@mnemoai/core 1.1.0 → 1.1.1

package/dist/src/adapters/qdrant.js

@@ -0,0 +1,132 @@
+import { registerAdapter } from "../storage-adapter.js";
+const COLLECTION = "mnemo_memories";
+class QdrantAdapter {
+  name = "qdrant";
+  client = null;
+  url = "http://localhost:6333";
+  apiKey;
+  vectorDim = 0;
+  constructor(config) {
+    if (config?.url) this.url = config.url;
+    if (config?.apiKey) this.apiKey = config.apiKey;
+  }
+  async connect() {
+    const { QdrantClient } = await import("@qdrant/js-client-rest");
+    this.client = new QdrantClient({
+      url: this.url,
+      ...this.apiKey ? { apiKey: this.apiKey } : {}
+    });
+  }
+  async ensureTable(vectorDimensions) {
+    this.vectorDim = vectorDimensions;
+    const collections = await this.client.getCollections();
+    const exists = collections.collections.some((c) => c.name === COLLECTION);
+    if (!exists) {
+      await this.client.createCollection(COLLECTION, {
+        vectors: { size: vectorDimensions, distance: "Cosine" }
+      });
+      await this.client.createPayloadIndex(COLLECTION, {
+        field_name: "scope",
+        field_schema: "keyword"
+      });
+      await this.client.createPayloadIndex(COLLECTION, {
+        field_name: "category",
+        field_schema: "keyword"
+      });
+    }
+  }
+  async add(records) {
+    const points = records.map((r) => ({
+      id: r.id,
+      vector: r.vector,
+      payload: {
+        text: r.text,
+        timestamp: r.timestamp,
+        scope: r.scope,
+        importance: r.importance,
+        category: r.category,
+        metadata: r.metadata
+      }
+    }));
+    await this.client.upsert(COLLECTION, { points });
+  }
+  async update(id, record) {
+    await this.add([record]);
+  }
+  async delete(filter) {
+    const idMatch = filter.match(/id\s*=\s*'([^']+)'/);
+    if (idMatch) {
+      await this.client.delete(COLLECTION, {
+        points: [idMatch[1]]
+      });
+    }
+  }
+  async vectorSearch(vector, limit, minScore = 0, scopeFilter) {
+    const filter = scopeFilter?.length ? { must: [{ key: "scope", match: { any: scopeFilter } }] } : void 0;
+    const results = await this.client.search(COLLECTION, {
+      vector,
+      limit,
+      with_payload: true,
+      score_threshold: minScore,
+      ...filter ? { filter } : {}
+    });
+    return results.map((r) => ({
+      record: this.toRecord(r.id, r.payload),
+      score: r.score
+    }));
+  }
+  async fullTextSearch(_query, _limit, _scopeFilter) {
+    return [];
+  }
+  async query(options) {
+    const filter = options.where ? this.parseFilter(options.where) : void 0;
+    const result = await this.client.scroll(COLLECTION, {
+      limit: options.limit || 100,
+      with_payload: true,
+      with_vectors: true,
+      ...filter ? { filter } : {}
+    });
+    return result.points.map((p) => this.toRecord(p.id, p.payload, p.vector));
+  }
+  async count(filter) {
+    const result = await this.client.count(COLLECTION, {
+      ...filter ? { filter: this.parseFilter(filter) } : {},
+      exact: true
+    });
+    return result.count;
+  }
+  async ensureFullTextIndex() {
+  }
+  hasFullTextSearch() {
+    return false;
+  }
+  async close() {
+    this.client = null;
+  }
+  // ── Helpers ──
+  toRecord(id, payload, vector) {
+    return {
+      id,
+      text: payload.text ?? "",
+      vector: vector ? Array.from(vector) : [],
+      timestamp: payload.timestamp ?? 0,
+      scope: payload.scope ?? "global",
+      importance: payload.importance ?? 0.5,
+      category: payload.category ?? "other",
+      metadata: payload.metadata ?? "{}"
+    };
+  }
+  parseFilter(where) {
+    const scopeMatch = where.match(/scope\s+IN\s*\(([^)]+)\)/i);
+    if (scopeMatch) {
+      const scopes = scopeMatch[1].split(",").map((s) => s.trim().replace(/'/g, ""));
+      return { must: [{ key: "scope", match: { any: scopes } }] };
+    }
+    return void 0;
+  }
+}
+registerAdapter("qdrant", (config) => new QdrantAdapter(config));
+export {
+  QdrantAdapter
+};
+//# sourceMappingURL=qdrant.js.map

