agentdb 1.0.12 → 1.1.0

package/dist/controllers/NightlyLearner.js

@@ -0,0 +1,383 @@
+/**
+ * Nightly Learner - Automated Causal Discovery and Consolidation
+ *
+ * Runs as a background job to:
+ * 1. Discover new causal edges from episode patterns
+ * 2. Run A/B experiments on promising hypotheses
+ * 3. Calculate uplift for completed experiments
+ * 4. Prune low-confidence edges
+ * 5. Update rerank weights based on performance
+ *
+ * Based on doubly robust learner:
+ * τ̂(x) = μ1(x) − μ0(x) + [a*(y−μ1(x)) / e(x)] − [(1−a)*(y−μ0(x)) / (1−e(x))]
+ */
+import { CausalMemoryGraph } from './CausalMemoryGraph.js';
+import { ReflexionMemory } from './ReflexionMemory.js';
+import { SkillLibrary } from './SkillLibrary.js';
+export class NightlyLearner {
+    config;
+    db;
+    causalGraph;
+    reflexion;
+    skillLibrary;
+    constructor(db, embedder, config = {
+        minSimilarity: 0.7,
+        minSampleSize: 30,
+        confidenceThreshold: 0.6,
+        upliftThreshold: 0.05,
+        pruneOldEdges: true,
+        edgeMaxAgeDays: 90,
+        autoExperiments: true,
+        experimentBudget: 10
+    }) {
+        this.config = config;
+        this.db = db;
+        this.causalGraph = new CausalMemoryGraph(db);
+        this.reflexion = new ReflexionMemory(db, embedder);
+        this.skillLibrary = new SkillLibrary(db, embedder);
+    }
+    /**
+     * Main learning job - runs all discovery and consolidation tasks
+     */
+    async run() {
+        console.log('\n🌙 Nightly Learner Starting...\n');
+        const startTime = Date.now();
+        const report = {
+            timestamp: startTime,
+            executionTimeMs: 0,
+            edgesDiscovered: 0,
+            edgesPruned: 0,
+            experimentsCompleted: 0,
+            experimentsCreated: 0,
+            avgUplift: 0,
+            avgConfidence: 0,
+            recommendations: []
+        };
+        try {
+            // Step 1: Discover new causal edges
+            console.log('📊 Discovering causal edges from episode patterns...');
+            report.edgesDiscovered = await this.discoverCausalEdges();
+            console.log(`   ✓ Discovered ${report.edgesDiscovered} new edges\n`);
+            // Step 2: Complete running experiments
+            console.log('🧪 Completing A/B experiments...');
+            report.experimentsCompleted = await this.completeExperiments();
+            console.log(`   ✓ Completed ${report.experimentsCompleted} experiments\n`);
+            // Step 3: Create new experiments (if enabled)
+            if (this.config.autoExperiments) {
+                console.log('🔬 Creating new A/B experiments...');
+                report.experimentsCreated = await this.createExperiments();
+                console.log(`   ✓ Created ${report.experimentsCreated} new experiments\n`);
+            }
+            // Step 4: Prune low-confidence edges
+            if (this.config.pruneOldEdges) {
+                console.log('🧹 Pruning low-confidence edges...');
+                report.edgesPruned = await this.pruneEdges();
+                console.log(`   ✓ Pruned ${report.edgesPruned} edges\n`);
+            }
+            // Step 5: Calculate statistics
+            const stats = this.calculateStats();
+            report.avgUplift = stats.avgUplift;
+            report.avgConfidence = stats.avgConfidence;
+            // Step 6: Generate recommendations
+            report.recommendations = this.generateRecommendations(report);
+            report.executionTimeMs = Date.now() - startTime;
+            console.log('✅ Nightly Learner Completed\n');
+            this.printReport(report);
+            return report;
+        }
+        catch (error) {
+            console.error('❌ Nightly Learner Failed:', error);
+            throw error;
+        }
+    }
+    /**
+     * Discover causal edges using doubly robust learner
+     *
+     * τ̂(x) = μ1(x) − μ0(x) + [a*(y−μ1(x)) / e(x)] − [(1−a)*(y−μ0(x)) / (1−e(x))]
+     *
+     * Where:
+     * - μ1(x) = outcome model for treatment
+     * - μ0(x) = outcome model for control
+     * - e(x) = propensity score (probability of treatment)
+     * - a = treatment indicator
+     * - y = observed outcome
+     */
+    async discover(config) {
+        return this.discoverCausalEdges();
+    }
+    async discoverCausalEdges() {
+        let discovered = 0;
+        // Find episode pairs with high similarity and temporal sequence
+        const candidatePairs = this.db.prepare(`
+      SELECT
+        e1.id as from_id,
+        e1.task as from_task,
+        e1.reward as from_reward,
+        e2.id as to_id,
+        e2.task as to_task,
+        e2.reward as to_reward,
+        e2.ts - e1.ts as time_diff
+      FROM episodes e1
+      JOIN episodes e2 ON e1.session_id = e2.session_id
+      WHERE e1.id != e2.id
+        AND e2.ts > e1.ts
+        AND e2.ts - e1.ts < 3600 -- Within 1 hour
+      ORDER BY e1.id, e2.ts
+      LIMIT 1000
+    `).all();
+        for (const pair of candidatePairs) {
+            // Check if edge already exists
+            const existing = this.db.prepare(`
+        SELECT id FROM causal_edges
+        WHERE from_memory_id = ? AND to_memory_id = ?
