agentic-flow 3.0.0-alpha.1 → 3.0.0

package/dist/packages/agentdb/src/controllers/ReasoningBank.js

@@ -1,489 +0,0 @@
-/**
- * ReasoningBank Controller - Pattern Storage and Retrieval
- *
- * Manages reasoning patterns with embeddings for semantic similarity search.
- * Integrates with ReasoningBank WASM for high-performance pattern matching.
- *
- * Pattern Structure:
- * - taskType: Type of task (e.g., "code_review", "data_analysis")
- * - approach: Description of the reasoning approach used
- * - successRate: Success rate of this pattern (0-1)
- * - embedding: Vector embedding of the pattern for similarity search
- * - metadata: Additional contextual information
- *
- * AgentDB v2 Migration:
- * - Uses VectorBackend abstraction for 8x faster search (RuVector/hnswlib)
- * - Optional GNN enhancement via LearningBackend
- * - 100% backward compatible with v1 API
- * - New features: useGNN option, recordOutcome for learning
- */
-import { cosineSimilarity } from '../utils/vector-math.js';
-export class ReasoningBank {
-    db;
-    embedder;
-    cache;
-    // v2: Optional vector backend (uses legacy if not provided)
-    vectorBackend;
-    learningBackend;
-    // Maps pattern ID (number) to vector backend ID (string) for hybrid mode
-    idMapping = new Map();
-    nextVectorId = 0;
-    /**
-     * Constructor supports both legacy (v1) and new (v2) modes
-     *
-     * Legacy mode (v1 - backward compatible):
-     *   new ReasoningBank(db, embedder)
-     *
-     * New mode (v2 - with VectorBackend):
-     *   new ReasoningBank(db, embedder, vectorBackend, learningBackend?)
-     */
-    constructor(db, embedder, vectorBackend, learningBackend) {
-        this.db = db;
-        this.embedder = embedder;
-        this.vectorBackend = vectorBackend;
-        this.learningBackend = learningBackend;
-        this.cache = new Map();
-        this.initializeSchema();
-    }
-    /**
-     * Initialize reasoning patterns schema
-     */
-    initializeSchema() {
-        // Create patterns table
-        this.db.exec(`
-      CREATE TABLE IF NOT EXISTS reasoning_patterns (
-        id INTEGER PRIMARY KEY AUTOINCREMENT,
-        ts INTEGER DEFAULT (strftime('%s', 'now')),
-        task_type TEXT NOT NULL,
-        approach TEXT NOT NULL,
-        success_rate REAL NOT NULL DEFAULT 0.0,
-        uses INTEGER DEFAULT 0,
-        avg_reward REAL DEFAULT 0.0,
-        tags TEXT,
-        metadata TEXT
-      );
-
-      CREATE INDEX IF NOT EXISTS idx_patterns_task_type ON reasoning_patterns(task_type);
-      CREATE INDEX IF NOT EXISTS idx_patterns_success_rate ON reasoning_patterns(success_rate);
-      CREATE INDEX IF NOT EXISTS idx_patterns_uses ON reasoning_patterns(uses);
-    `);
-        // Create pattern embeddings table
-        this.db.exec(`
-      CREATE TABLE IF NOT EXISTS pattern_embeddings (
-        pattern_id INTEGER PRIMARY KEY,
-        embedding BLOB NOT NULL,
-        FOREIGN KEY (pattern_id) REFERENCES reasoning_patterns(id) ON DELETE CASCADE
-      );
-    `);
-    }
-    /**
-     * Store a reasoning pattern with embedding
-     *
-     * v1 (legacy): Stores in SQLite with pattern_embeddings table
-     * v2 (VectorBackend): Stores metadata in SQLite, vectors in VectorBackend
-     */
-    async storePattern(pattern) {
-        // Generate embedding from approach text
-        const embedding = await this.embedder.embed(`${pattern.taskType}: ${pattern.approach}`);
-        // Insert pattern metadata into SQLite
-        const stmt = this.db.prepare(`
-      INSERT INTO reasoning_patterns (
-        task_type, approach, success_rate, uses, avg_reward, tags, metadata
-      ) VALUES (?, ?, ?, ?, ?, ?, ?)
-    `);
-        const result = stmt.run(pattern.taskType, pattern.approach, pattern.successRate, pattern.uses || 0, pattern.avgReward || 0.0, pattern.tags ? JSON.stringify(pattern.tags) : null, pattern.metadata ? JSON.stringify(pattern.metadata) : null);
-        const patternId = result.lastInsertRowid;
-        // Store embedding based on mode
-        if (this.vectorBackend) {
-            // v2: Use VectorBackend for high-performance search
-            try {
-                const vectorId = `pattern_${this.nextVectorId++}`;
-                this.idMapping.set(patternId, vectorId);
-                this.vectorBackend.insert(vectorId, embedding, {
-                    patternId,
-                    taskType: pattern.taskType,
-                    successRate: pattern.successRate,
-                });
-            }
-            catch {
-                // VectorBackend insert failed — fall back to SQLite storage
-                this.storePatternEmbedding(patternId, embedding);
-            }
-        }
-        else {
-            // v1: Use legacy SQLite storage (backward compatible)
-            this.storePatternEmbedding(patternId, embedding);
-        }
-        // Invalidate cache
-        this.cache.clear();
-        return patternId;
-    }
-    /**
-     * Store pattern embedding
-     */
-    storePatternEmbedding(patternId, embedding) {
-        const blob = Buffer.from(embedding.buffer);
-        const stmt = this.db.prepare(`
-      INSERT OR REPLACE INTO pattern_embeddings (pattern_id, embedding)
-      VALUES (?, ?)
