@agenticmail/enterprise 0.3.2 → 0.4.0

package/src/engine/agent-memory.ts

@@ -0,0 +1,1062 @@
+/**
+ * Agent Memory Manager -- Persistent Memory & Learning System
+ *
+ * Provides growing, persistent memory for AI agents that evolves over time,
+ * similar to how human employees learn and adapt in an organization.
+ *
+ * Features:
+ * - Category-based memory organization (org_knowledge, interaction_pattern, preference, correction, skill, context, reflection)
+ * - Importance levels (critical, high, normal, low)
+ * - Confidence scores that decay over time for unaccessed entries
+ * - Access tracking for frequently used knowledge
+ * - Context generation for injection into agent prompts
+ * - Pruning of expired/low-confidence entries
+ */
+
+import type { EngineDatabase } from './db-adapter.js';
+
+// ─── Types ──────────────────────────────────────────────
+
+export type MemoryCategory =
+  | 'org_knowledge'
+  | 'interaction_pattern'
+  | 'preference'
+  | 'correction'
+  | 'skill'
+  | 'context'
+  | 'reflection';
+
+export type MemoryImportance = 'critical' | 'high' | 'normal' | 'low';
+
+export type MemorySource =
+  | 'onboarding'
+  | 'interaction'
+  | 'admin'
+  | 'self_reflection'
+  | 'correction';
+
+export const MEMORY_CATEGORIES: Record<MemoryCategory, { label: string; description: string }> = {
+  org_knowledge: {
+    label: 'Organization Knowledge',
+    description: 'Policies, procedures, and organizational information',
+  },
+  interaction_pattern: {
+    label: 'Interaction Patterns',
+    description: 'Learned patterns from past interactions',
+  },
+  preference: {
+    label: 'Preferences',
+    description: 'User and organizational preferences',
+  },
+  correction: {
+    label: 'Corrections',
+    description: 'Corrections and feedback received',
+  },
+  skill: {
+    label: 'Skills',
+    description: 'Learned abilities and competencies',
+  },
+  context: {
+    label: 'Context',
+    description: 'Contextual information and background knowledge',
+  },
+  reflection: {
+    label: 'Reflections',
+    description: 'Self-reflective insights and learnings',
+  },
+};
+
+export interface AgentMemoryEntry {
+  id: string;
+  agentId: string;
+  orgId: string;
+  category: MemoryCategory;
+  title: string;
+  content: string;
+  source: MemorySource;
+  importance: MemoryImportance;
+  confidence: number; // 0.0-1.0
+  accessCount: number;
+  lastAccessedAt?: string;
+  expiresAt?: string;
+  tags: string[];
+  metadata: Record<string, any>;
+  createdAt: string;
+  updatedAt: string;
+}
+
+export interface MemoryStats {
+  totalEntries: number;
+  byCategory: Record<string, number>;
+  byImportance: Record<string, number>;
+  bySource: Record<string, number>;
+  avgConfidence: number;
+}
+
+/** Input shape for createMemory — id, timestamps, and accessCount are generated automatically. */
+export type CreateMemoryInput = Omit<AgentMemoryEntry, 'id' | 'createdAt' | 'updatedAt' | 'accessCount'>;
+
+/** Input shape for updateMemory — partial updates merged with existing entry. */
+export type UpdateMemoryInput = Partial<Omit<AgentMemoryEntry, 'id' | 'agentId' | 'orgId' | 'createdAt'>>;
+
+/** Query options for filtering memory entries. */
+export interface MemoryQueryOptions {
+  agentId: string;
+  category?: string;
+  importance?: string;
+  source?: string;
+  query?: string;
+  limit?: number;
+}
+
+/** Minimal policy shape expected by createFromPolicy (matches OrgPolicy fields used). */
+export interface PolicyForMemory {
+  id: string;
+  orgId: string;
+  name: string;
+  category: string;
+  content: string;
+  enforcement: string;
+}
+
+// ─── Importance Weight Map ──────────────────────────────
+
+const IMPORTANCE_WEIGHT: Record<MemoryImportance, number> = {
+  critical: 4,
+  high: 3,
+  normal: 2,
+  low: 1,
+};
+
+// ─── Full-Text Search Engine (BM25F + Stemming + Inverted Index) ─────
+//
+// Comprehensive text relevance scoring for agent memory retrieval.
+// Zero dependencies — pure TypeScript implementation.