+      `).get(pair.from_id, pair.to_id);
+            if (existing)
+                continue;
+            // Calculate propensity score e(x) - probability of treatment
+            // Simplified: use frequency of from_task in session
+            const propensity = this.calculatePropensity(pair.from_id);
+            // Calculate outcome models μ1(x) and μ0(x)
+            const mu1 = this.calculateOutcomeModel(pair.from_task, true); // With treatment
+            const mu0 = this.calculateOutcomeModel(pair.from_task, false); // Without treatment
+            // Calculate doubly robust estimator
+            const a = 1; // This is a treated observation
+            const y = pair.to_reward;
+            const doublyRobustEstimate = (mu1 - mu0) + (a * (y - mu1) / propensity);
+            // Calculate confidence based on sample size and variance
+            const sampleSize = this.getSampleSize(pair.from_task);
+            const confidence = this.calculateConfidence(sampleSize, doublyRobustEstimate);
+            // Only add if meets thresholds
+            if (Math.abs(doublyRobustEstimate) >= this.config.upliftThreshold && confidence >= this.config.confidenceThreshold) {
+                const edge = {
+                    fromMemoryId: pair.from_id,
+                    fromMemoryType: 'episode',
+                    toMemoryId: pair.to_id,
+                    toMemoryType: 'episode',
+                    similarity: 0.8, // Simplified - would use embedding similarity in production
+                    uplift: doublyRobustEstimate,
+                    confidence,
+                    sampleSize,
+                    mechanism: `${pair.from_task} → ${pair.to_task} (doubly robust)`,
+                    metadata: {
+                        propensity,
+                        mu1,
+                        mu0,
+                        discoveredAt: Date.now()
+                    }
+                };
+                this.causalGraph.addCausalEdge(edge);
+                discovered++;
+            }
+        }
+        return discovered;
+    }
+    /**
+     * Calculate propensity score e(x) - probability of treatment given context
+     */
+    calculatePropensity(episodeId) {
+        const episode = this.db.prepare('SELECT task, session_id FROM episodes WHERE id = ?').get(episodeId);
+        // Count occurrences of this task type in session
+        const counts = this.db.prepare(`
+      SELECT
+        COUNT(*) as total,
+        SUM(CASE WHEN task = ? THEN 1 ELSE 0 END) as task_count
+      FROM episodes
+      WHERE session_id = ?
+    `).get(episode.task, episode.session_id);
+        const propensity = counts.task_count / Math.max(counts.total, 1);
+        // Clip to avoid division by zero
+        return Math.max(0.01, Math.min(0.99, propensity));
+    }
+    /**
+     * Calculate outcome model μ(x) - expected outcome given treatment status
+     */
+    calculateOutcomeModel(task, treated) {
+        // Get average reward for episodes with/without this task in their history
+        const avgReward = this.db.prepare(`
+      SELECT AVG(reward) as avg_reward
+      FROM episodes
+      WHERE ${treated ? '' : 'NOT'} EXISTS (
+        SELECT 1 FROM episodes e2
+        WHERE e2.session_id = episodes.session_id
+          AND e2.task = ?
+          AND e2.ts < episodes.ts
+      )
+    `).get(task);
+        return avgReward?.avg_reward || 0.5;
+    }
+    /**
+     * Get sample size for a task type
+     */
+    getSampleSize(task) {
+        const count = this.db.prepare(`
+      SELECT COUNT(*) as count
+      FROM episodes
+      WHERE task = ?
+    `).get(task);
+        return count.count;
+    }
+    /**
+     * Calculate confidence based on sample size and effect size
+     */
+    calculateConfidence(sampleSize, uplift) {
+        // Simplified confidence calculation
+        // In production, use proper statistical methods (bootstrap, etc.)
+        const sampleFactor = Math.min(sampleSize / 100, 1.0); // Max at 100 samples
+        const effectSizeFactor = Math.min(Math.abs(uplift) / 0.5, 1.0); // Max at 0.5 uplift
+        return sampleFactor * effectSizeFactor;
+    }
+    /**
+     * Complete running A/B experiments and calculate uplift
+     */
+    async completeExperiments() {
+        const runningExperiments = this.db.prepare(`
+      SELECT id, start_time, sample_size
+      FROM causal_experiments
+      WHERE status = 'running'
+        AND sample_size >= ?