-    `);
-        stmt.run(patternId, blob);
-    }
-    /**
-     * Search patterns by semantic similarity
-     *
-     * v1 (legacy): Uses SQLite with cosine similarity computation
-     * v2 (VectorBackend): Uses high-performance vector search (8x faster)
-     * v2 + GNN: Optionally enhances query with learned patterns
-     */
-    async searchPatterns(query) {
-        const k = query.k || 10;
-        const threshold = query.threshold || 0.0;
-        // Generate embedding if task string provided (v1 API compatibility)
-        let queryEmbedding;
-        if (query.task && !query.taskEmbedding) {
-            queryEmbedding = await this.embedder.embed(query.task);
-        }
-        else if (query.taskEmbedding) {
-            queryEmbedding = query.taskEmbedding;
-        }
-        else {
-            throw new Error('PatternSearchQuery must provide either task (v1) or taskEmbedding (v2)');
-        }
-        // Create enriched query with embedding (ensure taskEmbedding is always defined)
-        const enrichedQuery = {
-            ...query,
-            taskEmbedding: queryEmbedding
-        };
-        // Use VectorBackend if available (v2 mode)
-        if (this.vectorBackend) {
-            return this.searchPatternsV2(enrichedQuery);
-        }
-        // Legacy v1 search (100% backward compatible)
-        return this.searchPatternsLegacy(enrichedQuery);
-    }
-    /**
-     * v2: Search using VectorBackend with optional GNN enhancement
-     */
-    async searchPatternsV2(query) {
-        const k = query.k || 10;
-        const threshold = query.threshold || 0.0;
-        let queryEmbedding = query.taskEmbedding;
-        try {
-            // Optional: Apply GNN enhancement
-            if (query.useGNN && this.learningBackend) {
-                // Get initial candidates for GNN context
-                const candidates = this.vectorBackend.search(queryEmbedding, k * 3, { threshold: 0.0 });
-                if (candidates.length > 0) {
-                    // Retrieve neighbor embeddings for GNN
-                    const neighborEmbeddings = await this.getEmbeddingsForVectorIds(candidates.map(c => c.id));
-                    const weights = candidates.map(c => c.similarity);
-                    // Enhance query using GNN
-                    queryEmbedding = this.learningBackend.enhance(queryEmbedding, neighborEmbeddings, weights);
-                }
-            }
-            // Perform vector search
-            const results = this.vectorBackend.search(queryEmbedding, k, { threshold });
-            // Hydrate with metadata from SQLite
-            return this.hydratePatterns(results);
-        }
-        catch {
-            // VectorBackend search failed — fall back to legacy SQLite search
-            return this.searchPatternsLegacy(query);
-        }
-    }
-    /**
-     * v1: Legacy search using SQLite (backward compatible)
-     */
-    async searchPatternsLegacy(query) {
-        const k = query.k || 10;
-        const threshold = query.threshold || 0.0;
-        // Build WHERE clause for filters
-        const conditions = [];
-        const params = [];
-        if (query.filters?.taskType) {
-            conditions.push('rp.task_type = ?');
-            params.push(query.filters.taskType);
-        }
-        if (query.filters?.minSuccessRate !== undefined) {
-            conditions.push('rp.success_rate >= ?');
-            params.push(query.filters.minSuccessRate);
-        }
-        if (query.filters?.tags && query.filters.tags.length > 0) {
-            // Check if any of the tags match
-            const tagConditions = query.filters.tags.map(() => 'rp.tags LIKE ?').join(' OR ');
-            conditions.push(`(${tagConditions})`);
-            query.filters.tags.forEach(tag => {
-                params.push(`%"${tag}"%`);
-            });
-        }
-        const whereClause = conditions.length > 0
-            ? `WHERE ${conditions.join(' AND ')}`
-            : '';
-        // Retrieve all candidate patterns
-        const stmt = this.db.prepare(`
-      SELECT
-        rp.id,
-        rp.ts,
-        rp.task_type,
-        rp.approach,
-        rp.success_rate,
-        rp.uses,
-        rp.avg_reward,
-        rp.tags,
-        rp.metadata,
-        pe.embedding
-      FROM reasoning_patterns rp
-      JOIN pattern_embeddings pe ON rp.id = pe.pattern_id
-      ${whereClause}
-    `);
-        const rows = stmt.all(...params);
-        // Calculate similarities
-        const candidates = rows.map(row => {
-            const embedding = new Float32Array(row.embedding.buffer, row.embedding.byteOffset, row.embedding.byteLength / 4);
-            const similarity = cosineSimilarity(query.taskEmbedding, embedding);
-            return {
-                id: row.id,
-                taskType: row.task_type,
-                approach: row.approach,
-                successRate: row.success_rate,
-                uses: row.uses,
-                avgReward: row.avg_reward,
-                tags: row.tags ? JSON.parse(row.tags) : [],
-                metadata: row.metadata ? JSON.parse(row.metadata) : {},
-                createdAt: row.ts,
-                embedding,
-                similarity,
-            };
-        });
-        // Filter by threshold and sort by similarity
-        const filtered = candidates
-            .filter(c => c.similarity >= threshold)
-            .sort((a, b) => b.similarity - a.similarity)
-            .slice(0, k);
-        return filtered;
-    }
-    /**
-     * Hydrate search results with metadata from SQLite
-     */
-    hydratePatterns(results) {
-        // Prepare statement OUTSIDE loop for better-sqlite3 best practice
-        const stmt = this.db.prepare(`
-      SELECT * FROM reasoning_patterns WHERE id = ?
-    `);
-        return results.map(result => {
-            const patternId = result.metadata?.patternId;
-            if (!patternId) {
-                throw new Error(`VectorBackend result missing patternId: ${result.id}`);
-            }
-            const row = stmt.get(patternId);
-            if (!row) {
-                throw new Error(`Pattern ${patternId} not found in database`);
-            }
-            return {
-                id: row.id,
-                taskType: row.task_type,
-                approach: row.approach,
-                successRate: row.success_rate,
-                uses: row.uses,
-                avgReward: row.avg_reward,
-                tags: row.tags ? JSON.parse(row.tags) : [],
-                metadata: row.metadata ? JSON.parse(row.metadata) : {},
-                createdAt: row.ts,
-                similarity: result.similarity,
-            };
-        });
-    }
-    /**
-     * Get embeddings for vector IDs (for GNN)
-     */
-    async getEmbeddingsForVectorIds(vectorIds) {
-        // In a full implementation, this would retrieve embeddings from VectorBackend
-        // For now, we regenerate them from the database
-        const embeddings = [];
-        for (const vectorId of vectorIds) {
-            // Find pattern ID from mapping
-            let patternId;
-            for (const [pid, vid] of this.idMapping.entries()) {
-                if (vid === vectorId) {
-                    patternId = pid;
-                    break;
-                }
-            }
-            if (patternId) {
-                const pattern = this.getPattern(patternId);
-                if (pattern?.approach) {
-                    const embedding = await this.embedder.embed(`${pattern.taskType}: ${pattern.approach}`);
-                    embeddings.push(embedding);
-                }
-            }
-        }
-        return embeddings;
-    }
-    /**
-     * Get pattern statistics
-     */
-    getPatternStats() {
-        // Check cache first
-        const cacheKey = 'pattern_stats';
-        if (this.cache.has(cacheKey)) {
-            return this.cache.get(cacheKey);
-        }
-        // Total patterns
-        const totalRow = this.db.prepare('SELECT COUNT(*) as count FROM reasoning_patterns').get();
-        // Average success rate and uses
-        const avgRow = this.db.prepare(`
-      SELECT
-        AVG(success_rate) as avg_success_rate,
-        AVG(uses) as avg_uses
-      FROM reasoning_patterns
-    `).get();
-        // Top task types
-        const topTaskTypes = this.db.prepare(`
-      SELECT
-        task_type,
-        COUNT(*) as count
-      FROM reasoning_patterns
-      GROUP BY task_type
-      ORDER BY count DESC
-      LIMIT 10
-    `).all();
-        // Recent patterns (last 7 days)
-        const recentRow = this.db.prepare(`
-      SELECT COUNT(*) as count
-      FROM reasoning_patterns
-      WHERE ts >= strftime('%s', 'now', '-7 days')
-    `).get();
-        // High performing patterns (success_rate >= 0.8)
-        const highPerfRow = this.db.prepare(`
-      SELECT COUNT(*) as count
-      FROM reasoning_patterns
-      WHERE success_rate >= 0.8
-    `).get();
-        const stats = {
-            totalPatterns: totalRow.count,
-            avgSuccessRate: avgRow.avg_success_rate || 0,
-            avgUses: avgRow.avg_uses || 0,
-            topTaskTypes: topTaskTypes.map(row => ({
-                taskType: row.task_type,
-                count: row.count,
-            })),
-            recentPatterns: recentRow.count,
-            highPerformingPatterns: highPerfRow.count,
-        };
-        // Cache for 5 minutes
-        this.cache.set(cacheKey, stats);
-        setTimeout(() => this.cache.delete(cacheKey), 5 * 60 * 1000);
-        return stats;
-    }
-    /**
-     * Update pattern statistics after use
-     */
-    updatePatternStats(patternId, success, reward) {
-        const stmt = this.db.prepare(`
-      UPDATE reasoning_patterns
-      SET
-        uses = uses + 1,
-        success_rate = (success_rate * uses + ?) / (uses + 1),
-        avg_reward = (avg_reward * uses + ?) / (uses + 1)
-      WHERE id = ?
-    `);
-        stmt.run(success ? 1 : 0, reward, patternId);
-        // Invalidate cache
-        this.cache.clear();
-    }
-    /**
-     * Record pattern outcome for GNN learning (v2 feature)
-     *
-     * Updates pattern stats and adds training sample to LearningBackend
-     * for future GNN model improvements.
