  ],
+  google: [
+    /<content>[\s\S]*?<\/content>/gi,
+    /[\Parts|thought]/gi,
+  ],
+};
+
+// ============================================================
+// Core normalizer
+// ============================================================
+
+/**
+ * Normalize text for cross-provider cache key generation.
+ * Removes provider-specific formatting while preserving semantic content.
+ */
+export function normalizeForCacheKey(
+  text: string,
+  options: CacheKeyOptions = {}
+): string {
+  let normalized = text;
+
+  // Step 1: Unicode normalization (NFC)
+  normalized = normalized.normalize('NFC');
+
+  // Step 2: Collapse whitespace
+  normalized = normalized.replace(/\s+/g, ' ');
+
+  // Step 3: Remove control characters
+  normalized = normalized.replace(/[\x00-\x1F\x7F]/g, '');
+
+  // Step 4: Strip provider-specific formatting
+  if (options.provider) {
+    const patterns = PROVIDER_SYSTEM_PATTERNS[options.provider] || [];
+    for (const pattern of patterns) {
+      normalized = normalized.replace(pattern, '');
+    }
+  }
+
+  // Step 5: General system/assistant role removal
+  normalized = normalized
+    .replace(/\b(system|user|assistant|human|bot)\s*:/gi, '')
+    .replace(/^(system|user|assistant|human|bot)\s*/gim, '');
+
+  // Step 6: Remove markdown formatting (often provider-specific)
+  normalized = normalized
+    .replace(/```[\s\S]*?```/g, '[CODE_BLOCK]')  // Preserve code block indicator
+    .replace(/`([^`]+)`/g, '$1')                 // Inline code content
+    .replace(/\*\*([^*]+)\*\*/g, '$1')           // Bold
+    .replace(/_([^_]+)_/g, '$1')                  // Italic
+    .replace(/#+\s*/g, '')                        // Headers
+    .replace(/^\s*[-*+]\s+/gm, '')                // List bullets
+    .replace(/^\s*\d+\.\s+/gm, '');              // Numbered lists
+
+  // Step 7: Apply custom rules
+  if (options.customRules) {
+    for (const rule of options.customRules) {
+      normalized = normalized.replace(rule.pattern, rule.replacement);
+    }
+  }
+
+  // Step 8: Collapse whitespace again after removals
+  normalized = normalized.replace(/\s+/g, ' ').trim();
+
+  return normalized;
+}
+
+/**
+ * Generate a deterministic cache key from a query.
+ * Same semantic content = same key across providers.
+ */
+export function generateCacheKey(
+  query: string,
+  options: CacheKeyOptions = {}
+): CacheKeyResult {
+  const originalLength = query.length;
+
+  // Normalize the query
+  let normalized = normalizeForCacheKey(query, {
+    ...options,
+    includeSystemPrompt: false, // Always exclude for user query matching
+  });
+
+  // Count rules that were applied (approximate)
+  let rulesApplied = 3; // Base normalizations
+  if (options.provider) rulesApplied += 2;
+  if (options.customRules) rulesApplied += options.customRules.length;
+
+  // Generate hash
+  const hash = crypto
+    .createHash('sha256')
+    .update(normalized)
+    .digest('hex')
+    .substring(0, 16); // First 16 chars = 64-bit key
+
+  // Final key format: v1:{hash}:{provider?[:model]?}
+  let key = `v1:${hash}`;
+  if (options.provider) {
+    key += `:${options.provider}`;
+    if (options.model) {
+      key += `:${options.model}`;
+    }
+  }
+
+  return {
+    key,
+    hash,
+    metadata: {
+      originalLength,
+      normalizedLength: normalized.length,
+      rulesApplied,
+      provider: options.provider,
+    },
+  };
+}
+
+// ============================================================
+// SemanticCache enhancement
+// ============================================================
+
+/**
+ * Add cross-provider cache key methods to existing SemanticCache.
+ * Call this to enhance the cache with provider-normalized lookups.
+ */
+export function createCacheKeyGenerator(
+  defaultOptions?: CacheKeyOptions
+): {
+  generateKey: (query: string, options?: CacheKeyOptions) => CacheKeyResult;
+  createNormalizedMatcher: (cache: Map<string, any>) => (query: string, options?: CacheKeyOptions) => string | null;
+} {
+  return {
+    /**
+     * Generate a cache key for a query.
+     */
+    generateKey: (query: string, options?: CacheKeyOptions): CacheKeyResult => {
+      return generateCacheKey(query, { ...defaultOptions, ...options });
+    },
+
+    /**
+     * Create a matcher function that finds existing cache entries
+     * by comparing normalized keys.
+     */
+    createNormalizedMatcher: (cache: Map<string, any>) => {
+      return (query: string, options?: CacheKeyOptions): string | null => {
+        const { key } = generateCacheKey(query, { ...defaultOptions, ...options });
+
+        // Check exact match
+        if (cache.has(key)) {
+          return key;
+        }
+
+        // Check hash-only match (v1:{hash} prefix)
+        const hashPrefix = key.split(':').slice(0, 2).join(':');
+        for (const cachedKey of cache.keys()) {
+          if (cachedKey.startsWith(hashPrefix + ':')) {
+            return cachedKey;
+          }
+        }
+
+        return null;
+      };
+    },
+  };
+}
+
+// ============================================================
+// Convenience exports
+// ============================================================
+
+/**
+ * Quick cache key generation (simplified API).
+ * Use this for simple cross-provider cache lookups.
+ *
+ * @example
+ * const key1 = toCacheKey("What is Python?", "openai");
+ * const key2 = toCacheKey("What is Python?", "anthropic");
+ * console.log(key1 === key2); // true
+ */
+export function toCacheKey(query: string, provider?: string): string {
+  return generateCacheKey(query, { provider }).key;
+}