+    `).all(this.config.minSampleSize);
+        let completed = 0;
+        for (const exp of runningExperiments) {
+            try {
+                this.causalGraph.calculateUplift(exp.id);
+                completed++;
+            }
+            catch (error) {
+                console.error(`   ⚠ Failed to calculate uplift for experiment ${exp.id}:`, error);
+            }
+        }
+        return completed;
+    }
+    /**
+     * Create new A/B experiments for promising hypotheses
+     */
+    async createExperiments() {
+        const currentExperiments = this.db.prepare(`
+      SELECT COUNT(*) as count
+      FROM causal_experiments
+      WHERE status = 'running'
+    `).get();
+        const available = this.config.experimentBudget - currentExperiments.count;
+        if (available <= 0) {
+            return 0;
+        }
+        // Find promising task pairs that don't have experiments yet
+        const candidates = this.db.prepare(`
+      SELECT DISTINCT
+        e1.task as treatment_task,
+        e1.id as treatment_id,
+        COUNT(e2.id) as potential_outcomes
+      FROM episodes e1
+      JOIN episodes e2 ON e1.session_id = e2.session_id
+      WHERE e2.ts > e1.ts
+        AND NOT EXISTS (
+          SELECT 1 FROM causal_experiments
+          WHERE treatment_id = e1.id
+        )
+      GROUP BY e1.task, e1.id
+      HAVING COUNT(e2.id) >= ?
+      ORDER BY COUNT(e2.id) DESC
+      LIMIT ?
+    `).all(this.config.minSampleSize, available);
+        let created = 0;
+        for (const candidate of candidates) {
+            const expId = this.causalGraph.createExperiment({
+                name: `Auto: ${candidate.treatment_task} Impact`,
+                hypothesis: `${candidate.treatment_task} affects downstream outcomes`,
+                treatmentId: candidate.treatment_id,
+                treatmentType: 'episode',
+                startTime: Date.now(),
+                sampleSize: 0,
+                status: 'running',
+                metadata: {
+                    autoGenerated: true,
+                    potentialOutcomes: candidate.potential_outcomes
+                }
+            });
+            created++;
+        }
+        return created;
+    }
+    /**
+     * Prune old or low-confidence edges
+     */
+    async pruneEdges() {
+        const maxAgeMs = this.config.edgeMaxAgeDays * 24 * 60 * 60 * 1000;
+        const cutoffTime = Date.now() / 1000 - maxAgeMs / 1000;
+        const result = this.db.prepare(`
+      DELETE FROM causal_edges
+      WHERE confidence < ?
+        OR created_at < ?
+    `).run(this.config.confidenceThreshold, cutoffTime);
+        return result.changes;
+    }
+    /**
+     * Calculate overall statistics
+     */
+    calculateStats() {
+        const stats = this.db.prepare(`
+      SELECT
+        AVG(ABS(uplift)) as avg_uplift,
+        AVG(confidence) as avg_confidence
+      FROM causal_edges
+      WHERE uplift IS NOT NULL
+    `).get();
+        return {
+            avgUplift: stats?.avg_uplift || 0,
+            avgConfidence: stats?.avg_confidence || 0
+        };
+    }
+    /**
+     * Generate recommendations based on learning results
+     */
+    generateRecommendations(report) {
+        const recommendations = [];
+        if (report.edgesDiscovered === 0) {
+            recommendations.push('No new causal edges discovered. Consider collecting more diverse episode data.');
+        }
+        if (report.avgUplift < 0.1) {
+            recommendations.push('Average uplift is low. Review task sequences for optimization opportunities.');
+        }
+        if (report.avgConfidence < 0.7) {
+            recommendations.push('Average confidence is below target. Increase sample sizes or refine hypothesis selection.');
+        }
+        if (report.experimentsCompleted > 0) {