-     *
-     * @param patternId - Pattern ID to update
-     * @param success - Whether the pattern was successful
-     * @param reward - Optional reward value (default: 1 for success, 0 for failure)
-     */
-    async recordOutcome(patternId, success, reward) {
-        // Update pattern statistics
-        const actualReward = reward !== undefined ? reward : (success ? 1.0 : 0.0);
-        this.updatePatternStats(patternId, success, actualReward);
-        // Add to GNN training buffer if available
-        if (this.learningBackend) {
-            const pattern = this.getPattern(patternId);
-            if (pattern?.approach) {
-                const embedding = await this.embedder.embed(`${pattern.taskType}: ${pattern.approach}`);
-                this.learningBackend.addSample(embedding, success);
-            }
-        }
-    }
-    /**
-     * Train GNN model on collected samples (v2 feature)
-     *
-     * Trains the learning backend using accumulated pattern outcomes.
-     * Requires LearningBackend to be configured.
-     *
-     * @param options - Training options (epochs, batchSize)
-     * @returns Training results with epochs and final loss
-     * @throws Error if LearningBackend not available
-     */
-    async trainGNN(options) {
-        if (!this.learningBackend) {
-            throw new Error('GNN learning not available. Initialize ReasoningBank with LearningBackend.');
-        }
-        return this.learningBackend.train(options);
-    }
-    /**
-     * Get pattern by ID
-     */
-    getPattern(patternId) {
-        const stmt = this.db.prepare(`
-      SELECT
-        rp.id,
-        rp.ts,
-        rp.task_type,
-        rp.approach,
-        rp.success_rate,
-        rp.uses,
-        rp.avg_reward,
-        rp.tags,
-        rp.metadata,
-        pe.embedding
-      FROM reasoning_patterns rp
-      LEFT JOIN pattern_embeddings pe ON rp.id = pe.pattern_id
-      WHERE rp.id = ?
-    `);
-        const row = stmt.get(patternId);
-        if (!row)
-            return null;
-        return {
-            id: row.id,
-            taskType: row.task_type,
-            approach: row.approach,
-            successRate: row.success_rate,
-            uses: row.uses,
-            avgReward: row.avg_reward,
-            tags: row.tags ? JSON.parse(row.tags) : [],
-            metadata: row.metadata ? JSON.parse(row.metadata) : {},
-            createdAt: row.ts,
-            embedding: row.embedding
-                ? new Float32Array(row.embedding.buffer, row.embedding.byteOffset, row.embedding.byteLength / 4)
-                : undefined,
-        };
-    }
-    /**
-     * Delete pattern by ID
-     */
-    deletePattern(patternId) {
-        const stmt = this.db.prepare('DELETE FROM reasoning_patterns WHERE id = ?');
-        const result = stmt.run(patternId);
-        // Invalidate cache
-        this.cache.clear();
-        return result.changes > 0;
-    }
-    /**
-     * Clear query cache
-     */
-    clearCache() {
-        this.cache.clear();
-    }
-}
-//# sourceMappingURL=ReasoningBank.js.map

package/dist/packages/agentdb/src/controllers/ReasoningBank.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"ReasoningBank.js","sourceRoot":"","sources":["../../../../../../packages/agentdb/src/controllers/ReasoningBank.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAMH,OAAO,EAAE,gBAAgB,EAAE,MAAM,yBAAyB,CAAC;AAgE3D,MAAM,OAAO,aAAa;IAChB,EAAE,CAAW;IACb,QAAQ,CAAmB;IAC3B,KAAK,CAAmB;IAEhC,4DAA4D;IACpD,aAAa,CAAiB;IAC9B,eAAe,CAAmB;IAE1C,yEAAyE;IACjE,SAAS,GAAwB,IAAI,GAAG,EAAE,CAAC;IAC3C,YAAY,GAAG,CAAC,CAAC;IAEzB;;;;;;;;OAQG;IACH,YACE,EAAY,EACZ,QAA0B,EAC1B,aAA6B,EAC7B,eAAiC;QAEjC,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC;QACb,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACzB,IAAI,CAAC,aAAa,GAAG,aAAa,CAAC;QACnC,IAAI,CAAC,eAAe,GAAG,eAAe,CAAC;QACvC,IAAI,CAAC,KAAK,GAAG,IAAI,GAAG,EAAE,CAAC;QACvB,IAAI,CAAC,gBAAgB,EAAE,CAAC;IAC1B,CAAC;IAED;;OAEG;IACK,gBAAgB;QACtB,wBAAwB;QACxB,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC;;;;;;;;;;;;;;;;KAgBZ,CAAC,CAAC;QAEH,kCAAkC;QAClC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC;;;;;;KAMZ,CAAC,CAAC;IACL,CAAC;IAED;;;;;OAKG;IACH,KAAK,CAAC,YAAY,CAAC,OAAyB;QAC1C,wCAAwC;QACxC,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,KAAK,CACzC,GAAG,OAAO,CAAC,QAAQ,KAAK,OAAO,CAAC,QAAQ,EAAE,CAC3C,CAAC;QAEF,sCAAsC;QACtC,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;KAI5B,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CACrB,OAAO,CAAC,QAAQ,EAChB,OAAO,CAAC,QAAQ,EAChB,OAAO,CAAC,WAAW,EACnB,OAAO,CAAC,IAAI,IAAI,CAAC,EACjB,OAAO,CAAC,SAAS,IAAI,GAAG,EACxB,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,EAClD,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,CAC3D,CAAC;QAEF,MAAM,SAAS,GAAG,MAAM,CAAC,eAAyB,CAAC;QAEnD,gCAAgC;QAChC,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;YACvB,oDAAoD;YACpD,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,WAAW,IAAI,CAAC,YAAY,EAAE,EAAE,CAAC;gBAClD,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;gBAExC,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,QAAQ,EAAE,SAAS,EAAE;oBAC7C,SAAS;oBACT,QAAQ,EAAE,OAAO,CAAC,QAAQ;oBAC1B,WAAW,EAAE,OAAO,CAAC,WAAW;iBACjC,CAAC,CAAC;YACL,CAAC;YAAC,MAAM,CAAC;gBACP,4DAA4D;gBAC5D,IAAI,CAAC,qBAAqB,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC;YACnD,CAAC;QACH,CAAC;aAAM,CAAC;YACN,sDAAsD;YACtD,IAAI,CAAC,qBAAqB,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC;QACnD,CAAC;QAED,mBAAmB;QACnB,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;QAEnB,OAAO,SAAS,CAAC;IACnB,CAAC;IAED;;OAEG;IACK,qBAAqB,CAAC,SAAiB,EAAE,SAAuB;QACtE,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;QAE3C,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;KAG5B,CAAC,CAAC;QAEH,IAAI,CAAC,GAAG,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;IAC5B,CAAC;IAED;;;;;;OAMG;IACH,KAAK,CAAC,cAAc,CAAC,KAAyB;QAC5C,MAAM,CAAC,GAAG,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC;QACxB,MAAM,SAAS,GAAG,KAAK,CAAC,SAAS,IAAI,GAAG,CAAC;QAEzC,oEAAoE;QACpE,IAAI,cAA4B,CAAC;QACjC,IAAI,KAAK,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,CAAC;YACvC,cAAc,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QACzD,CAAC;aAAM,IAAI,KAAK,CAAC,aAAa,EAAE,CAAC;YAC/B,cAAc,GAAG,KAAK,CAAC,aAAa,CAAC;QACvC,CAAC;aAAM,CAAC;YACN,MAAM,IAAI,KAAK,CAAC,wEAAwE,CAAC,CAAC;QAC5F,CAAC;QAED,gFAAgF;QAChF,MAAM,aAAa,GAAyD;YAC1E,GAAG,KAAK;YACR,aAAa,EAAE,cAAc;SAC9B,CAAC;QAEF,2CAA2C;QAC3C,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;YACvB,OAAO,IAAI,CAAC,gBAAgB,CAAC,aAAa,CAAC,CAAC;QAC9C,CAAC;QAED,8CAA8C;QAC9C,OAAO,IAAI,CAAC,oBAAoB,CAAC,aAAa,CAAC,CAAC;IAClD,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,gBAAgB,CAAC,KAA2D;QACxF,MAAM,CAAC,GAAG,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC;QACxB,MAAM,SAAS,GAAG,KAAK,CAAC,SAAS,IAAI,GAAG,CAAC;QACzC,IAAI,cAAc,GAAG,KAAK,CAAC,aAAa,CAAC;QAEzC,IAAI,CAAC;YACH,kCAAkC;YAClC,IAAI,KAAK,CAAC,MAAM,IAAI,IAAI,CAAC,eAAe,EAAE,CAAC;gBACzC,yCAAyC;gBACzC,MAAM,UAAU,GAAG,IAAI,CAAC,aAAc,CAAC,MAAM,CAAC,cAAc,EAAE,CAAC,GAAG,CAAC,EAAE,EAAE,SAAS,EAAE,GAAG,EAAE,CAAC,CAAC;gBAEzF,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;oBAC1B,uCAAuC;oBACvC,MAAM,kBAAkB,GAAG,MAAM,IAAI,CAAC,yBAAyB,CAC7D,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAC1B,CAAC;oBACF,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC;oBAElD,0BAA0B;oBAC1B,cAAc,GAAG,IAAI,CAAC,eAAe,CAAC,OAAO,CAAC,cAAc,EAAE,kBAAkB,EAAE,OAAO,CAAC,CAAC;gBAC7F,CAAC;YACH,CAAC;YAED,wBAAwB;YACxB,MAAM,OAAO,GAAG,IAAI,CAAC,aAAc,CAAC,MAAM,CAAC,cAAc,EAAE,CAAC,EAAE,EAAE,SAAS,EAAE,CAAC,CAAC;YAE7E,oCAAoC;YACpC,OAAO,IAAI,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC;QACvC,CAAC;QAAC,MAAM,CAAC;YACP,kEAAkE;YAClE,OAAO,IAAI,CAAC,oBAAoB,CAAC,KAAK,CAAC,CAAC;QAC1C,CAAC;IACH,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,oBAAoB,CAAC,KAA2D;QAC5F,MAAM,CAAC,GAAG,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC;QACxB,MAAM,SAAS,GAAG,KAAK,CAAC,SAAS,IAAI,GAAG,CAAC;QAEzC,iCAAiC;QACjC,MAAM,UAAU,GAAa,EAAE,CAAC;QAChC,MAAM,MAAM,GAAU,EAAE,CAAC;QAEzB,IAAI,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC;YAC5B,UAAU,CAAC,IAAI,CAAC,kBAAkB,CAAC,CAAC;YACpC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;QACtC,CAAC;QAED,IAAI,KAAK,CAAC,OAAO,EAAE,cAAc,KAAK,SAAS,EAAE,CAAC;YAChD,UAAU,CAAC,IAAI,CAAC,sBAAsB,CAAC,CAAC;YACxC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC;QAC5C,CAAC;QAED,IAAI,KAAK,CAAC,OAAO,EAAE,IAAI,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACzD,iCAAiC;YACjC,MAAM,aAAa,GAAG,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,gBAAgB,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YAClF,UAAU,CAAC,IAAI,CAAC,IAAI,aAAa,GAAG,CAAC,CAAC;YACtC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE;gBAC/B,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,CAAC;YAC5B,CAAC,CAAC,CAAC;QACL,CAAC;QAED,MAAM,WAAW,GAAG,UAAU,CAAC,MAAM,GAAG,CAAC;YACvC,CAAC,CAAC,SAAS,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE;YACrC,CAAC,CAAC,EAAE,CAAC;QAEP,kCAAkC;QAClC,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;;;;;;;;QAczB,WAAW;KACd,CAAC,CAAC;QAEH,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,MAAM,CAAU,CAAC;QAE1C,yBAAyB;QACzB,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE;YAChC,MAAM,SAAS,GAAG,IAAI,YAAY,CAC/B,GAAG,CAAC,SAAoB,CAAC,MAAM,EAC/B,GAAG,CAAC,SAAoB,CAAC,UAAU,EACnC,GAAG,CAAC,SAAoB,CAAC,UAAU,GAAG,CAAC,CACzC,CAAC;YAEF,MAAM,UAAU,GAAG,gBAAgB,CAAC,KAAK,CAAC,aAAa,EAAE,SAAS,CAAC,CAAC;YAEpE,OAAO;gBACL,EAAE,EAAE,GAAG,CAAC,EAAE;gBACV,QAAQ,EAAE,GAAG,CAAC,SAAS;gBACvB,QAAQ,EAAE,GAAG,CAAC,QAAQ;gBACtB,WAAW,EAAE,GAAG,CAAC,YAAY;gBAC7B,IAAI,EAAE,GAAG,CAAC,IAAI;gBACd,SAAS,EAAE,GAAG,CAAC,UAAU;gBACzB,IAAI,EAAE,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE;gBAC1C,QAAQ,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE;gBACtD,SAAS,EAAE,GAAG,CAAC,EAAE;gBACjB,SAAS;gBACT,UAAU;aACX,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,6CAA6C;QAC7C,MAAM,QAAQ,GAAG,UAAU;aACxB,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,UAAU,IAAI,SAAS,CAAC;aACtC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,GAAG,CAAC,CAAC,UAAU,CAAC;aAC3C,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QAEf,OAAO,QAAQ,CAAC;IAClB,CAAC;IAED;;OAEG;IACK,eAAe,CAAC,OAAuB;QAC7C,kEAAkE;QAClE,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;KAE5B,CAAC,CAAC;QAEH,OAAO,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE;YAC1B,MAAM,SAAS,GAAG,MAAM,CAAC,QAAQ,EAAE,SAAS,CAAC;YAC7C,IAAI,CAAC,SAAS,EAAE,CAAC;gBACf,MAAM,IAAI,KAAK,CAAC,2CAA2C,MAAM,CAAC,EAAE,EAAE,CAAC,CAAC;YAC1E,CAAC;YAED,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,SAAS,CAAQ,CAAC;YAEvC,IAAI,CAAC,GAAG,EAAE,CAAC;gBACT,MAAM,IAAI,KAAK,CAAC,WAAW,SAAS,wBAAwB,CAAC,CAAC;YAChE,CAAC;YAED,OAAO;gBACL,EAAE,EAAE,GAAG,CAAC,EAAE;gBACV,QAAQ,EAAE,GAAG,CAAC,SAAS;gBACvB,QAAQ,EAAE,GAAG,CAAC,QAAQ;gBACtB,WAAW,EAAE,GAAG,CAAC,YAAY;gBAC7B,IAAI,EAAE,GAAG,CAAC,IAAI;gBACd,SAAS,EAAE,GAAG,CAAC,UAAU;gBACzB,IAAI,EAAE,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE;gBAC1C,QAAQ,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE;gBACtD,SAAS,EAAE,GAAG,CAAC,EAAE;gBACjB,UAAU,EAAE,MAAM,CAAC,UAAU;aAC9B,CAAC;QACJ,CAAC,CAAC,CAAC;IACL,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,yBAAyB,CAAC,SAAmB;QACzD,8EAA8E;QAC9E,gDAAgD;QAChD,MAAM,UAAU,GAAmB,EAAE,CAAC;QAEtC,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;YACjC,+BAA+B;YAC/B,IAAI,SAA6B,CAAC;YAClC,KAAK,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,IAAI,CAAC,SAAS,CAAC,OAAO,EAAE,EAAE,CAAC;gBAClD,IAAI,GAAG,KAAK,QAAQ,EAAE,CAAC;oBACrB,SAAS,GAAG,GAAG,CAAC;oBAChB,MAAM;gBACR,CAAC;YACH,CAAC;YAED,IAAI,SAAS,EAAE,CAAC;gBACd,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;gBAC3C,IAAI,OAAO,EAAE,QAAQ,EAAE,CAAC;oBACtB,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,KAAK,CACzC,GAAG,OAAO,CAAC,QAAQ,KAAK,OAAO,CAAC,QAAQ,EAAE,CAC3C,CAAC;oBACF,UAAU,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBAC7B,CAAC;YACH,CAAC;QACH,CAAC;QAED,OAAO,UAAU,CAAC;IACpB,CAAC;IAED;;OAEG;IACH,eAAe;QACb,oBAAoB;QACpB,MAAM,QAAQ,GAAG,eAAe,CAAC;QACjC,IAAI,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC7B,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;QAClC,CAAC;QAED,iBAAiB;QACjB,MAAM,QAAQ,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAC9B,kDAAkD,CACnD,CAAC,GAAG,EAAS,CAAC;QAEf,gCAAgC;QAChC,MAAM,MAAM,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;KAK9B,CAAC,CAAC,GAAG,EAAS,CAAC;QAEhB,iBAAiB;QACjB,MAAM,YAAY,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;;KAQpC,CAAC,CAAC,GAAG,EAAW,CAAC;QAElB,gCAAgC;QAChC,MAAM,SAAS,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;KAIjC,CAAC,CAAC,GAAG,EAAS,CAAC;QAEhB,iDAAiD;QACjD,MAAM,WAAW,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;KAInC,CAAC,CAAC,GAAG,EAAS,CAAC;QAEhB,MAAM,KAAK,GAAiB;YAC1B,aAAa,EAAE,QAAQ,CAAC,KAAK;YAC7B,cAAc,EAAE,MAAM,CAAC,gBAAgB,IAAI,CAAC;YAC5C,OAAO,EAAE,MAAM,CAAC,QAAQ,IAAI,CAAC;YAC7B,YAAY,EAAE,YAAY,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;gBACrC,QAAQ,EAAE,GAAG,CAAC,SAAS;gBACvB,KAAK,EAAE,GAAG,CAAC,KAAK;aACjB,CAAC,CAAC;YACH,cAAc,EAAE,SAAS,CAAC,KAAK;YAC/B,sBAAsB,EAAE,WAAW,CAAC,KAAK;SAC1C,CAAC;QAEF,sBAAsB;QACtB,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;QAChC,UAAU,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC;QAE7D,OAAO,KAAK,CAAC;IACf,CAAC;IAED;;OAEG;IACH,kBAAkB,CAChB,SAAiB,EACjB,OAAgB,EAChB,MAAc;QAEd,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;KAO5B,CAAC,CAAC;QAEH,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC;QAE7C,mBAAmB;QACnB,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;IACrB,CAAC;IAED;;;;;;;;;OASG;IACH,KAAK,CAAC,aAAa,CACjB,SAAiB,EACjB,OAAgB,EAChB,MAAe;QAEf,4BAA4B;QAC5B,MAAM,YAAY,GAAG,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QAC3E,IAAI,CAAC,kBAAkB,CAAC,SAAS,EAAE,OAAO,EAAE,YAAY,CAAC,CAAC;QAE1D,0CAA0C;QAC1C,IAAI,IAAI,CAAC,eAAe,EAAE,CAAC;YACzB,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;YAC3C,IAAI,OAAO,EAAE,QAAQ,EAAE,CAAC;gBACtB,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,KAAK,CACzC,GAAG,OAAO,CAAC,QAAQ,KAAK,OAAO,CAAC,QAAQ,EAAE,CAC3C,CAAC;gBACF,IAAI,CAAC,eAAe,CAAC,SAAS,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;YACrD,CAAC;QACH,CAAC;IACH,CAAC;IAED;;;;;;;;;OASG;IACH,KAAK,CAAC,QAAQ,CAAC,OAAiD;QAI9D,IAAI,CAAC,IAAI,CAAC,eAAe,EAAE,CAAC;YAC1B,MAAM,IAAI,KAAK,CAAC,4EAA4E,CAAC,CAAC;QAChG,CAAC;QAED,OAAO,IAAI,CAAC,eAAe,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;IAC7C,CAAC;IAED;;OAEG;IACH,UAAU,CAAC,SAAiB;QAC1B,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;;;;;;;;;KAe5B,CAAC,CAAC;QAEH,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,SAAS,CAAQ,CAAC;QACvC,IAAI,CAAC,GAAG;YAAE,OAAO,IAAI,CAAC;QAEtB,OAAO;YACL,EAAE,EAAE,GAAG,CAAC,EAAE;YACV,QAAQ,EAAE,GAAG,CAAC,SAAS;YACvB,QAAQ,EAAE,GAAG,CAAC,QAAQ;YACtB,WAAW,EAAE,GAAG,CAAC,YAAY;YAC7B,IAAI,EAAE,GAAG,CAAC,IAAI;YACd,SAAS,EAAE,GAAG,CAAC,UAAU;YACzB,IAAI,EAAE,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE;YAC1C,QAAQ,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE;YACtD,SAAS,EAAE,GAAG,CAAC,EAAE;YACjB,SAAS,EAAE,GAAG,CAAC,SAAS;gBACtB,CAAC,CAAC,IAAI,YAAY,CACd,GAAG,CAAC,SAAS,CAAC,MAAM,EACpB,GAAG,CAAC,SAAS,CAAC,UAAU,EACxB,GAAG,CAAC,SAAS,CAAC,UAAU,GAAG,CAAC,CAC7B;gBACH,CAAC,CAAC,SAAS;SACd,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,aAAa,CAAC,SAAiB;QAC7B,MAAM,IAAI,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,6CAA6C,CAAC,CAAC;QAC5E,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QAEnC,mBAAmB;QACnB,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;QAEnB,OAAO,MAAM,CAAC,OAAO,GAAG,CAAC,CAAC;IAC5B,CAAC;IAED;;OAEG;IACH,UAAU;QACR,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;IACrB,CAAC;CAEF","sourcesContent":["/**\n * ReasoningBank Controller - Pattern Storage and Retrieval\n *\n * Manages reasoning patterns with embeddings for semantic similarity search.\n * Integrates with ReasoningBank WASM for high-performance pattern matching.\n *\n * Pattern Structure:\n * - taskType: Type of task (e.g., \"code_review\", \"data_analysis\")\n * - approach: Description of the reasoning approach used\n * - successRate: Success rate of this pattern (0-1)\n * - embedding: Vector embedding of the pattern for similarity search\n * - metadata: Additional contextual information\n *\n * AgentDB v2 Migration:\n * - Uses VectorBackend abstraction for 8x faster search (RuVector/hnswlib)\n * - Optional GNN enhancement via LearningBackend\n * - 100% backward compatible with v1 API\n * - New features: useGNN option, recordOutcome for learning\n */\n\n// Database type from db-fallback\ntype Database = any;\nimport { EmbeddingService } from './EmbeddingService.js';\nimport type { VectorBackend, SearchResult } from '../backends/VectorBackend.js';\nimport { cosineSimilarity } from '../utils/vector-math.js';\n\nexport interface ReasoningPattern {\n  id?: number;\n  taskType: string;\n  approach: string;\n  successRate: number;\n  embedding?: Float32Array;\n  uses?: number;\n  avgReward?: number;\n  tags?: string[];\n  metadata?: Record<string, any>;\n  createdAt?: number;\n  similarity?: number; // Cosine similarity score (for search results)\n}\n\nexport interface PatternSearchQuery {\n  /** v1 API: Task string (will be embedded automatically) */\n  task?: string;\n  /** v2 API: Pre-computed embedding */\n  taskEmbedding?: Float32Array;\n  k?: number;\n  threshold?: number;\n  /** Enable GNN-based query enhancement (requires LearningBackend) */\n  useGNN?: boolean;\n  filters?: {\n    taskType?: string;\n    minSuccessRate?: number;\n    tags?: string[];\n  };\n}\n\nexport interface PatternStats {\n  totalPatterns: number;\n  avgSuccessRate: number;\n  avgUses: number;\n  topTaskTypes: Array<{ taskType: string; count: number }>;\n  recentPatterns: number;\n  highPerformingPatterns: number;\n}\n\n/**\n * Optional GNN Learning Backend for query enhancement\n */\nexport interface LearningBackend {\n  /**\n   * Enhance query embedding using GNN and neighbor context\n   */\n  enhance(query: Float32Array, neighbors: Float32Array[], weights: number[]): Float32Array;\n\n  /**\n   * Add training sample for future learning\n   */\n  addSample(embedding: Float32Array, success: boolean): void;\n\n  /**\n   * Train the GNN model\n   */\n  train(options?: { epochs?: number; batchSize?: number }): Promise<{\n    epochs: number;\n    finalLoss: number;\n  }>;\n}\n\nexport class ReasoningBank {\n  private db: Database;\n  private embedder: EmbeddingService;\n  private cache: Map<string, any>;\n\n  // v2: Optional vector backend (uses legacy if not provided)\n  private vectorBackend?: VectorBackend;\n  private learningBackend?: LearningBackend;\n\n  // Maps pattern ID (number) to vector backend ID (string) for hybrid mode\n  private idMapping: Map<number, string> = new Map();\n  private nextVectorId = 0;\n\n  /**\n   * Constructor supports both legacy (v1) and new (v2) modes\n   *\n   * Legacy mode (v1 - backward compatible):\n   *   new ReasoningBank(db, embedder)\n   *\n   * New mode (v2 - with VectorBackend):\n   *   new ReasoningBank(db, embedder, vectorBackend, learningBackend?)