+//
+// Features:
+// - Pre-built inverted index maintained incrementally (no re-indexing on query)
+// - Lightweight Porter-style stemmer (suffix stripping for English)
+// - Field weighting via BM25F: title ×3, tags ×2, content ×1
+// - Pre-computed IDF values updated on index mutations
+// - Prefix matching: "deploy" matches "deployment", "deployments"
+// - Per-agent partitioning for scoped searches
+// - Bigram proximity boost: terms appearing adjacent score higher
+
+// ── BM25 Parameters ──
+
+const BM25_K1 = 1.2;   // Term frequency saturation
+const BM25_B = 0.75;    // Document length normalization
+const FIELD_WEIGHT_TITLE = 3.0;
+const FIELD_WEIGHT_TAGS = 2.0;
+const FIELD_WEIGHT_CONTENT = 1.0;
+const PREFIX_MATCH_PENALTY = 0.7; // Prefix matches score 70% of exact matches
+
+// ── Stop Words ──
+
+const STOP_WORDS = new Set([
+  'a', 'about', 'above', 'after', 'again', 'against', 'all', 'am', 'an',
+  'and', 'any', 'are', 'as', 'at', 'be', 'because', 'been', 'before',
+  'being', 'below', 'between', 'both', 'but', 'by', 'can', 'could', 'did',
+  'do', 'does', 'doing', 'down', 'during', 'each', 'either', 'every',
+  'few', 'for', 'from', 'further', 'get', 'got', 'had', 'has', 'have',
+  'having', 'he', 'her', 'here', 'hers', 'herself', 'him', 'himself',
+  'his', 'how', 'i', 'if', 'in', 'into', 'is', 'it', 'its', 'itself',
+  'just', 'may', 'me', 'might', 'more', 'most', 'must', 'my', 'myself',
+  'neither', 'no', 'nor', 'not', 'now', 'of', 'off', 'on', 'once', 'only',
+  'or', 'other', 'ought', 'our', 'ours', 'ourselves', 'out', 'over', 'own',
+  'same', 'shall', 'she', 'should', 'so', 'some', 'such', 'than', 'that',
+  'the', 'their', 'theirs', 'them', 'themselves', 'then', 'there', 'these',
+  'they', 'this', 'those', 'through', 'to', 'too', 'under', 'until', 'up',
+  'us', 'very', 'was', 'we', 'were', 'what', 'when', 'where', 'which',
+  'while', 'who', 'whom', 'why', 'will', 'with', 'would', 'yet', 'you',
+  'your', 'yours', 'yourself', 'yourselves',
+]);
+
+// ── Porter Stemmer (lightweight suffix stripping) ──
+// Handles common English suffixes to normalize "deployments" → "deploy",
+// "running" → "run", "policies" → "polici", "configured" → "configur".
+// Not a full Porter stemmer — covers the 80/20 of suffixes that matter most.
+
+const STEM_RULES: [RegExp, string, number][] = [
+  // Step 1: plurals and past participles
+  [/ies$/, 'i', 3],            // policies → polici,eries → eri
+  [/sses$/, 'ss', 4],          // addresses → address
+  [/([^s])s$/, '$1', 3],       // items → item, but not "ss"
+  [/eed$/, 'ee', 4],           // agreed → agree
+  [/ed$/, '', 3],              // configured → configur, but min length 3
+  [/ing$/, '', 4],             // running → runn → run (handled below)
+  // Step 2: derivational suffixes
+  [/ational$/, 'ate', 6],      // relational → relate
+  [/tion$/, 't', 5],           // adoption → adopt
+  [/ness$/, '', 5],            // awareness → aware
+  [/ment$/, '', 5],            // deployment → deploy
+  [/able$/, '', 5],            // configurable → configur
+  [/ible$/, '', 5],            // accessible → access
+  [/ful$/, '', 5],             // powerful → power
+  [/ous$/, '', 5],             // dangerous → danger
+  [/ive$/, '', 5],             // interactive → interact
+  [/ize$/, '', 4],             // normalize → normal
+  [/ise$/, '', 4],             // organise → organ
+  [/ally$/, '', 5],            // automatically → automat
+  [/ly$/, '', 4],              // quickly → quick
+  [/er$/, '', 4],              // handler → handl
+];
+
+/** Clean up common doubling artifacts after suffix stripping. */
+const DOUBLE_CONSONANT = /([^aeiou])\1$/;
+
+function stem(word: string): string {
+  if (word.length < 3) return word;
+  let stemmed = word;
+  for (const [pattern, replacement, minLen] of STEM_RULES) {
+    if (stemmed.length >= minLen && pattern.test(stemmed)) {
+      stemmed = stemmed.replace(pattern, replacement);
+      break; // Apply only the first matching rule
+    }
+  }
+  // Clean doubled consonants: runn → run, configurr → configur
+  if (stemmed.length > 2 && DOUBLE_CONSONANT.test(stemmed)) {
+    stemmed = stemmed.slice(0, -1);
+  }
+  return stemmed;
+}
+
+// ── Tokenizer ──
+
+/** Tokenize text into stemmed, lowercase terms, filtering stop words. */
+function tokenize(text: string): string[] {
+  return text.toLowerCase()
+    .split(/[^a-z0-9]+/)
+    .filter((t) => t.length > 1 && !STOP_WORDS.has(t))
+    .map(stem);
+}
+
+/** Tokenize preserving original (unstemmed) forms alongside stems. */
+function tokenizeWithOriginals(text: string): { stem: string; original: string }[] {
+  return text.toLowerCase()
+    .split(/[^a-z0-9]+/)
+    .filter((t) => t.length > 1 && !STOP_WORDS.has(t))
+    .map((t) => ({ stem: stem(t), original: t }));
+}
+
+// ── Inverted Index Data Structures ──
+
+interface DocRecord {
+  /** Weighted term frequencies across all fields: title (3x), tags (2x), content (1x) */
+  weightedTf: Map<string, number>;
+  /** Total weighted document length (for BM25 length normalization) */
+  weightedLen: number;
+  /** All unique stems in the document (for prefix matching) */
+  allStems: Set<string>;
+  /** Ordered list of stems for bigram proximity detection */
+  stemSequence: string[];
+}
+
+/**
+ * Pre-built inverted index for fast text search.