package/dist/src/adapters/qdrant.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../../src/adapters/qdrant.ts"],
+  "sourcesContent": ["// SPDX-License-Identifier: MIT\n/**\n * Qdrant Storage Adapter for Mnemo\n *\n * Requirements:\n *   npm install @qdrant/js-client-rest\n *\n * Config:\n *   storage: \"qdrant\"\n *   storageConfig: { url: \"http://localhost:6333\", apiKey?: \"...\" }\n */\n\nimport type {\n  StorageAdapter,\n  MemoryRecord,\n  SearchResult,\n  QueryOptions,\n} from \"../storage-adapter.js\";\nimport { registerAdapter } from \"../storage-adapter.js\";\n\nconst COLLECTION = \"mnemo_memories\";\n\nexport class QdrantAdapter implements StorageAdapter {\n  readonly name = \"qdrant\";\n\n  private client: any = null;\n  private url: string = \"http://localhost:6333\";\n  private apiKey?: string;\n  private vectorDim = 0;\n\n  constructor(config?: Record<string, unknown>) {\n    if (config?.url) this.url = config.url as string;\n    if (config?.apiKey) this.apiKey = config.apiKey as string;\n  }\n\n  async connect(): Promise<void> {\n    const { QdrantClient } = await import(\"@qdrant/js-client-rest\");\n    this.client = new QdrantClient({\n      url: this.url,\n      ...(this.apiKey ? { apiKey: this.apiKey } : {}),\n    });\n  }\n\n  async ensureTable(vectorDimensions: number): Promise<void> {\n    this.vectorDim = vectorDimensions;\n    const collections = await this.client.getCollections();\n    const exists = collections.collections.some((c: any) => c.name === COLLECTION);\n\n    if (!exists) {\n      await this.client.createCollection(COLLECTION, {\n        vectors: { size: vectorDimensions, distance: \"Cosine\" },\n      });\n      // Create payload indices for filtering\n      await this.client.createPayloadIndex(COLLECTION, {\n        field_name: \"scope\",\n        field_schema: \"keyword\",\n      });\n      await this.client.createPayloadIndex(COLLECTION, {\n        field_name: \"category\",\n        field_schema: \"keyword\",\n      });\n    }\n  }\n\n  async add(records: MemoryRecord[]): Promise<void> {\n    const points = records.map((r) => ({\n      id: r.id,\n      vector: r.vector,\n      payload: {\n        text: r.text,\n        timestamp: r.timestamp,\n        scope: r.scope,\n        importance: r.importance,\n        category: r.category,\n        metadata: r.metadata,\n      },\n    }));\n    await this.client.upsert(COLLECTION, { points });\n  }\n\n  async update(id: string, record: MemoryRecord): Promise<void> {\n    await this.add([record]);\n  }\n\n  async delete(filter: string): Promise<void> {\n    // Parse simple \"id = 'xxx'\" filter\n    const idMatch = filter.match(/id\\s*=\\s*'([^']+)'/);\n    if (idMatch) {\n      await this.client.delete(COLLECTION, {\n        points: [idMatch[1]],\n      });\n    }\n  }\n\n  async vectorSearch(\n    vector: number[],\n    limit: number,\n    minScore = 0,\n    scopeFilter?: string[],\n  ): Promise<SearchResult[]> {\n    const filter = scopeFilter?.length\n      ? { must: [{ key: \"scope\", match: { any: scopeFilter } }] }\n      : undefined;\n\n    const results = await this.client.search(COLLECTION, {\n      vector,\n      limit,\n      with_payload: true,\n      score_threshold: minScore,\n      ...(filter ? { filter } : {}),\n    });\n\n    return results.map((r: any) => ({\n      record: this.toRecord(r.id, r.payload),\n      score: r.score,\n    }));\n  }\n\n  async fullTextSearch(\n    _query: string,\n    _limit: number,\n    _scopeFilter?: string[],\n  ): Promise<SearchResult[]> {\n    // Qdrant doesn't have native BM25 \u2014 fall back to empty\n    // Users should pair with a separate FTS engine or use vector search\n    return [];\n  }\n\n  async query(options: QueryOptions): Promise<MemoryRecord[]> {\n    const filter = options.where\n      ? this.parseFilter(options.where)\n      : undefined;\n\n    const result = await this.client.scroll(COLLECTION, {\n      limit: options.limit || 100,\n      with_payload: true,\n      with_vectors: true,\n      ...(filter ? { filter } : {}),\n    });\n\n    return result.points.map((p: any) => this.toRecord(p.id, p.payload, p.vector));\n  }\n\n  async count(filter?: string): Promise<number> {\n    const result = await this.client.count(COLLECTION, {\n      ...