\n   */\n  constructor(\n    db: Database,\n    embedder: EmbeddingService,\n    vectorBackend?: VectorBackend,\n    learningBackend?: LearningBackend\n  ) {\n    this.db = db;\n    this.embedder = embedder;\n    this.vectorBackend = vectorBackend;\n    this.learningBackend = learningBackend;\n    this.cache = new Map();\n    this.initializeSchema();\n  }\n\n  /**\n   * Initialize reasoning patterns schema\n   */\n  private initializeSchema(): void {\n    // Create patterns table\n    this.db.exec(`\n      CREATE TABLE IF NOT EXISTS reasoning_patterns (\n        id INTEGER PRIMARY KEY AUTOINCREMENT,\n        ts INTEGER DEFAULT (strftime('%s', 'now')),\n        task_type TEXT NOT NULL,\n        approach TEXT NOT NULL,\n        success_rate REAL NOT NULL DEFAULT 0.0,\n        uses INTEGER DEFAULT 0,\n        avg_reward REAL DEFAULT 0.0,\n        tags TEXT,\n        metadata TEXT\n      );\n\n      CREATE INDEX IF NOT EXISTS idx_patterns_task_type ON reasoning_patterns(task_type);\n      CREATE INDEX IF NOT EXISTS idx_patterns_success_rate ON reasoning_patterns(success_rate);\n      CREATE INDEX IF NOT EXISTS idx_patterns_uses ON reasoning_patterns(uses);\n    `);\n\n    // Create pattern embeddings table\n    this.db.exec(`\n      CREATE TABLE IF NOT EXISTS pattern_embeddings (\n        pattern_id INTEGER PRIMARY KEY,\n        embedding BLOB NOT NULL,\n        FOREIGN KEY (pattern_id) REFERENCES reasoning_patterns(id) ON DELETE CASCADE\n      );\n    `);\n  }\n\n  /**\n   * Store a reasoning pattern with embedding\n   *\n   * v1 (legacy): Stores in SQLite with pattern_embeddings table\n   * v2 (VectorBackend): Stores metadata in SQLite, vectors in VectorBackend\n   */\n  async storePattern(pattern: ReasoningPattern): Promise<number> {\n    // Generate embedding from approach text\n    const embedding = await this.embedder.embed(\n      `${pattern.taskType}: ${pattern.approach}`\n    );\n\n    // Insert pattern metadata into SQLite\n    const stmt = this.db.prepare(`\n      INSERT INTO reasoning_patterns (\n        task_type, approach, success_rate, uses, avg_reward, tags, metadata\n      ) VALUES (?, ?, ?, ?, ?, ?, ?)\n    `);\n\n    const result = stmt.run(\n      pattern.taskType,\n      pattern.approach,\n      pattern.successRate,\n      pattern.uses || 0,\n      pattern.avgReward || 0.0,\n      pattern.tags ? JSON.stringify(pattern.tags) : null,\n      pattern.metadata ? JSON.stringify(pattern.metadata) : null\n    );\n\n    const patternId = result.lastInsertRowid as number;\n\n    // Store embedding based on mode\n    if (this.vectorBackend) {\n      // v2: Use VectorBackend for high-performance search\n      try {\n        const vectorId = `pattern_${this.nextVectorId++}`;\n        this.idMapping.set(patternId, vectorId);\n\n        this.vectorBackend.insert(vectorId, embedding, {\n          patternId,\n          taskType: pattern.taskType,\n          successRate: pattern.successRate,\n        });\n      } catch {\n        // VectorBackend insert failed — fall back to SQLite storage\n        this.storePatternEmbedding(patternId, embedding);\n      }\n    } else {\n      // v1: Use legacy SQLite storage (backward compatible)\n      this.storePatternEmbedding(patternId, embedding);\n    }\n\n    // Invalidate cache\n    this.cache.clear();\n\n    return patternId;\n  }\n\n  /**\n   * Store pattern embedding\n   */\n  private storePatternEmbedding(patternId: number, embedding: Float32Array): void {\n    const blob = Buffer.from(embedding.buffer);\n\n    const stmt = this.db.prepare(`\n      INSERT OR REPLACE INTO pattern_embeddings (pattern_id, embedding)\n      VALUES (?, ?)\n    `);\n\n    stmt.run(patternId, blob);\n  }\n\n  /**\n   * Search patterns by semantic similarity\n   *\n   * v1 (legacy): Uses SQLite with cosine similarity computation\n   * v2 (VectorBackend): Uses high-performance vector search (8x faster)\n   * v2 + GNN: Optionally enhances query with learned patterns\n   */\n  async searchPatterns(query: PatternSearchQuery): Promise<ReasoningPattern[]> {\n    const k = query.k || 10;\n    const threshold = query.threshold || 0.0;\n\n    // Generate embedding if task string provided (v1 API compatibility)\n    let queryEmbedding: Float32Array;\n    if (query.task && !query.taskEmbedding) {\n      queryEmbedding = await this.embedder.embed(query.task);\n    } else if (query.taskEmbedding) {\n      queryEmbedding = query.taskEmbedding;\n    } else {\n      throw new Error('PatternSearchQuery must provide either task (v1) or taskEmbedding (v2)');\n    }\n\n    // Create enriched query with embedding (ensure taskEmbedding is always defined)\n    const enrichedQuery: PatternSearchQuery & { taskEmbedding: Float32Array } = {\n      ...query,\n      taskEmbedding: queryEmbedding\n    };\n\n    // Use VectorBackend if available (v2 mode)\n    if (this.vectorBackend) {\n      return this.searchPatternsV2(enrichedQuery);\n    }\n\n    // Legacy v1 search (100% backward compatible)\n    return this.searchPatternsLegacy(enrichedQuery);\n  }\n\n  /**\n   * v2: Search using VectorBackend with optional GNN enhancement\n   */\n  private async searchPatternsV2(query: PatternSearchQuery & { taskEmbedding: Float32Array }): Promise<ReasoningPattern[]> {\n    const k = query.k || 10;\n    const threshold = query.threshold || 0.0;\n    let queryEmbedding = query.taskEmbedding;\n\n    try {\n      // Optional: Apply GNN enhancement\n      if (query.useGNN && this.learningBackend) {\n        // Get initial candidates for GNN context\n        const candidates = this.vectorBackend!.search(queryEmbedding, k * 3, { threshold: 0.0 });\n\n        if (candidates.length > 0) {\n          // Retrieve neighbor embeddings for GNN\n          const neighborEmbeddings = await this.getEmbeddingsForVectorIds(\n            candidates.map(c => c.id)\n          );\n          const weights = candidates.map(c => c.similarity);\n\n          // Enhance query using GNN\n          queryEmbedding = this.learningBackend.enhance(queryEmbedding, neighborEmbeddings, weights);\n        }\n      }\n\n      // Perform vector search\n      const results = this.vectorBackend!.search(queryEmbedding, k, { threshold });\n\n      // Hydrate with metadata from SQLite\n      return this.hydratePatterns(results);\n    } catch {\n      // VectorBackend search failed — fall back to legacy SQLite search\n      return this.searchPatternsLegacy(query);\n    }\n  }\n\n  /**\n   * v1: Legacy search using SQLite (backward compatible)\n   */\n  private async searchPatternsLegacy(query: PatternSearchQuery & { taskEmbedding: Float32Array }): Promise<ReasoningPattern[]> {\n    const k = query.k || 10;\n    const threshold = query.threshold || 0.0;\n\n    // Build WHERE clause for filters\n    const conditions: string[] = [];\n    const params: any[] = [];\n\n    if (query.filters?.taskType) {\n      conditions.push('rp.task_type = ?');\n      params.push(query.filters.taskType);\n    }\n\n    if (query.filters?.minSuccessRate !