+ * Maintained incrementally — no re-indexing needed on queries.
+ *
+ * Structure:
+ *   term → Set<docId>              (posting list — which docs contain this term)
+ *   prefixMap: prefix → Set<stem>  (3-char prefixes → full stems for prefix matching)
+ *   docs: docId → DocRecord        (per-doc weighted TF and length)
+ *   idf: term → number             (pre-computed IDF, refreshed on mutations)
+ */
+class MemorySearchIndex {
+  /** Posting lists: stemmed term → Set of memory IDs containing it */
+  private postings = new Map<string, Set<string>>();
+  /** Per-document metadata for BM25 scoring */
+  private docs = new Map<string, DocRecord>();
+  /** Pre-computed IDF values. Stale flag triggers lazy recomputation. */
+  private idf = new Map<string, number>();
+  private idfStale = true;
+  /** 3-character prefix map for prefix matching: prefix → Set of full stems */
+  private prefixMap = new Map<string, Set<string>>();
+  /** Total weighted document length (for computing average) */
+  private totalWeightedLen = 0;
+
+  get docCount(): number { return this.docs.size; }
+  get avgDocLen(): number { return this.docs.size > 0 ? this.totalWeightedLen / this.docs.size : 1; }
+
+  /**
+   * Index a memory entry. Extracts stems from title, content, and tags
+   * with field-specific weighting and builds posting lists.
+   */
+  addDocument(id: string, entry: { title: string; content: string; tags: string[] }): void {
+    // Remove old version if updating
+    if (this.docs.has(id)) this.removeDocument(id);
+
+    const titleTokens = tokenize(entry.title);
+    const contentTokens = tokenize(entry.content);
+    const tagTokens = entry.tags.flatMap((t) => tokenize(t));
+
+    // Build weighted term frequency map
+    const weightedTf = new Map<string, number>();
+    for (const t of titleTokens) weightedTf.set(t, (weightedTf.get(t) || 0) + FIELD_WEIGHT_TITLE);
+    for (const t of tagTokens) weightedTf.set(t, (weightedTf.get(t) || 0) + FIELD_WEIGHT_TAGS);
+    for (const t of contentTokens) weightedTf.set(t, (weightedTf.get(t) || 0) + FIELD_WEIGHT_CONTENT);
+
+    const weightedLen = titleTokens.length * FIELD_WEIGHT_TITLE
+      + tagTokens.length * FIELD_WEIGHT_TAGS
+      + contentTokens.length * FIELD_WEIGHT_CONTENT;
+
+    const allStems = new Set<string>();
+    for (const t of weightedTf.keys()) allStems.add(t);
+
+    // Stem sequence for bigram proximity (title first, then content — most important ordering)
+    const stemSequence = [...titleTokens, ...contentTokens];
+
+    const docRecord: DocRecord = { weightedTf, weightedLen, allStems, stemSequence };
+    this.docs.set(id, docRecord);
+    this.totalWeightedLen += weightedLen;
+
+    // Update posting lists
+    for (const term of allStems) {
+      let posting = this.postings.get(term);
+      if (!posting) { posting = new Set(); this.postings.set(term, posting); }
+      posting.add(id);
+
+      // Update prefix map (3-char prefixes for prefix matching)
+      if (term.length >= 3) {
+        const prefix = term.slice(0, 3);
+        let prefixSet = this.prefixMap.get(prefix);
+        if (!prefixSet) { prefixSet = new Set(); this.prefixMap.set(prefix, prefixSet); }
+        prefixSet.add(term);
+      }
+    }
+
+    this.idfStale = true;
+  }
+
+  /** Remove a document from the index. */
+  removeDocument(id: string): void {
+    const doc = this.docs.get(id);
+    if (!doc) return;
+
+    this.totalWeightedLen -= doc.weightedLen;
+    this.docs.delete(id);
+
+    // Remove from posting lists
+    for (const term of doc.allStems) {
+      const posting = this.postings.get(term);
+      if (posting) {
+        posting.delete(id);
+        if (posting.size === 0) {
+          this.postings.delete(term);
+          // Clean prefix map
+          if (term.length >= 3) {
+            const prefixSet = this.prefixMap.get(term.slice(0, 3));
+            if (prefixSet) { prefixSet.delete(term); if (prefixSet.size === 0) this.prefixMap.delete(term.slice(0, 3)); }
+          }
+        }
+      }
+    }
+
+    this.idfStale = true;
+  }
+
+  /** Recompute IDF values for all terms. Called lazily before search. */
+  private refreshIdf(): void {
+    if (!this.idfStale) return;
+    const N = this.docs.size;
+    this.idf.clear();
+    for (const [term, posting] of this.postings) {
+      const df = posting.size;
+      // BM25 IDF: log((N - df + 0.5) / (df + 0.5) + 1)
+      this.idf.set(term, Math.log((N - df + 0.5) / (df + 0.5) + 1));
+    }
+    this.idfStale = false;
+  }
+
+  /**
+   * Expand query terms with prefix matches.