(filter ? { filter: this.parseFilter(filter) } : {}),\n      exact: true,\n    });\n    return result.count;\n  }\n\n  async ensureFullTextIndex(): Promise<void> {\n    // Qdrant uses payload indices, not FTS indices\n    // Text search via Qdrant requires external FTS or payload keyword match\n  }\n\n  hasFullTextSearch(): boolean {\n    return false; // Qdrant doesn't have native BM25\n  }\n\n  async close(): Promise<void> {\n    this.client = null;\n  }\n\n  // \u2500\u2500 Helpers \u2500\u2500\n\n  private toRecord(id: string, payload: any, vector?: number[]): MemoryRecord {\n    return {\n      id,\n      text: payload.text ?? \"\",\n      vector: vector ? Array.from(vector) : [],\n      timestamp: payload.timestamp ?? 0,\n      scope: payload.scope ?? \"global\",\n      importance: payload.importance ?? 0.5,\n      category: payload.category ?? \"other\",\n      metadata: payload.metadata ?? \"{}\",\n    };\n  }\n\n  private parseFilter(where: string): any {\n    // Simple parser for common filters\n    const scopeMatch = where.match(/scope\\s+IN\\s*\\(([^)]+)\\)/i);\n    if (scopeMatch) {\n      const scopes = scopeMatch[1].split(\",\").map((s) => s.trim().replace(/'/g, \"\"));\n      return { must: [{ key: \"scope\", match: { any: scopes } }] };\n    }\n    return undefined;\n  }\n}\n\nregisterAdapter(\"qdrant\", (config) => new QdrantAdapter(config));\n"],
+  "mappings": "AAkBA,SAAS,uBAAuB;AAEhC,MAAM,aAAa;AAEZ,MAAM,cAAwC;AAAA,EAC1C,OAAO;AAAA,EAER,SAAc;AAAA,EACd,MAAc;AAAA,EACd;AAAA,EACA,YAAY;AAAA,EAEpB,YAAY,QAAkC;AAC5C,QAAI,QAAQ,IAAK,MAAK,MAAM,OAAO;AACnC,QAAI,QAAQ,OAAQ,MAAK,SAAS,OAAO;AAAA,EAC3C;AAAA,EAEA,MAAM,UAAyB;AAC7B,UAAM,EAAE,aAAa,IAAI,MAAM,OAAO,wBAAwB;AAC9D,SAAK,SAAS,IAAI,aAAa;AAAA,MAC7B,KAAK,KAAK;AAAA,MACV,GAAI,KAAK,SAAS,EAAE,QAAQ,KAAK,OAAO,IAAI,CAAC;AAAA,IAC/C,CAAC;AAAA,EACH;AAAA,EAEA,MAAM,YAAY,kBAAyC;AACzD,SAAK,YAAY;AACjB,UAAM,cAAc,MAAM,KAAK,OAAO,eAAe;AACrD,UAAM,SAAS,YAAY,YAAY,KAAK,CAAC,MAAW,EAAE,SAAS,UAAU;AAE7E,QAAI,CAAC,QAAQ;AACX,YAAM,KAAK,OAAO,iBAAiB,YAAY;AAAA,QAC7C,SAAS,EAAE,MAAM,kBAAkB,UAAU,SAAS;AAAA,MACxD,CAAC;AAED,YAAM,KAAK,OAAO,mBAAmB,YAAY;AAAA,QAC/C,YAAY;AAAA,QACZ,cAAc;AAAA,MAChB,CAAC;AACD,YAAM,KAAK,OAAO,mBAAmB,YAAY;AAAA,QAC/C,YAAY;AAAA,QACZ,cAAc;AAAA,MAChB,CAAC;AAAA,IACH;AAAA,EACF;AAAA,EAEA,MAAM,IAAI,SAAwC;AAChD,UAAM,SAAS,QAAQ,IAAI,CAAC,OAAO;AAAA,MACjC,IAAI,EAAE;AAAA,MACN,QAAQ,EAAE;AAAA,MACV,SAAS;AAAA,QACP,MAAM,EAAE;AAAA,QACR,WAAW,EAAE;AAAA,QACb,OAAO,EAAE;AAAA,QACT,YAAY,EAAE;AAAA,QACd,UAAU,EAAE;AAAA,QACZ,UAAU,EAAE;AAAA,MACd;AAAA,IACF,EAAE;AACF,UAAM,KAAK,OAAO,OAAO,YAAY,EAAE,OAAO,CAAC;AAAA,EACjD;AAAA,EAEA,MAAM,OAAO,IAAY,QAAqC;AAC5D,UAAM,KAAK,IAAI,CAAC,MAAM,CAAC;AAAA,EACzB;AAAA,EAEA,MAAM,OAAO,QAA+B;AAE1C,UAAM,UAAU,OAAO,MAAM,oBAAoB;AACjD,QAAI,SAAS;AACX,YAAM,KAAK,OAAO,OAAO,YAAY;AAAA,QACnC,QAAQ,CAAC,QAAQ,CAAC,CAAC;AAAA,MACrB,CAAC;AAAA,IACH;AAAA,EACF;AAAA,EAEA,MAAM,aACJ,QACA,OACA,WAAW,GACX,aACyB;AACzB,UAAM,SAAS,aAAa,SACxB,EAAE,MAAM,CAAC,EAAE,KAAK,SAAS,OAAO,EAAE,KAAK,YAAY,EAAE,CAAC,EAAE,IACxD;AAEJ,UAAM,UAAU,MAAM,KAAK,OAAO,OAAO,YAAY;AAAA,MACnD;AAAA,MACA;AAAA,MACA,cAAc;AAAA,MACd,iBAAiB;AAAA,MACjB,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;AAAA,IAC7B,CAAC;AAED,WAAO,QAAQ,IAAI,CAAC,OAAY;AAAA,MAC9B,QAAQ,KAAK,SAAS,EAAE,IAAI,EAAE,OAAO;AAAA,MACrC,OAAO,EAAE;AAAA,IACX,EAAE;AAAA,EACJ;AAAA,EAEA,MAAM,eACJ,QACA,QACA,cACyB;AAGzB,WAAO,CAAC;AAAA,EACV;AAAA,EAEA,MAAM,MAAM,SAAgD;AAC1D,UAAM,SAAS,QAAQ,QACnB,KAAK,YAAY,QAAQ,KAAK,IAC9B;AAEJ,UAAM,SAAS,MAAM,KAAK,OAAO,OAAO,YAAY;AAAA,MAClD,OAAO,QAAQ,SAAS;AAAA,MACxB,cAAc;AAAA,MACd,cAAc;AAAA,MACd,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;AAAA,IAC7B,CAAC;AAED,WAAO,OAAO,OAAO,IAAI,CAAC,MAAW,KAAK,SAAS,EAAE,IAAI,EAAE,SAAS,EAAE,MAAM,CAAC;AAAA,EAC/E;AAAA,EAEA,MAAM,MAAM,QAAkC;AAC5C,UAAM,SAAS,MAAM,KAAK,OAAO,MAAM,YAAY;AAAA,MACjD,GAAI,SAAS,EAAE,QAAQ,KAAK,YAAY,MAAM,EAAE,IAAI,CAAC;AAAA,MACrD,OAAO;AAAA,IACT,CAAC;AACD,WAAO,OAAO;AAAA,EAChB;AAAA,EAEA,MAAM,sBAAqC;AAAA,EAG3C;AAAA,EAEA,oBAA6B;AAC3B,WAAO;AAAA,EACT;AAAA,EAEA,MAAM,QAAuB;AAC3B,SAAK,SAAS;AAAA,EAChB;AAAA;AAAA,EAIQ,SAAS,IAAY,SAAc,QAAiC;AAC1E,WAAO;AAAA,MACL;AAAA,MACA,MAAM,QAAQ,QAAQ;AAAA,MACtB,QAAQ,SAAS,MAAM,KAAK,MAAM,IAAI,CAAC;AAAA,MACvC,WAAW,QAAQ,aAAa;AAAA,MAChC,OAAO,QAAQ,SAAS;AAAA,MACxB,YAAY,QAAQ,cAAc;AAAA,MAClC,UAAU,QAAQ,YAAY;AAAA,MAC9B,UAAU,QAAQ,YAAY;AAAA,IAChC;AAAA,EACF;AAAA,EAEQ,YAAY,OAAoB;AAEtC,UAAM,aAAa,MAAM,MAAM,2BAA2B;AAC1D,QAAI,YAAY;AACd,YAAM,SAAS,WAAW,CAAC,EAAE,MAAM,GAAG,EAAE,IAAI,CAAC,MAAM,EAAE,KAAK,EAAE,QAAQ,MAAM,EAAE,CAAC;AAC7E,aAAO,EAAE,MAAM,CAAC,EAAE,KAAK,SAAS,OAAO,EAAE,KAAK,OAAO,EAAE,CAAC,EAAE;AAAA,IAC5D;AACA,WAAO;AAAA,EACT;AACF;AAEA,gBAAgB,UAAU,CAAC,WAAW,IAAI,cAAc,MAAM,CAAC;",
+  "names": []
+}