== undefined) {\n      conditions.push('rp.success_rate >= ?');\n      params.push(query.filters.minSuccessRate);\n    }\n\n    if (query.filters?.tags && query.filters.tags.length > 0) {\n      // Check if any of the tags match\n      const tagConditions = query.filters.tags.map(() => 'rp.tags LIKE ?').join(' OR ');\n      conditions.push(`(${tagConditions})`);\n      query.filters.tags.forEach(tag => {\n        params.push(`%\"${tag}\"%`);\n      });\n    }\n\n    const whereClause = conditions.length > 0\n      ? `WHERE ${conditions.join(' AND ')}`\n      : '';\n\n    // Retrieve all candidate patterns\n    const stmt = this.db.prepare(`\n      SELECT\n        rp.id,\n        rp.ts,\n        rp.task_type,\n        rp.approach,\n        rp.success_rate,\n        rp.uses,\n        rp.avg_reward,\n        rp.tags,\n        rp.metadata,\n        pe.embedding\n      FROM reasoning_patterns rp\n      JOIN pattern_embeddings pe ON rp.id = pe.pattern_id\n      ${whereClause}\n    `);\n\n    const rows = stmt.all(...params) as any[];\n\n    // Calculate similarities\n    const candidates = rows.map(row => {\n      const embedding = new Float32Array(\n        (row.embedding as Buffer).buffer,\n        (row.embedding as Buffer).byteOffset,\n        (row.embedding as Buffer).byteLength / 4\n      );\n\n      const similarity = cosineSimilarity(query.taskEmbedding, embedding);\n\n      return {\n        id: row.id,\n        taskType: row.task_type,\n        approach: row.approach,\n        successRate: row.success_rate,\n        uses: row.uses,\n        avgReward: row.avg_reward,\n        tags: row.tags ? JSON.parse(row.tags) : [],\n        metadata: row.metadata ? JSON.parse(row.metadata) : {},\n        createdAt: row.ts,\n        embedding,\n        similarity,\n      };\n    });\n\n    // Filter by threshold and sort by similarity\n    const filtered = candidates\n      .filter(c => c.similarity >= threshold)\n      .sort((a, b) => b.similarity - a.similarity)\n      .slice(0, k);\n\n    return filtered;\n  }\n\n  /**\n   * Hydrate search results with metadata from SQLite\n   */\n  private hydratePatterns(results: SearchResult[]): ReasoningPattern[] {\n    // Prepare statement OUTSIDE loop for better-sqlite3 best practice\n    const stmt = this.db.prepare(`\n      SELECT * FROM reasoning_patterns WHERE id = ?\n    `);\n\n    return results.map(result => {\n      const patternId = result.metadata?.patternId;\n      if (!patternId) {\n        throw new Error(`VectorBackend result missing patternId: ${result.id}`);\n      }\n\n      const row = stmt.get(patternId) as any;\n\n      if (!row) {\n        throw new Error(`Pattern ${patternId} not found in database`);\n      }\n\n      return {\n        id: row.id,\n        taskType: row.task_type,\n        approach: row.approach,\n        successRate: row.success_rate,\n        uses: row.uses,\n        avgReward: row.avg_reward,\n        tags: row.tags ? JSON.parse(row.tags) : [],\n        metadata: row.metadata ? JSON.parse(row.metadata) : {},\n        createdAt: row.ts,\n        similarity: result.similarity,\n      };\n    });\n  }\n\n  /**\n   * Get embeddings for vector IDs (for GNN)\n   */\n  private async getEmbeddingsForVectorIds(vectorIds: string[]): Promise<Float32Array[]> {\n    // In a full implementation, this would retrieve embeddings from VectorBackend\n    // For now, we regenerate them from the database\n    const embeddings: Float32Array[] = [];\n\n    for (const vectorId of vectorIds) {\n      // Find pattern ID from mapping\n      let patternId: number | undefined;\n      for (const [pid, vid] of this.idMapping.entries()) {\n        if (vid === vectorId) {\n          patternId = pid;\n          break;\n        }\n      }\n\n      if (patternId) {\n        const pattern = this.getPattern(patternId);\n        if (pattern?.approach) {\n          const embedding = await this.embedder.embed(\n            `${pattern.taskType}: ${pattern.approach}`\n          );\n          embeddings.push(embedding);\n        }\n      }\n    }\n\n    return embeddings;\n  }\n\n  /**\n   * Get pattern statistics\n   */\n  getPatternStats(): PatternStats {\n    // Check cache first\n    const cacheKey = 'pattern_stats';\n    if (this.cache.has(cacheKey)) {\n      return this.cache.get(cacheKey);\n    }\n\n    // Total patterns\n    const totalRow = this.db.prepare(\n      'SELECT COUNT(*) as count FROM reasoning_patterns'\n    ).get() as any;\n\n    // Average success rate and uses\n    const avgRow = this.db.prepare(`\n      SELECT\n        AVG(success_rate) as avg_success_rate,\n        AVG(uses) as avg_uses\n      FROM reasoning_patterns\n    `).get() as any;\n\n    // Top task types\n    const topTaskTypes = this.db.prepare(`\n      SELECT\n        task_type,\n        COUNT(*) as count\n      FROM reasoning_patterns\n      GROUP BY task_type\n      ORDER BY count DESC\n      LIMIT 10\n    `).all() as any[];\n\n    // Recent patterns (last 7 days)\n    const recentRow = this.db.prepare(`\n      SELECT COUNT(*) as count\n      FROM reasoning_patterns\n      WHERE ts >= strftime('%s', 'now', '-7 days')\n    `).get() as any;\n\n    // High performing patterns (success_rate >= 0.8)\n    const highPerfRow = this.db.prepare(`\n      SELECT COUNT(*) as count\n      FROM reasoning_patterns\n      WHERE success_rate >= 0.8\n    `).get() as any;\n\n    const stats: PatternStats = {\n      totalPatterns: totalRow.count,\n      avgSuccessRate: avgRow.avg_success_rate || 0,\n      avgUses: avgRow.avg_uses || 0,\n      topTaskTypes: topTaskTypes.map(row => ({\n        taskType: row.task_type,\n        count: row.count,\n      })),\n      recentPatterns: recentRow.count,\n      highPerformingPatterns: highPerfRow.count,\n    };\n\n    // Cache for 5 minutes\n    this.cache.set(cacheKey, stats);\n    setTimeout(() => this.cache.delete(cacheKey), 5 * 60 * 1000);\n\n    return stats;\n  }\n\n  /**\n   * Update pattern statistics after use\n   */\n  updatePatternStats(\n    patternId: number,\n    success: boolean,\n    reward: number\n  ): void {\n    const stmt = this.db.prepare(`\n      UPDATE reasoning_patterns\n      SET\n        uses = uses + 1,\n        success_rate = (success_rate * uses + ?) / (uses + 1),\n        avg_reward = (avg_reward * uses + ?) / (uses + 1)\n      WHERE id = ?\n    `);\n\n    stmt.run(success ? 1 : 0, reward, patternId);\n\n    // Invalidate cache\n    this.cache.clear();\n  }\n\n  /**\n   * Record pattern outcome for GNN learning (v2 feature)\n   *\n   * Updates pattern stats and adds training sample to LearningBackend\n   * for future GNN model improvements.\n   *\n   * @param patternId - Pattern ID to update\n   * @param success - Whether the pattern was successful\n   * @param reward - Optional reward value (default: 1 for success, 0 for failure)\n   */\n  async recordOutcome(\n    patternId: number,\n    success: boolean,\n    reward?: number\n  ): Promise<void> {\n    // Update pattern statistics\n    const actualReward = reward !== undefined ? reward : (success ? 