+   * "deploy" → ["deploy", "deployment", "deploying", ...] (if they exist in the index)
+   */
+  private expandQueryTerms(queryStems: string[]): Map<string, number> {
+    const expanded = new Map<string, number>();
+
+    for (const qs of queryStems) {
+      // Exact match always gets full weight
+      if (this.postings.has(qs)) {
+        expanded.set(qs, Math.max(expanded.get(qs) || 0, 1.0));
+      }
+
+      // Prefix expansion: find all stems that start with the query stem (min 3 chars)
+      if (qs.length >= 3) {
+        const prefix = qs.slice(0, 3);
+        const candidates = this.prefixMap.get(prefix);
+        if (candidates) {
+          for (const candidate of candidates) {
+            if (candidate !== qs && candidate.startsWith(qs)) {
+              expanded.set(candidate, Math.max(expanded.get(candidate) || 0, PREFIX_MATCH_PENALTY));
+            }
+          }
+        }
+      }
+    }
+
+    return expanded;
+  }
+
+  /**
+   * Compute bigram proximity boost: if two query terms appear adjacent
+   * in the document's stem sequence, boost the score.
+   */
+  private bigramProximityBoost(docId: string, queryStems: string[]): number {
+    if (queryStems.length < 2) return 0;
+    const doc = this.docs.get(docId);
+    if (!doc || doc.stemSequence.length < 2) return 0;
+
+    let boost = 0;
+    const seq = doc.stemSequence;
+    const querySet = new Set(queryStems);
+
+    for (let i = 0; i < seq.length - 1; i++) {
+      if (querySet.has(seq[i]) && querySet.has(seq[i + 1]) && seq[i] !== seq[i + 1]) {
+        boost += 0.5; // Each adjacent pair of query terms adds 0.5
+      }
+    }
+
+    return Math.min(boost, 2.0); // Cap at 2.0 bonus
+  }
+
+  /**
+   * Search the index for documents matching a query.
+   * Returns scored results sorted by BM25F relevance.
+   *
+   * @param query - Raw query string
+   * @param candidateIds - Optional: only score these document IDs (for agent-scoped search)
+   * @returns Array of { id, score } sorted by descending score
+   */
+  search(query: string, candidateIds?: Set<string>): Array<{ id: string; score: number }> {
+    const queryStems = tokenize(query);
+    if (queryStems.length === 0) return [];
+
+    this.refreshIdf();
+
+    const expandedTerms = this.expandQueryTerms(queryStems);
+    if (expandedTerms.size === 0) return [];
+
+    const avgDl = this.avgDocLen;
+
+    // Collect candidate document IDs from posting lists
+    const candidates = new Set<string>();
+    for (const term of expandedTerms.keys()) {
+      const posting = this.postings.get(term);
+      if (posting) {
+        for (const docId of posting) {
+          if (!candidateIds || candidateIds.has(docId)) candidates.add(docId);
+        }
+      }
+    }
+
+    // Score each candidate
+    const results: Array<{ id: string; score: number }> = [];
+
+    for (const docId of candidates) {
+      const doc = this.docs.get(docId);
+      if (!doc) continue;
+
+      let score = 0;
+
+      for (const [term, weight] of expandedTerms) {
+        const tf = doc.weightedTf.get(term) || 0;
+        if (tf === 0) continue;
+        const termIdf = this.idf.get(term) || 0;
+
+        // BM25F: IDF × (weightedTF × (k1 + 1)) / (weightedTF + k1 × (1 - b + b × docLen/avgDocLen))
+        const numerator = tf * (BM25_K1 + 1);
+        const denominator = tf + BM25_K1 * (1 - BM25_B + BM25_B * (doc.weightedLen / avgDl));
+        score += termIdf * (numerator / denominator) * weight;
+      }
+
+      // Bigram proximity boost
+      score += this.bigramProximityBoost(docId, queryStems);
+
+      if (score > 0) results.push({ id: docId, score });
+    }
+
+    results.sort((a, b) => b.score - a.score);
+    return results;
+  }
+
+  /** Check if a document exists in the index. */
+  has(id: string): boolean { return this.docs.has(id); }
+}
+
+// ─── Agent Memory Manager ───────────────────────────────
+
+export class AgentMemoryManager {
+  private memories = new Map<string, AgentMemoryEntry>();
+  /** Per-agent index: agentId → Set of memory IDs for O(1) agent lookups */
+  private agentIndex = new Map<string, Set<string>>();
+  /** Full-text search index (BM25F + stemming + inverted index) */
+  private searchIndex = new MemorySearchIndex();
+  private engineDb?: EngineDatabase;
+
+  // ─── Database Lifecycle ─────────────────────────────
+
+  async setDb(db: EngineDatabase): Promise<void> {
+    this.engineDb = db;
+    await this.loadFromDb();
+  }
+
+  private async loadFromDb(): Promise<void> {
+    if (!this.engineDb) return;
+    try {
+      const rows = await this.engineDb.query<any>('SELECT * FROM agent_memory');
+      for (const r of rows) {
+        const entry = this.rowToEntry(r);
+        this.memories.set(entry.id, entry);
+        this.indexAdd(entry.agentId, entry.id);