package/dist/src/adaptive-retrieval.d.ts

@@ -0,0 +1,14 @@
+/**
+ * Adaptive Retrieval
+ * Determines whether a query needs memory retrieval at all.
+ * Skips retrieval for greetings, commands, simple instructions, and system messages.
+ * Saves embedding API calls and reduces noise injection.
+ */
+/**
+ * Determine if a query should skip memory retrieval.
+ * Returns true if retrieval should be skipped.
+ * @param query The raw prompt text
+ * @param minLength Optional minimum length override (if set, overrides built-in thresholds)
+ */
+export declare function shouldSkipRetrieval(query: string, minLength?: number): boolean;
+//# sourceMappingURL=adaptive-retrieval.d.ts.map

package/dist/src/adaptive-retrieval.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"adaptive-retrieval.d.ts","sourceRoot":"","sources":["../../src/adaptive-retrieval.ts"],"names":[],"mappings":"AACA;;;;;GAKG;AAkDH;;;;;GAKG;AACH,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,OAAO,CA2B9E"}

package/dist/src/adaptive-retrieval.js

@@ -0,0 +1,52 @@
+const SKIP_PATTERNS = [
+  // Greetings & pleasantries
+  /^(hi|hello|hey|good\s*(morning|afternoon|evening|night)|greetings|yo|sup|howdy|what'?s up)\b/i,
+  // System/bot commands
+  /^\//,
+  // slash commands
+  /^(run|build|test|ls|cd|git|npm|pip|docker|curl|cat|grep|find|make|sudo)\b/i,
+  // Simple affirmations/negations
+  /^(yes|no|yep|nope|ok|okay|sure|fine|thanks|thank you|thx|ty|got it|understood|cool|nice|great|good|perfect|awesome)\s*[.!]?$/i,
+  // Continuation prompts
+  /^(go ahead|continue|proceed|do it|start|begin|next)\s*[.!]?$/i,
+  // Pure emoji
+  /^[\p{Emoji}\s]+$/u,
+  // Heartbeat/system (match anywhere, not just at start, to handle prefixed formats)
+  /HEARTBEAT/i,
+  /^\[System/i,
+  // Single-word utility pings
+  /^(ping|pong|test|debug)\s*[.!?]?$/i
+];
+const FORCE_RETRIEVE_PATTERNS = [
+  /\b(remember|recall|forgot|memory|memories)\b/i,
+  /\b(last time|before|previously|earlier|yesterday|ago)\b/i,
+  /\b(my (name|email|phone|address|birthday|preference))\b/i,
+  /\b(what did (i|we)|did i (tell|say|mention))\b/i
+];
+function normalizeQuery(query) {
+  let s = query.trim();
+  const metadataPattern = /^(Conversation info|Sender) \(untrusted metadata\):[\s\S]*?\n\s*\n/gim;
+  s = s.replace(metadataPattern, "");
+  s = s.trim().replace(/^\[cron:[^\]]+\]\s*/i, "");
+  s = s.trim().replace(/^\[[A-Za-z]{3}\s\d{4}-\d{2}-\d{2}\s\d{2}:\d{2}\s[^\]]+\]\s*/, "");
+  const result = s.trim();
+  return result;
+}
+function shouldSkipRetrieval(query, minLength) {
+  const trimmed = normalizeQuery(query);
+  if (FORCE_RETRIEVE_PATTERNS.some((p) => p.test(trimmed))) return false;
+  if (trimmed.length < 5) return true;
+  if (SKIP_PATTERNS.some((p) => p.test(trimmed))) return true;
+  if (minLength !== void 0 && minLength > 0) {
+    if (trimmed.length < minLength && !trimmed.includes("?")) return true;
+    return false;
+  }
+  const hasCJK = /[\u4e00-\u9fff\u3040-\u309f\u30a0-\u30ff\uac00-\ud7af]/.test(trimmed);
+  const defaultMinLength = hasCJK ? 6 : 15;
+  if (trimmed.length < defaultMinLength && !trimmed.includes("?")) return true;
+  return false;
+}
+export {
+  shouldSkipRetrieval
+};
+//# sourceMappingURL=adaptive-retrieval.js.map

package/dist/src/adaptive-retrieval.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../src/adaptive-retrieval.ts"],
+  "sourcesContent": ["// SPDX-License-Identifier: MIT\n/**\n * Adaptive Retrieval\n * Determines whether a query needs memory retrieval at all.\n * Skips retrieval for greetings, commands, simple instructions, and system messages.\n * Saves embedding API calls and reduces noise injection.\n */\n\n// Queries that are clearly NOT memory-retrieval candidates\nconst SKIP_PATTERNS = [\n  // Greetings & pleasantries\n  /^(hi|hello|hey|good\\s*(morning|afternoon|evening|night)|greetings|yo|sup|howdy|what'?s up)\\b/i,\n  // System/bot commands\n  /^\\//,  // slash commands\n  /^(run|build|test|ls|cd|git|npm|pip|docker|curl|cat|grep|find|make|sudo)\\b/i,\n  // Simple affirmations/negations\n  /^(yes|no|yep|nope|ok|okay|sure|fine|thanks|thank you|thx|ty|got it|understood|cool|nice|great|good|perfect|awesome)\\s*[.!]?$/i,\n  // Continuation prompts\n  /^(go ahead|continue|proceed|do it|start|begin|next)\\s*[.!]?$/i,\n  // Pure emoji\n  /^[\\p{Emoji}\\s]+$/u,\n  // Heartbeat/system (match anywhere, not just at start, to handle prefixed formats)\n  /HEARTBEAT/i,\n  /^\\[System/i,\n  // Single-word utility pings\n  /^(ping|pong|test|debug)\\s*[.!?]?$/i,\n];\n\n// Queries that SHOULD trigger retrieval even if short\nconst FORCE_RETRIEVE_PATTERNS = [\n  /\\b(remember|recall|forgot|memory|memories)\\b/i,\n  /\\b(last time|before|previously|earlier|yesterday|ago)\\b/i,\n  /\\b(my (name|email|phone|address|birthday|preference))\\b/i,\n  /\\b(what did (i|we)|did i (tell|say|mention))\\b/i,\n];\n\n/**\n * Normalize the raw prompt before applying skip/force rules.\n */\nfunction normalizeQuery(query: string): string {\n  let s = query.trim();\n\n  // 1. Strip injected metadata headers.\n  const metadataPattern = /^(Conversation info|Sender) \\(untrusted metadata\\):[\\s\\S]*?\\n\\s*\\n/gim;\n  s = s.replace(metadataPattern, \"\");\n\n  // 2. Strip cron wrapper prefix.\n  s = s.trim().replace(/^\\[cron:[^\\]]+\\]\\s*/i, \"\");\n\n  // 3. Strip timestamp prefix [Mon 2026-03-02 04:21 GMT+8].\n  s = s.trim().replace(/^\\[[A-Za-z]{3}\\s\\d{4}-\\d{2}-\\d{2}\\s\\d{2}:\\d{2}\\s[^\\]]+\\]\\s*/, \"\");\n\n  const result = s.trim();\n  return result;\n}\n\n/**\n * Determine if a query should skip memory retrieval.\n * Returns true if retrieval should be skipped.\n * @param query The raw prompt text\n * @param minLength Optional minimum length override (if set, overrides built-in thresholds)\n */\nexport function shouldSkipRetrieval(query: string, minLength?: number): boolean {\n  const trimmed = normalizeQuery(query);\n\n  // Force retrieve if query has memory-related intent (checked FIRST,\n  // before length check, so short CJK queries aren't skipped)\n  if (FORCE_RETRIEVE_PATTERNS.some(p => p.test(trimmed))) return false;\n\n  // Too short to be meaningful\n  if (trimmed.length < 5) return true;\n\n  // Skip if matches any skip pattern\n  if (SKIP_PATTERNS.some(p => p.test(trimmed))) return true;\n\n  // If caller provides a custom minimum length, use it\n  if (minLength !== undefined && minLength > 0) {\n    if (trimmed.length < minLength && !trimmed.includes('?')) return true;\n    return false;\n  }\n\n  // Skip very short non-question messages (likely commands or affirmations)\n  // CJK characters carry more meaning per character, so use a lower threshold\n  const hasCJK = /[\\u4e00-\\u9fff\\u3040-\\u309f\\u30a0-\\u30ff\\uac00-\\ud7af]/.test(trimmed);\n  const defaultMinLength = hasCJK ? 6 : 15;\n  if (trimmed.length < defaultMinLength && !trimmed.includes('?')) return true;\n\n  // Default: do retrieve\n  return false;\n}\n"],
+  "mappings": "AASA,MAAM,gBAAgB;AAAA;AAAA,EAEpB;AAAA;AAAA,EAEA;AAAA;AAAA,EACA;AAAA;AAAA,EAEA;AAAA;AAAA,EAEA;AAAA;AAAA,EAEA;AAAA;AAAA,EAEA;AAAA,EACA;AAAA;AAAA,EAEA;AACF;AAGA,MAAM,0BAA0B;AAAA,EAC9B;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAKA,SAAS,eAAe,OAAuB;AAC7C,MAAI,IAAI,MAAM,KAAK;AAGnB,QAAM,kBAAkB;AACxB,MAAI,EAAE,QAAQ,iBAAiB,EAAE;AAGjC,MAAI,EAAE,KAAK,EAAE,QAAQ,wBAAwB,EAAE;AAG/C,MAAI,EAAE,KAAK,EAAE,QAAQ,+DAA+D,EAAE;AAEtF,QAAM,SAAS,EAAE,KAAK;AACtB,SAAO;AACT;AAQO,SAAS,oBAAoB,OAAe,WAA6B;AAC9E,QAAM,UAAU,eAAe,KAAK;AAIpC,MAAI,wBAAwB,KAAK,OAAK,EAAE,KAAK,OAAO,CAAC,EAAG,QAAO;AAG/D,MAAI,QAAQ,SAAS,EAAG,QAAO;AAG/B,MAAI,cAAc,KAAK,OAAK,EAAE,KAAK,OAAO,CAAC,EAAG,QAAO;AAGrD,MAAI,cAAc,UAAa,YAAY,GAAG;AAC5C,QAAI,QAAQ,SAAS,aAAa,CAAC,QAAQ,SAAS,GAAG,EAAG,QAAO;AACjE,WAAO;AAAA,EACT;AAIA,QAAM,SAAS,yDAAyD,KAAK,OAAO;AACpF,QAAM,mBAAmB,SAAS,IAAI;AACtC,MAAI,QAAQ,SAAS,oBAAoB,CAAC,QAAQ,SAAS,GAAG,EAAG,QAAO;AAGxE,SAAO;AACT;",
+  "names": []
+}

package/dist/src/audit-log.d.ts

@@ -0,0 +1,56 @@
+/**
+ * Mnemo Audit Log — GDPR/EU AI Act compliance
+ *
+ * Records all memory CRUD operations with:
+ * - WHO: agent/user identity
+ * - WHAT: operation type + affected memory IDs
+ * - WHEN: ISO timestamp
+ * - WHY: source/trigger (auto-capture, manual, contradiction, etc.)
+ *
+ * Stored as append-only JSONL file. Supports retention policies.
+ */
+export type AuditAction = "create" | "update" | "delete" | "bulk_delete" | "expire" | "merge" | "recall" | "export";
+export interface AuditEntry {
+    timestamp: string;
+    action: AuditAction;
+    actor: string;
+    memoryIds: string[];
+    scope?: string;
+    reason?: string;
+    details?: string;
+    ip?: string;
+}
+/**
+ * Append an audit entry. Fire-and-forget — never blocks the main flow.
+ */
+export declare function audit(entry: AuditEntry): Promise<void>;
+/**
+ * Convenience: log a memory creation.
+ */
+export declare function auditCreate(memoryId: string, actor: string, scope: string, reason: string, textPreview?: string): void;
+/**
+ * Convenience: log a memory deletion.
+ */
+export declare function auditDelete(memoryIds: string[], actor: string, reason: string): void;
+/**
+ * Convenience: log a memory update (e.g., importance change, tier change).
+ */
+export declare function auditUpdate(memoryId: string, actor: string, reason: string, details?: string): void;
+/**
+ * Convenience: log a memory expiration (contradiction resolution).
+ */
+export declare function auditExpire(memoryId: string, actor: string, reason: string, details?: string): void;
+/**
+ * Convenience: log a memory recall (for access audit trail).
+ */
+export declare function auditRecall(memoryIds: string[], actor: string, query?: string): void;
+/**
+ * Read audit log entries for a date range.
+ * Useful for compliance exports.
+ */
+export declare function readAuditLog(startDate: string, endDate: string): Promise<AuditEntry[]>;
+/**
+ * Enable or disable audit logging.
+ */
+export declare function setAuditEnabled(enabled: boolean): void;
+//# sourceMappingURL=audit-log.d.ts.map

package/dist/src/audit-log.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"audit-log.d.ts","sourceRoot":"","sources":["../../src/audit-log.ts"],"names":[],"mappings":"AACA;;;;;;;;;;GAUG;AASH,MAAM,MAAM,WAAW,GACnB,QAAQ,GACR,QAAQ,GACR,QAAQ,GACR,aAAa,GACb,QAAQ,GACR,OAAO,GACP,QAAQ,GACR,QAAQ,CAAC;AAEb,MAAM,WAAW,UAAU;IACzB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,WAAW,CAAC;IACpB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,EAAE,CAAC,EAAE,MAAM,CAAC;CACb;AAyBD;;GAEG;AACH,wBAAsB,KAAK,CAAC,KAAK,EAAE,UAAU,GAAG,OAAO,CAAC,IAAI,CAAC,CAgC5D;AAED;;GAEG;AACH,wBAAgB,WAAW,CACzB,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,MAAM,EACb,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,EACd,WAAW,CAAC,EAAE,MAAM,GACnB,IAAI,CAUN;AAED;;GAEG;AACH,wBAAgB,WAAW,CACzB,SAAS,EAAE,MAAM,EAAE,EACnB,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,GACb,IAAI,CAQN;AAED;;GAEG;AACH,wBAAgB,WAAW,CACzB,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,EACd,OAAO,CAAC,EAAE,MAAM,GACf,IAAI,CASN;AAED;;GAEG;AACH,wBAAgB,WAAW,CACzB,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,EACd,OAAO,CAAC,EAAE,MAAM,GACf,IAAI,CASN;AAED;;GAEG;AACH,wBAAgB,WAAW,CACzB,SAAS,EAAE,MAAM,EAAE,EACnB,KAAK,EAAE,MAAM,EACb,KAAK,CAAC,EAAE,MAAM,GACb,IAAI,CASN;AAED;;;GAGG;AACH,wBAAsB,YAAY,CAChC,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,MAAM,GACd,OAAO,CAAC,UAAU,EAAE,CAAC,CA4BvB;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,OAAO,EAAE,OAAO,GAAG,IAAI,CAEtD"}

package/dist/src/audit-log.js

@@ -0,0 +1,139 @@
+import { appendFile, mkdir, readFile, stat } from "node:fs/promises";
+import { join } from "node:path";
+import { homedir } from "node:os";
+const AUDIT_DIR = join(homedir(), ".mnemo", "audit");
+const MAX_FILE_SIZE = 10 * 1024 * 1024;
+let _initialized = false;
+let _currentFile = "";
+let _enabled = true;
+async function ensureDir() {
+  if (_initialized) return;
+  try {
+    await mkdir(AUDIT_DIR, { recursive: true });
+    _currentFile = getLogFileName();
+    _initialized = true;
+  } catch {
+    _enabled = false;
+  }
+}
+function getLogFileName() {
+  const date = (/* @__PURE__ */ new Date()).toISOString().slice(0, 10);
+  return join(AUDIT_DIR, `audit-${date}.jsonl`);
+}
+async function audit(entry) {
+  if (!_enabled) return;
+  try {
+    await ensureDir();
+    const expectedFile = getLogFileName();
+    if (expectedFile !== _currentFile) {
+      _currentFile = expectedFile;
+    }
+    try {
+      const stats = await stat(_currentFile);
+      if (stats.size > MAX_FILE_SIZE) {
+        const rotatedName = _currentFile.replace(".jsonl", `-${Date.now()}.jsonl`);
+        _currentFile = rotatedName;
+      }
+    } catch {
+    }
+    const line = JSON.stringify({
+      ...entry,
+      timestamp: entry.timestamp || (/* @__PURE__ */ new Date()).toISOString()
+    }) + "\n";
+    await appendFile(_currentFile, line);
+  } catch {
+  }
+}
+function auditCreate(memoryId, actor, scope, reason, textPreview) {
+  audit({
+    timestamp: (/* @__PURE__ */ new Date()).toISOString(),
+    action: "create",
+    actor,
+    memoryIds: [memoryId],
+    scope,
+    reason,
+    details: textPreview ? textPreview.slice(0, 200) : void 0
+  }).catch(() => {
+  });
+}
+function auditDelete(memoryIds, actor, reason) {
+  audit({
+    timestamp: (/* @__PURE__ */ new Date()).toISOString(),
+    action: memoryIds.length > 1 ? "bulk_delete" : "delete",
+    actor,
+    memoryIds,
+    reason
+  }).catch(() => {
+  });
+}
+function auditUpdate(memoryId, actor, reason, details) {
+  audit({
+    timestamp: (/* @__PURE__ */ new Date()).toISOString(),
+    action: "update",
+    actor,
+    memoryIds: [memoryId],
+    reason,
+    details
+  }).catch(() => {
+  });
+}
+function auditExpire(memoryId, actor, reason, details) {
+  audit({
+    timestamp: (/* @__PURE__ */ new Date()).toISOString(),
+    action: "expire",
+    actor,
+    memoryIds: [memoryId],
+    reason,
+    details
+  }).catch(() => {
+  });
+}
+function auditRecall(memoryIds, actor, query) {
+  audit({
+    timestamp: (/* @__PURE__ */ new Date()).toISOString(),
+    action: "recall",
+    actor,
+    memoryIds,
+    reason: "retrieval",
+    details: query ? query.slice(0, 200) : void 0
+  }).catch(() => {
+  });
+}
+async function readAuditLog(startDate, endDate) {
+  await ensureDir();
+  const entries = [];
+  const start = new Date(startDate);
+  const end = new Date(endDate);
+  const current = new Date(start);
+  while (current <= end) {
+    const dateStr = current.toISOString().slice(0, 10);
+    const filePath = join(AUDIT_DIR, `audit-${dateStr}.jsonl`);
+    try {
+      const content = await readFile(filePath, "utf8");
+      const lines = content.trim().split("\n").filter(Boolean);
+      for (const line of lines) {
+        try {
+          entries.push(JSON.parse(line));
+        } catch {
+        }
+      }
+    } catch {
+    }
+    current.setDate(current.getDate() + 1);
+  }
+  return entries;
+}
+function setAuditEnabled(enabled) {
+  _enabled = enabled;
+}
+export {
+  audit,
+  auditCreate,
+  auditDelete,
+  auditExpire,
+  auditRecall,
+  auditUpdate,
+  readAuditLog,
+  setAuditEnabled
+};
+//# sourceMappingURL=audit-log.js.map