1.0 : 0.0);\n    this.updatePatternStats(patternId, success, actualReward);\n\n    // Add to GNN training buffer if available\n    if (this.learningBackend) {\n      const pattern = this.getPattern(patternId);\n      if (pattern?.approach) {\n        const embedding = await this.embedder.embed(\n          `${pattern.taskType}: ${pattern.approach}`\n        );\n        this.learningBackend.addSample(embedding, success);\n      }\n    }\n  }\n\n  /**\n   * Train GNN model on collected samples (v2 feature)\n   *\n   * Trains the learning backend using accumulated pattern outcomes.\n   * Requires LearningBackend to be configured.\n   *\n   * @param options - Training options (epochs, batchSize)\n   * @returns Training results with epochs and final loss\n   * @throws Error if LearningBackend not available\n   */\n  async trainGNN(options?: { epochs?: number; batchSize?: number }): Promise<{\n    epochs: number;\n    finalLoss: number;\n  }> {\n    if (!this.learningBackend) {\n      throw new Error('GNN learning not available. Initialize ReasoningBank with LearningBackend.');\n    }\n\n    return this.learningBackend.train(options);\n  }\n\n  /**\n   * Get pattern by ID\n   */\n  getPattern(patternId: number): ReasoningPattern | null {\n    const stmt = this.db.prepare(`\n      SELECT\n        rp.id,\n        rp.ts,\n        rp.task_type,\n        rp.approach,\n        rp.success_rate,\n        rp.uses,\n        rp.avg_reward,\n        rp.tags,\n        rp.metadata,\n        pe.embedding\n      FROM reasoning_patterns rp\n      LEFT JOIN pattern_embeddings pe ON rp.id = pe.pattern_id\n      WHERE rp.id = ?\n    `);\n\n    const row = stmt.get(patternId) as any;\n    if (!row) return null;\n\n    return {\n      id: row.id,\n      taskType: row.task_type,\n      approach: row.approach,\n      successRate: row.success_rate,\n      uses: row.uses,\n      avgReward: row.avg_reward,\n      tags: row.tags ? JSON.parse(row.tags) : [],\n      metadata: row.metadata ? JSON.parse(row.metadata) : {},\n      createdAt: row.ts,\n      embedding: row.embedding\n        ? new Float32Array(\n            row.embedding.buffer,\n            row.embedding.byteOffset,\n            row.embedding.byteLength / 4\n          )\n        : undefined,\n    };\n  }\n\n  /**\n   * Delete pattern by ID\n   */\n  deletePattern(patternId: number): boolean {\n    const stmt = this.db.prepare('DELETE FROM reasoning_patterns WHERE id = ?');\n    const result = stmt.run(patternId);\n\n    // Invalidate cache\n    this.cache.clear();\n\n    return result.changes > 0;\n  }\n\n  /**\n   * Clear query cache\n   */\n  clearCache(): void {\n    this.cache.clear();\n  }\n\n}\n"]}

package/dist/packages/agentdb/src/controllers/ReflexionMemory.d.ts

@@ -1,124 +0,0 @@
-/**
- * ReflexionMemory - Episodic Replay Memory System
- *
- * Implements reflexion-style episodic replay for agent self-improvement.
- * Stores self-critiques and outcomes, retrieves relevant past experiences.
- *
- * Based on: "Reflexion: Language Agents with Verbal Reinforcement Learning"
- * https://arxiv.org/abs/2303.11366
- */
-type Database = any;
-import { EmbeddingService } from './EmbeddingService.js';
-import type { VectorBackend } from '../backends/VectorBackend.js';
-import type { LearningBackend } from '../backends/LearningBackend.js';
-import type { GraphBackend } from '../backends/GraphBackend.js';
-export interface Episode {
-    id?: number;
-    ts?: number;
-    sessionId: string;
-    task: string;
-    input?: string;
-    output?: string;
-    critique?: string;
-    reward: number;
-    success: boolean;
-    latencyMs?: number;
-    tokensUsed?: number;
-    tags?: string[];
-    metadata?: Record<string, any>;
-}
-export interface EpisodeWithEmbedding extends Episode {
-    embedding?: Float32Array;
-    similarity?: number;
-}
-export interface ReflexionQuery {
-    task: string;
-    currentState?: string;
-    k?: number;
-    minReward?: number;
-    onlyFailures?: boolean;
-    onlySuccesses?: boolean;
-    timeWindowDays?: number;
-}
-export declare class ReflexionMemory {
-    private db;
-    private embedder;
-    private vectorBackend?;
-    private learningBackend?;
-    private graphBackend?;
-    constructor(db: Database, embedder: EmbeddingService, vectorBackend?: VectorBackend, learningBackend?: LearningBackend, graphBackend?: GraphBackend);
-    /**
-     * Store a new episode with its critique and outcome
-     */
-    storeEpisode(episode: Episode): Promise<number>;
-    /**
-     * Retrieve relevant past episodes for a new task attempt
-     */
-    retrieveRelevant(query: ReflexionQuery): Promise<EpisodeWithEmbedding[]>;
-    /**
-     * Get statistics for a task
-     */
-    getTaskStats(task: string, timeWindowDays?: number): {
-        totalAttempts: number;
-        successRate: number;
-        avgReward: number;
-        avgLatency: number;
-        improvementTrend: number;
-    };
-    /**
-     * Build critique summary from similar failed episodes
-     */
-    getCritiqueSummary(query: ReflexionQuery): Promise<string>;
-    /**
-     * Get successful strategies for a task
-     */
-    getSuccessStrategies(query: ReflexionQuery): Promise<string>;
-    /**
-     * Get recent episodes for a session
-     */
-    getRecentEpisodes(sessionId: string, limit?: number): Promise<Episode[]>;
-    /**
-     * Prune low-quality episodes based on TTL and quality threshold
-     */
-    pruneEpisodes(config: {
-        minReward?: number;
-        maxAgeDays?: number;
-        keepMinPerTask?: number;
-    }): number;
-    private buildEpisodeText;
-    private storeEmbedding;
-    private serializeEmbedding;
-    private deserializeEmbedding;
-    /**
-     * Create graph node for episode with relationships
-     */
-    private createEpisodeGraphNode;
-    /**
-     * Enhance query embedding using GNN attention mechanism
-     */
-    private enhanceQueryWithGNN;
-    /**
-     * Get graph-based episode relationships
-     */
-    getEpisodeRelationships(episodeId: number): Promise<{
-        similar: number[];
-        session: string;
-        learnedFrom: number[];
-    }>;
-    /**
-     * Train GNN model on accumulated samples
-     */
-    trainGNN(options?: {
-        epochs?: number;
-    }): Promise<void>;
-    /**
-     * Get learning backend statistics
-     */
-    getLearningStats(): import("../backends/LearningBackend.js").LearningStats;
-    /**
-     * Get graph backend statistics
-     */
-    getGraphStats(): import("../backends/GraphBackend.js").GraphStats;
-}
-export {};
-//# sourceMappingURL=ReflexionMemory.d.ts.map