+        this.searchIndex.addDocument(entry.id, entry);
+      }
+    } catch {
+      /* table may not exist yet */
+    }
+  }
+
+  /** Add a memory ID to the per-agent index. */
+  private indexAdd(agentId: string, memoryId: string): void {
+    let set = this.agentIndex.get(agentId);
+    if (!set) { set = new Set(); this.agentIndex.set(agentId, set); }
+    set.add(memoryId);
+  }
+
+  /** Remove a memory ID from the per-agent index. */
+  private indexRemove(agentId: string, memoryId: string): void {
+    const set = this.agentIndex.get(agentId);
+    if (set) { set.delete(memoryId); if (set.size === 0) this.agentIndex.delete(agentId); }
+  }
+
+  /** Get all memory entries for an agent via the index — O(n) where n is that agent's entries, not total. */
+  private getAgentMemories(agentId: string): AgentMemoryEntry[] {
+    const ids = this.agentIndex.get(agentId);
+    if (!ids || ids.size === 0) return [];
+    const result: AgentMemoryEntry[] = [];
+    for (const id of ids) {
+      const entry = this.memories.get(id);
+      if (entry) result.push(entry);
+    }
+    return result;
+  }
+
+  // ─── CRUD Operations ────────────────────────────────
+
+  /**
+   * Creates a new memory entry with auto-generated id and timestamps.
+   * Persists to both the in-memory Map and the database.
+   */
+  async createMemory(input: CreateMemoryInput): Promise<AgentMemoryEntry> {
+    const now = new Date().toISOString();
+    const entry: AgentMemoryEntry = {
+      ...input,
+      id: crypto.randomUUID(),
+      accessCount: 0,
+      createdAt: now,
+      updatedAt: now,
+    };
+
+    this.memories.set(entry.id, entry);
+    this.indexAdd(entry.agentId, entry.id);
+    this.searchIndex.addDocument(entry.id, entry);
+
+    await this.engineDb?.execute(
+      `INSERT INTO agent_memory (id, agent_id, org_id, category, title, content, source, importance, confidence, access_count, last_accessed_at, expires_at, tags, metadata, created_at, updated_at)
+       VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,
+      [
+        entry.id, entry.agentId, entry.orgId, entry.category,
+        entry.title, entry.content, entry.source, entry.importance,
+        entry.confidence, entry.accessCount, entry.lastAccessedAt || null,
+        entry.expiresAt || null, JSON.stringify(entry.tags),
+        JSON.stringify(entry.metadata), entry.createdAt, entry.updatedAt,
+      ]
+    ).catch((err) => {
+      console.error('[agent-memory] Failed to persist memory entry:', err);
+    });
+
+    return entry;
+  }
+
+  /**
+   * Updates an existing memory entry by merging provided fields.
+   * Returns the updated entry or null if not found.
+   */
+  async updateMemory(id: string, updates: UpdateMemoryInput): Promise<AgentMemoryEntry | null> {
+    const existing = this.memories.get(id);
+    if (!existing) return null;
+
+    const now = new Date().toISOString();
+    const updated: AgentMemoryEntry = {
+      ...existing,
+      ...updates,
+      id: existing.id,
+      agentId: existing.agentId,
+      orgId: existing.orgId,
+      createdAt: existing.createdAt,
+      updatedAt: now,
+    };
+
+    this.memories.set(id, updated);
+
+    // Re-index if text fields changed
+    if (updates.title !== undefined || updates.content !== undefined || updates.tags !== undefined) {
+      this.searchIndex.addDocument(id, updated);
+    }
+
+    await this.engineDb?.execute(
+      `UPDATE agent_memory SET
+        category = ?, title = ?, content = ?, source = ?,
+        importance = ?, confidence = ?, access_count = ?,
+        last_accessed_at = ?, expires_at = ?, tags = ?,
+        metadata = ?, updated_at = ?
+       WHERE id = ?`,
+      [
+        updated.category, updated.title, updated.content, updated.source,
+        updated.importance, updated.confidence, updated.accessCount,
+        updated.lastAccessedAt || null, updated.expiresAt || null,
+        JSON.stringify(updated.tags), JSON.stringify(updated.metadata),
+        updated.updatedAt, id,
+      ]
+    ).catch((err) => {
+      console.error('[agent-memory] Failed to update memory entry:', err);
+    });
+
+    return updated;
+  }
+
+  /**
+   * Deletes a memory entry from both the in-memory Map and the database.
+   * Returns true if the entry existed and was deleted.
+   */
+  async deleteMemory(id: string): Promise<boolean> {
+    const entry = this.memories.get(id);
+    const existed = this.memories.delete(id);
+    if (entry) this.indexRemove(entry.agentId, id);
+    this.searchIndex.removeDocument(id);
+
+    await this.engineDb?.execute(
+      'DELETE FROM agent_memory WHERE id = ?',
+      [id]
+    ).catch((err) => {
+      console.error('[agent-memory] Failed to delete memory entry:', err);
+    });
+
+    return existed;
+  }
+
+  /**
+   * Retrieves a single memory entry by id.