package/dist/src/audit-log.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../src/audit-log.ts"],
+  "sourcesContent": ["// SPDX-License-Identifier: LicenseRef-Mnemo-Pro\n/**\n * Mnemo Audit Log \u2014 GDPR/EU AI Act compliance\n *\n * Records all memory CRUD operations with:\n * - WHO: agent/user identity\n * - WHAT: operation type + affected memory IDs\n * - WHEN: ISO timestamp\n * - WHY: source/trigger (auto-capture, manual, contradiction, etc.)\n *\n * Stored as append-only JSONL file. Supports retention policies.\n */\n\nimport { appendFile, mkdir, readFile, stat } from \"node:fs/promises\";\nimport { join } from \"node:path\";\nimport { homedir } from \"node:os\";\n\nconst AUDIT_DIR = join(homedir(), \".mnemo\", \"audit\");\nconst MAX_FILE_SIZE = 10 * 1024 * 1024; // 10MB per file, then rotate\n\nexport type AuditAction =\n  | \"create\"\n  | \"update\"\n  | \"delete\"\n  | \"bulk_delete\"\n  | \"expire\"\n  | \"merge\"\n  | \"recall\"\n  | \"export\";\n\nexport interface AuditEntry {\n  timestamp: string;\n  action: AuditAction;\n  actor: string;          // agent ID, user ID, or \"system\"\n  memoryIds: string[];    // affected memory IDs\n  scope?: string;\n  reason?: string;        // \"auto-capture\", \"contradiction\", \"user-request\", \"decay\", etc.\n  details?: string;       // additional context (text preview, old\u2192new value, etc.)\n  ip?: string;            // for API-based access\n}\n\nlet _initialized = false;\nlet _currentFile = \"\";\nlet _enabled = true;\n\n/**\n * Initialize the audit log directory.\n */\nasync function ensureDir(): Promise<void> {\n  if (_initialized) return;\n  try {\n    await mkdir(AUDIT_DIR, { recursive: true });\n    _currentFile = getLogFileName();\n    _initialized = true;\n  } catch {\n    _enabled = false;\n  }\n}\n\nfunction getLogFileName(): string {\n  const date = new Date().toISOString().slice(0, 10); // YYYY-MM-DD\n  return join(AUDIT_DIR, `audit-${date}.jsonl`);\n}\n\n/**\n * Append an audit entry. Fire-and-forget \u2014 never blocks the main flow.\n */\nexport async function audit(entry: AuditEntry): Promise<void> {\n  if (!_enabled) return;\n\n  try {\n    await ensureDir();\n\n    // Rotate file daily\n    const expectedFile = getLogFileName();\n    if (expectedFile !== _currentFile) {\n      _currentFile = expectedFile;\n    }\n\n    // Check file size for rotation\n    try {\n      const stats = await stat(_currentFile);\n      if (stats.size > MAX_FILE_SIZE) {\n        const rotatedName = _currentFile.replace(\".jsonl\", `-${Date.now()}.jsonl`);\n        _currentFile = rotatedName;\n      }\n    } catch {\n      // File doesn't exist yet, that's fine\n    }\n\n    const line = JSON.stringify({\n      ...entry,\n      timestamp: entry.timestamp || new Date().toISOString(),\n    }) + \"\\n\";\n\n    await appendFile(_currentFile, line);\n  } catch {\n    // Audit log failure should never break the main flow\n  }\n}\n\n/**\n * Convenience: log a memory creation.\n */\nexport function auditCreate(\n  memoryId: string,\n  actor: string,\n  scope: string,\n  reason: string,\n  textPreview?: string,\n): void {\n  audit({\n    timestamp: new Date().toISOString(),\n    action: \"create\",\n    actor,\n    memoryIds: [memoryId],\n    scope,\n    reason,\n    details: textPreview ? textPreview.slice(0, 200) : undefined,\n  }).catch(() => {});\n}\n\n/**\n * Convenience: log a memory deletion.\n */\nexport function auditDelete(\n  memoryIds: string[],\n  actor: string,\n  reason: string,\n): void {\n  audit({\n    timestamp: new Date().toISOString(),\n    action: memoryIds.length > 1 ? \"bulk_delete\" : \"delete\",\n    actor,\n    memoryIds,\n    reason,\n  }).catch(() => {});\n}\n\n/**\n * Convenience: log a memory update (e.g., importance change, tier change).\n */\nexport function auditUpdate(\n  memoryId: string,\n  actor: string,\n  reason: string,\n  details?: string,\n): void {\n  audit({\n    timestamp: new Date().toISOString(),\n    action: \"update\",\n    actor,\n    memoryIds: [memoryId],\n    reason,\n    details,\n  }).catch(() => {});\n}\n\n/**\n * Convenience: log a memory expiration (contradiction resolution).\n */\nexport function auditExpire(\n  memoryId: string,\n  actor: string,\n  reason: string,\n  details?: string,\n): void {\n  audit({\n    timestamp: new Date().toISOString(),\n    action: \"expire\",\n    actor,\n    memoryIds: [memoryId],\n    reason,\n    details,\n  }).catch(() => {});\n}\n\n/**\n * Convenience: log a memory recall (for access audit trail).\n */\nexport function auditRecall(\n  memoryIds: string[],\n  actor: string,\n  query?: string,\n): void {\n  audit({\n    timestamp: new Date().toISOString(),\n    action: \"recall\",\n    actor,\n    memoryIds,\n    reason: \"retrieval\",\n    details: query ? query.slice(0, 200) : undefined,\n  }).catch(() => {});\n}\n\n/**\n * Read audit log entries for a date range.\n * Useful for compliance exports.\n */\nexport async function readAuditLog(\n  startDate: string,\n  endDate: string,\n): Promise<AuditEntry[]> {\n  await ensureDir();\n  const entries: AuditEntry[] = [];\n\n  const start = new Date(startDate);\n  const end = new Date(endDate);\n  const current = new Date(start);\n\n  while (current <= end) {\n    const dateStr = current.toISOString().slice(0, 10);\n    const filePath = join(AUDIT_DIR, `audit-${dateStr}.jsonl`);\n\n    try {\n      const content = await readFile(filePath, \"utf8\");\n      const lines = content.trim().split(\"\\n\").filter(Boolean);\n      for (const line of lines) {\n        try {\n          entries.push(JSON.parse(line));\n        } catch { /* skip malformed */ }\n      }\n    } catch {\n      // File doesn't exist for this date, skip\n    }\n\n    current.setDate(current.getDate() + 1);\n  }\n\n  return entries;\n}\n\n/**\n * Enable or disable audit logging.\n */\nexport function setAuditEnabled(enabled: boolean): void {\n  _enabled = enabled;\n}\n"],
+  "mappings": "AAaA,SAAS,YAAY,OAAO,UAAU,YAAY;AAClD,SAAS,YAAY;AACrB,SAAS,eAAe;AAExB,MAAM,YAAY,KAAK,QAAQ,GAAG,UAAU,OAAO;AACnD,MAAM,gBAAgB,KAAK,OAAO;AAuBlC,IAAI,eAAe;AACnB,IAAI,eAAe;AACnB,IAAI,WAAW;AAKf,eAAe,YAA2B;AACxC,MAAI,aAAc;AAClB,MAAI;AACF,UAAM,MAAM,WAAW,EAAE,WAAW,KAAK,CAAC;AAC1C,mBAAe,eAAe;AAC9B,mBAAe;AAAA,EACjB,QAAQ;AACN,eAAW;AAAA,EACb;AACF;AAEA,SAAS,iBAAyB;AAChC,QAAM,QAAO,oBAAI,KAAK,GAAE,YAAY,EAAE,MAAM,GAAG,EAAE;AACjD,SAAO,KAAK,WAAW,SAAS,IAAI,QAAQ;AAC9C;AAKA,eAAsB,MAAM,OAAkC;AAC5D,MAAI,CAAC,SAAU;AAEf,MAAI;AACF,UAAM,UAAU;AAGhB,UAAM,eAAe,eAAe;AACpC,QAAI,iBAAiB,cAAc;AACjC,qBAAe;AAAA,IACjB;AAGA,QAAI;AACF,YAAM,QAAQ,MAAM,KAAK,YAAY;AACrC,UAAI,MAAM,OAAO,eAAe;AAC9B,cAAM,cAAc,aAAa,QAAQ,UAAU,IAAI,KAAK,IAAI,CAAC,QAAQ;AACzE,uBAAe;AAAA,MACjB;AAAA,IACF,QAAQ;AAAA,IAER;AAEA,UAAM,OAAO,KAAK,UAAU;AAAA,MAC1B,GAAG;AAAA,MACH,WAAW,MAAM,cAAa,oBAAI,KAAK,GAAE,YAAY;AAAA,IACvD,CAAC,IAAI;AAEL,UAAM,WAAW,cAAc,IAAI;AAAA,EACrC,QAAQ;AAAA,EAER;AACF;AAKO,SAAS,YACd,UACA,OACA,OACA,QACA,aACM;AACN,QAAM;AAAA,IACJ,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC,QAAQ;AAAA,IACR;AAAA,IACA,WAAW,CAAC,QAAQ;AAAA,IACpB;AAAA,IACA;AAAA,IACA,SAAS,cAAc,YAAY,MAAM,GAAG,GAAG,IAAI;AAAA,EACrD,CAAC,EAAE,MAAM,MAAM;AAAA,EAAC,CAAC;AACnB;AAKO,SAAS,YACd,WACA,OACA,QACM;AACN,QAAM;AAAA,IACJ,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC,QAAQ,UAAU,SAAS,IAAI,gBAAgB;AAAA,IAC/C;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC,EAAE,MAAM,MAAM;AAAA,EAAC,CAAC;AACnB;AAKO,SAAS,YACd,UACA,OACA,QACA,SACM;AACN,QAAM;AAAA,IACJ,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC,QAAQ;AAAA,IACR;AAAA,IACA,WAAW,CAAC,QAAQ;AAAA,IACpB;AAAA,IACA;AAAA,EACF,CAAC,EAAE,MAAM,MAAM;AAAA,EAAC,CAAC;AACnB;AAKO,SAAS,YACd,UACA,OACA,QACA,SACM;AACN,QAAM;AAAA,IACJ,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC,QAAQ;AAAA,IACR;AAAA,IACA,WAAW,CAAC,QAAQ;AAAA,IACpB;AAAA,IACA;AAAA,EACF,CAAC,EAAE,MAAM,MAAM;AAAA,EAAC,CAAC;AACnB;AAKO,SAAS,YACd,WACA,OACA,OACM;AACN,QAAM;AAAA,IACJ,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC,QAAQ;AAAA,IACR;AAAA,IACA;AAAA,IACA,QAAQ;AAAA,IACR,SAAS,QAAQ,MAAM,MAAM,GAAG,GAAG,IAAI;AAAA,EACzC,CAAC,EAAE,MAAM,MAAM;AAAA,EAAC,CAAC;AACnB;AAMA,eAAsB,aACpB,WACA,SACuB;AACvB,QAAM,UAAU;AAChB,QAAM,UAAwB,CAAC;AAE/B,QAAM,QAAQ,IAAI,KAAK,SAAS;AAChC,QAAM,MAAM,IAAI,KAAK,OAAO;AAC5B,QAAM,UAAU,IAAI,KAAK,KAAK;AAE9B,SAAO,WAAW,KAAK;AACrB,UAAM,UAAU,QAAQ,YAAY,EAAE,MAAM,GAAG,EAAE;AACjD,UAAM,WAAW,KAAK,WAAW,SAAS,OAAO,QAAQ;AAEzD,QAAI;AACF,YAAM,UAAU,MAAM,SAAS,UAAU,MAAM;AAC/C,YAAM,QAAQ,QAAQ,KAAK,EAAE,MAAM,IAAI,EAAE,OAAO,OAAO;AACvD,iBAAW,QAAQ,OAAO;AACxB,YAAI;AACF,kBAAQ,KAAK,KAAK,MAAM,IAAI,CAAC;AAAA,QAC/B,QAAQ;AAAA,QAAuB;AAAA,MACjC;AAAA,IACF,QAAQ;AAAA,IAER;AAEA,YAAQ,QAAQ,QAAQ,QAAQ,IAAI,CAAC;AAAA,EACvC;AAEA,SAAO;AACT;AAKO,SAAS,gBAAgB,SAAwB;AACtD,aAAW;AACb;",
+  "names": []
+}