+   * Returns from the in-memory Map (async for interface compatibility).
+   */
+  async getMemory(id: string): Promise<AgentMemoryEntry | undefined> {
+    return this.memories.get(id);
+  }
+
+  // ─── Query Operations ───────────────────────────────
+
+  /**
+   * Queries memory entries for a specific agent with optional filters.
+   * Supports filtering by category, importance, source, and text search on title/content.
+   */
+  async queryMemories(opts: MemoryQueryOptions): Promise<AgentMemoryEntry[]> {
+    let results = this.getAgentMemories(opts.agentId);
+
+    if (opts.category) {
+      results = results.filter((m) => m.category === opts.category);
+    }
+
+    if (opts.importance) {
+      results = results.filter((m) => m.importance === opts.importance);
+    }
+
+    if (opts.source) {
+      results = results.filter((m) => m.source === opts.source);
+    }
+
+    if (opts.query) {
+      // Use the pre-built inverted index for BM25F search
+      const candidateIds = new Set(results.map((m) => m.id));
+      const searchResults = this.searchIndex.search(opts.query, candidateIds);
+
+      if (searchResults.length > 0) {
+        // Combine BM25F relevance with importance weight
+        const scored = searchResults
+          .map((r) => {
+            const entry = this.memories.get(r.id);
+            return entry ? { entry, score: r.score * IMPORTANCE_WEIGHT[entry.importance] } : null;
+          })
+          .filter((r): r is { entry: AgentMemoryEntry; score: number } => r !== null);
+
+        scored.sort((a, b) => b.score - a.score);
+        return scored.slice(0, opts.limit || 100).map((d) => d.entry);
+      }
+    }
+
+    // No query — sort by importance weight descending, then by creation date descending
+    results.sort((a, b) => {
+      const weightDiff = IMPORTANCE_WEIGHT[b.importance] - IMPORTANCE_WEIGHT[a.importance];
+      if (weightDiff !== 0) return weightDiff;
+      return new Date(b.createdAt).getTime() - new Date(a.createdAt).getTime();
+    });
+
+    const limit = opts.limit || 100;
+    return results.slice(0, limit);
+  }
+
+  /**
+   * Returns memories created within the last N hours for an agent.
+   */
+  async getRecentMemories(agentId: string, hours: number = 24): Promise<AgentMemoryEntry[]> {
+    const cutoff = new Date(Date.now() - hours * 3600_000).toISOString();
+    return this.getAgentMemories(agentId)
+      .filter((m) => m.createdAt >= cutoff)
+      .sort((a, b) => new Date(b.createdAt).getTime() - new Date(a.createdAt).getTime());
+  }
+
+  // ─── Access Tracking ────────────────────────────────
+
+  /**
+   * Increments the access count and updates lastAccessedAt for a memory entry.
+   * Used to track frequently referenced knowledge.
+   */
+  async recordAccess(memoryId: string): Promise<void> {
+    const entry = this.memories.get(memoryId);
+    if (!entry) return;
+
+    const now = new Date().toISOString();
+    entry.accessCount += 1;
+    entry.lastAccessedAt = now;
+    entry.updatedAt = now;
+
+    await this.engineDb?.execute(
+      `UPDATE agent_memory SET access_count = ?, last_accessed_at = ?, updated_at = ? WHERE id = ?`,
+      [entry.accessCount, entry.lastAccessedAt, entry.updatedAt, memoryId]
+    ).catch((err) => {
+      console.error('[agent-memory] Failed to record access:', err);
+    });
+  }
+
+  // ─── Context Generation ─────────────────────────────
+
+  /**
+   * Generates a markdown context string suitable for injection into an agent's system prompt.