package/dist/src/chunker.d.ts

@@ -0,0 +1,45 @@
+/**
+ * Long Context Chunking System
+ *
+ * Goal: split documents that exceed embedding model context limits into smaller,
+ * semantically coherent chunks with overlap.
+ *
+ * Notes:
+ * - We use *character counts* as a conservative proxy for tokens.
+ * - The embedder triggers this only after a provider throws a context-length error.
+ */
+export interface ChunkMetadata {
+    startIndex: number;
+    endIndex: number;
+    length: number;
+}
+export interface ChunkResult {
+    chunks: string[];
+    metadatas: ChunkMetadata[];
+    totalOriginalLength: number;
+    chunkCount: number;
+}
+export interface ChunkerConfig {
+    /** Maximum characters per chunk. */
+    maxChunkSize: number;
+    /** Overlap between chunks in characters. */
+    overlapSize: number;
+    /** Minimum chunk size (except the final chunk). */
+    minChunkSize: number;
+    /** Attempt to split on sentence boundaries for better semantic coherence. */
+    semanticSplit: boolean;
+    /** Max lines per chunk before we try to split earlier on a line boundary. */
+    maxLinesPerChunk: number;
+}
+export declare const EMBEDDING_CONTEXT_LIMITS: Record<string, number>;
+export declare const DEFAULT_CHUNKER_CONFIG: ChunkerConfig;
+export declare function chunkDocument(text: string, config?: ChunkerConfig): ChunkResult;
+/**
+ * Smart chunker that adapts to model context limits.
+ *
+ * We intentionally pick conservative char limits (70% of the reported limit)
+ * since token/char ratios vary.
+ */
+export declare function smartChunk(text: string, embedderModel?: string): ChunkResult;
+export default chunkDocument;
+//# sourceMappingURL=chunker.d.ts.map

package/dist/src/chunker.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"chunker.d.ts","sourceRoot":"","sources":["../../src/chunker.ts"],"names":[],"mappings":"AACA;;;;;;;;;GASG;AAMH,MAAM,WAAW,aAAa;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,WAAW;IAC1B,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,EAAE,aAAa,EAAE,CAAC;IAC3B,mBAAmB,EAAE,MAAM,CAAC;IAC5B,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,aAAa;IAC5B,oCAAoC;IACpC,YAAY,EAAE,MAAM,CAAC;IACrB,4CAA4C;IAC5C,WAAW,EAAE,MAAM,CAAC;IACpB,mDAAmD;IACnD,YAAY,EAAE,MAAM,CAAC;IACrB,6EAA6E;IAC7E,aAAa,EAAE,OAAO,CAAC;IACvB,6EAA6E;IAC7E,gBAAgB,EAAE,MAAM,CAAC;CAC1B;AAID,eAAO,MAAM,wBAAwB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAiB3D,CAAC;AAEF,eAAO,MAAM,sBAAsB,EAAE,aAMpC,CAAC;AAoGF,wBAAgB,aAAa,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,GAAE,aAAsC,GAAG,WAAW,CA6DvG;AAED;;;;;GAKG;AACH,wBAAgB,UAAU,CAAC,IAAI,EAAE,MAAM,EAAE,aAAa,CAAC,EAAE,MAAM,GAAG,WAAW,CAa5E;AAED,eAAe,aAAa,CAAC"}