+   *
+   * Ranking strategy:
+   * 1. Critical importance entries always come first
+   * 2. Remaining entries scored by: confidence * accessWeight * recencyWeight
+   * 3. If a query is provided, entries matching query terms get a relevance boost
+   * 4. Output is grouped by category with markdown headers
+   * 5. Truncated to maxTokens (estimated at ~4 chars per token)
+   */
+  async generateMemoryContext(
+    agentId: string,
+    query?: string,
+    maxTokens: number = 1500,
+  ): Promise<string> {
+    const entries = this.getAgentMemories(agentId)
+      .filter((m) => m.confidence >= 0.1);
+
+    if (entries.length === 0) return '';
+
+    const now = Date.now();
+
+    // Pre-compute BM25F relevance scores via the inverted index if query provided
+    let relevanceMap: Map<string, number> | undefined;
+    if (query) {
+      const candidateIds = new Set(entries.map((e) => e.id));
+      const searchResults = this.searchIndex.search(query, candidateIds);
+      if (searchResults.length > 0) {
+        relevanceMap = new Map();
+        // Normalize scores: divide by max so top result = 1.0
+        const maxScore = searchResults[0].score;
+        for (const r of searchResults) {
+          relevanceMap.set(r.id, maxScore > 0 ? r.score / maxScore : 0);
+        }
+      }
+    }
+
+    // Score each entry for ranking: combines recency, access, confidence, importance, and BM25F relevance
+    const scored = entries.map((entry) => {
+      // Access weight: logarithmic scaling of access count
+      const accessWeight = 1 + Math.log1p(entry.accessCount) * 0.3;
+
+      // Recency weight: entries accessed/created more recently score higher
+      const lastTouch = entry.lastAccessedAt || entry.createdAt;
+      const ageHours = Math.max(1, (now - new Date(lastTouch).getTime()) / 3600_000);
+      const recencyWeight = 1 / (1 + Math.log1p(ageHours / 24) * 0.2);
+
+      // Base score from confidence, access weight, and recency
+      let score = entry.confidence * accessWeight * recencyWeight;
+
+      // Importance multiplier
+      score *= IMPORTANCE_WEIGHT[entry.importance];
+
+      // BM25F query relevance boost (from pre-built inverted index)
+      if (relevanceMap) {
+        const relevance = relevanceMap.get(entry.id) || 0;
+        if (relevance > 0) {
+          // Normalized relevance (0–1) → multiplier (1x–4x boost)
+          score *= 1 + relevance * 3;
+        }
+      }
+
+      return { entry, score };
+    });
+
+    // Sort by score descending
+    scored.sort((a, b) => b.score - a.score);
+
+    // Group by category preserving score order
+    const grouped = new Map<MemoryCategory, AgentMemoryEntry[]>();
+    for (const { entry } of scored) {
+      const group = grouped.get(entry.category) || [];
+      group.push(entry);
+      grouped.set(entry.category, group);
+    }
+
+    // Build markdown output
+    const maxChars = maxTokens * 4;
+    const lines: string[] = ['## Agent Memory', ''];
+    let charCount = lines.join('\n').length;
+
+    for (const [category, categoryEntries] of Array.from(grouped.entries())) {
+      const meta = MEMORY_CATEGORIES[category];
+      if (!meta) continue;
+
+      const header = `### ${meta.label}`;
+      if (charCount + header.length + 2 > maxChars) break;
+
+      lines.push(header);
+      lines.push('');
+      charCount += header.length + 2;
+
+      for (const entry of categoryEntries) {
+        const badge = entry.importance === 'critical' ? '[CRITICAL] '
+          : entry.importance === 'high' ? '[HIGH] '
+          : '';
+        const entryLine = `- **${badge}${entry.title}**: ${entry.content}`;
+
+        if (charCount + entryLine.length + 1 > maxChars) break;
+
+        lines.push(entryLine);
+        charCount += entryLine.length + 1;
+      }
+
+      lines.push('');
+      charCount += 1;
+    }
+
+    return lines.join('\n').trim();
+  }
+
+  // ─── Memory Lifecycle ───────────────────────────────
+
+  /**
+   * Decays confidence scores for entries that have not been accessed in 7+ days.
+   * Critical importance entries are exempt from decay.
+   *
+   * @param agentId - The agent whose memories to decay
+   * @param decayRate - How much to reduce confidence (default 0.05)
+   * @returns Number of entries that were decayed
+   */
+  async decayConfidence(agentId: string, decayRate: number = 0.05): Promise<number> {
+    const cutoff = new Date(Date.now() - 7 * 86_400_000).toISOString();
+    const now = new Date().toISOString();
+    const decayed: { id: string; confidence: number }[] = [];
+
+    for (const entry of this.getAgentMemories(agentId)) {
+      if (entry.importance === 'critical') continue;
+
+      const lastTouch = entry.lastAccessedAt || entry.createdAt;
+      if (lastTouch >= cutoff) continue;
+
+      const newConfidence = Math.max(0, entry.confidence - decayRate);
+      if (newConfidence === entry.confidence) continue;
+
+      entry.confidence = parseFloat(newConfidence.toFixed(4));
+      entry.updatedAt = now;
+      decayed.push({ id: entry.id, confidence: entry.confidence });
+    }
+
+    // Batch DB update — single statement per entry but without await-per-row overhead
+    if (decayed.length > 0 && this.engineDb) {
+      await Promise.all(
+        decayed.map((d) =>
+          this.engineDb!.execute(
+            'UPDATE agent_memory SET confidence = ?, updated_at = ? WHERE id = ?',
+            [d.confidence, now, d.id]
+          ).catch((err) => console.error('[agent-memory] Failed to decay confidence:', err))
+        )
+      );
+    }
+
+    return decayed.length;
+  }
+
+  /**
+   * Prunes (deletes) memory entries that are expired or have very low confidence.
+   * An entry is pruned if:
+   * - confidence < 0.1, OR
+   * - expiresAt is set and has passed
+   *
+   * @param agentId - Optional: limit pruning to a specific agent. If omitted, prunes all.
+   * @returns Number of entries pruned
+   */
+  async pruneExpired(agentId?: string): Promise<number> {
+    const now = new Date().toISOString();
+    const toDelete: { id: string; agentId: string }[] = [];
+
+    const entries = agentId
+      ? this.getAgentMemories(agentId)
+      : Array.from(this.memories.values());
+
+    for (const entry of entries) {
+      const isLowConfidence = entry.confidence < 0.1;
+      const isExpired = entry.expiresAt && entry.expiresAt <= now;
+
+      if (isLowConfidence || isExpired) {
+        toDelete.push({ id: entry.id, agentId: entry.agentId });
+      }
+    }
+
+    // Delete from Map + agent index + search index
+    for (const item of toDelete) {
+      this.memories.delete(item.id);
+      this.indexRemove(item.agentId, item.id);
+      this.searchIndex.removeDocument(item.id);
+    }
+
+    // Batch DB deletes
+    if (toDelete.length > 0 && this.engineDb) {
+      await Promise.all(
+        toDelete.map((item) =>
+          this.engineDb!.execute('DELETE FROM agent_memory WHERE id = ?', [item.id])
+            .catch((err) => console.error('[agent-memory] Failed to prune memory entry:', err))
+        )
+      );
+    }
+
+    return toDelete.length;
+  }
+
+  // ─── Statistics ─────────────────────────────────────
+
+  /**
+   * Returns aggregate statistics for a specific agent's memory entries.
+   */
+  async getStats(agentId: string): Promise<MemoryStats> {
+    return this.computeStats(this.getAgentMemories(agentId));
+  }
+
+  /**
+   * Returns per-agent memory statistics for all agents in an organization.
+   */
+  async getStatsByOrg(orgId: string): Promise<Array<{ agentId: string; stats: MemoryStats }>> {
+    // Group entries by agentId for the given org
+    const agentMap = new Map<string, AgentMemoryEntry[]>();
+
+    for (const entry of this.memories.values()) {
+      if (entry.orgId !== orgId) continue;
+      const group = agentMap.get(entry.agentId) || [];
+      group.push(entry);
+      agentMap.set(entry.agentId, group);
+    }
+
+    const results: Array<{ agentId: string; stats: MemoryStats }> = [];
+    for (const [agentId, entries] of Array.from(agentMap.entries())) {
+      results.push({ agentId, stats: this.computeStats(entries) });
+    }
+
+    return results;
+  }
+
+  /**
+   * Computes statistics from a set of memory entries.
+   */
+  private computeStats(entries: AgentMemoryEntry[]): MemoryStats {
+    const byCategory: Record<string, number> = {};
+    const byImportance: Record<string, number> = {};
+    const bySource: Record<string, number> = {};
+    let totalConfidence = 0;
+
+    for (const entry of entries) {
+      byCategory[entry.category] = (byCategory[entry.category] || 0) + 1;
+      byImportance[entry.importance] = (byImportance[entry.importance] || 0) + 1;
+      bySource[entry.source] = (bySource[entry.source] || 0) + 1;
+      totalConfidence += entry.confidence;
+    }
+
+    return {
+      totalEntries: entries.length,
+      byCategory,
+      byImportance,
+      bySource,
+      avgConfidence: entries.length > 0
+        ? parseFloat((totalConfidence / entries.length).toFixed(4))
+        : 0,
+    };
+  }
+
+  // ─── Policy Integration ─────────────────────────────
+
+  /**
+   * Creates a memory entry from an organization policy.
+   * Used during onboarding to seed agent memory with policy knowledge.
+   *
+   * Importance is derived from policy enforcement level:
+   * - mandatory  -> critical
+   * - recommended -> high
+   * - informational / other -> normal
+   */
+  async createFromPolicy(agentId: string, policy: PolicyForMemory): Promise<AgentMemoryEntry> {
+    const importance: MemoryImportance =
+      policy.enforcement === 'mandatory' ? 'critical'
+      : policy.enforcement === 'recommended' ? 'high'
+      : 'normal';
+
+    return this.createMemory({
+      agentId,
+      orgId: policy.orgId,
+      category: 'org_knowledge',
+      title: policy.name,
+      content: policy.content,
+      source: 'onboarding',
+      importance,
+      confidence: 1.0,
+      lastAccessedAt: undefined,
+      expiresAt: undefined,
+      tags: ['policy', policy.category],
+      metadata: {
+        policyId: policy.id,
+        enforcement: policy.enforcement,
+      },
+    });
+  }
+
+  // ─── Row Mapper ─────────────────────────────────────
+
+  /**
+   * Converts a database row into an AgentMemoryEntry.
+   */
+  private rowToEntry(row: any): AgentMemoryEntry {
+    return {
+      id: row.id,
+      agentId: row.agent_id,
+      orgId: row.org_id,
+      category: row.category as MemoryCategory,
+      title: row.title,
+      content: row.content,
+      source: row.source as MemorySource,
+      importance: row.importance as MemoryImportance,
+      confidence: row.confidence,
+      accessCount: row.access_count || 0,
+      lastAccessedAt: row.last_accessed_at || undefined,
+      expiresAt: row.expires_at || undefined,
+      tags: JSON.parse(row.tags || '[]'),
+      metadata: JSON.parse(row.metadata || '{}'),
+      createdAt: row.created_at,
+      updatedAt: row.updated_at,
+    };
+  }
+}