+            recommendations.push(`${report.experimentsCompleted} experiments completed. Review results for actionable insights.`);
+        }
+        if (report.edgesPruned > report.edgesDiscovered) {
+            recommendations.push('More edges pruned than discovered. Consider adjusting confidence thresholds.');
+        }
+        return recommendations;
+    }
+    /**
+     * Print report to console
+     */
+    printReport(report) {
+        console.log('═══════════════════════════════════════════════════════════');
+        console.log('  Nightly Learner Report');
+        console.log('═══════════════════════════════════════════════════════════\n');
+        console.log(`  Execution Time: ${report.executionTimeMs}ms`);
+        console.log(`  Timestamp: ${new Date(report.timestamp).toISOString()}\n`);
+        console.log('  Results:');
+        console.log(`    • Edges Discovered: ${report.edgesDiscovered}`);
+        console.log(`    • Edges Pruned: ${report.edgesPruned}`);
+        console.log(`    • Experiments Completed: ${report.experimentsCompleted}`);
+        console.log(`    • Experiments Created: ${report.experimentsCreated}\n`);
+        console.log('  Statistics:');
+        console.log(`    • Avg Uplift: ${report.avgUplift.toFixed(3)}`);
+        console.log(`    • Avg Confidence: ${report.avgConfidence.toFixed(3)}\n`);
+        if (report.recommendations.length > 0) {
+            console.log('  Recommendations:');
+            report.recommendations.forEach(rec => console.log(`    • ${rec}`));
+            console.log('');
+        }
+        console.log('═══════════════════════════════════════════════════════════\n');
+    }
+    /**
+     * Update learner configuration
+     */
+    updateConfig(config) {
+        this.config = { ...this.config, ...config };
+    }
+}
+//# sourceMappingURL=NightlyLearner.js.map

package/dist/controllers/NightlyLearner.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"NightlyLearner.js","sourceRoot":"","sources":["../../src/controllers/NightlyLearner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAGH,OAAO,EAAE,iBAAiB,EAAc,MAAM,wBAAwB,CAAC;AACvE,OAAO,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAC;AACvD,OAAO,EAAE,YAAY,EAAE,MAAM,mBAAmB,CAAC;AA0BjD,MAAM,OAAO,cAAc;IASf;IARF,EAAE,CAAW;IACb,WAAW,CAAoB;IAC/B,SAAS,CAAkB;IAC3B,YAAY,CAAe;IAEnC,YACE,EAAY,EACZ,QAA0B,EAClB,SAAwB;QAC9B,aAAa,EAAE,GAAG;QAClB,aAAa,EAAE,EAAE;QACjB,mBAAmB,EAAE,GAAG;QACxB,eAAe,EAAE,IAAI;QACrB,aAAa,EAAE,IAAI;QACnB,cAAc,EAAE,EAAE;QAClB,eAAe,EAAE,IAAI;QACrB,gBAAgB,EAAE,EAAE;KACrB;QATO,WAAM,GAAN,MAAM,CASb;QAED,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC;QACb,IAAI,CAAC,WAAW,GAAG,IAAI,iBAAiB,CAAC,EAAE,CAAC,CAAC;QAC7C,IAAI,CAAC,SAAS,GAAG,IAAI,eAAe,CAAC,EAAE,EAAE,QAAQ,CAAC,CAAC;QACnD,IAAI,CAAC,YAAY,GAAG,IAAI,YAAY,CAAC,EAAE,EAAE,QAAQ,CAAC,CAAC;IACrD,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,GAAG;QACP,OAAO,CAAC,GAAG,CAAC,oCAAoC,CAAC,CAAC;QAClD,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAE7B,MAAM,MAAM,GAAkB;YAC5B,SAAS,EAAE,SAAS;YACpB,eAAe,EAAE,CAAC;YAClB,eAAe,EAAE,CAAC;YAClB,WAAW,EAAE,CAAC;YACd,oBAAoB,EAAE,CAAC;YACvB,kBAAkB,EAAE,CAAC;YACrB,SAAS,EAAE,CAAC;YACZ,aAAa,EAAE,CAAC;YAChB,eAAe,EAAE,EAAE;SACpB,CAAC;QAEF,IAAI,CAAC;YACH,oCAAoC;YACpC,OAAO,CAAC,GAAG,CAAC,sDAAsD,CAAC,CAAC;YACpE,MAAM,CAAC,eAAe,GAAG,MAAM,IAAI,CAAC,mBAAmB,EAAE,CAAC;YAC1D,OAAO,CAAC,GAAG,CAAC,mBAAmB,MAAM,CAAC,eAAe,cAAc,CAAC,CAAC;YAErE,uCAAuC;YACvC,OAAO,CAAC,GAAG,CAAC,kCAAkC,CAAC,CAAC;YAChD,MAAM,CAAC,oBAAoB,GAAG,MAAM,IAAI,CAAC,mBAAmB,EAAE,CAAC;YAC/D,OAAO,CAAC,GAAG,CAAC,kBAAkB,MAAM,CAAC,oBAAoB,gBAAgB,CAAC,CAAC;YAE3E,8CAA8C;YAC9C,IAAI,IAAI,CAAC,MAAM,CAAC,eAAe,EAAE,CAAC;gBAChC,OAAO,CAAC,GAAG,CAAC,oCAAoC,CAAC,CAAC;gBAClD,MAAM,CAAC,kBAAkB,GAAG,MAAM,IAAI,CAAC,iBAAiB,EAAE,CAAC;gBAC3D,OAAO,CAAC,GAAG,CAAC,gBAAgB,MAAM,CAAC,kBAAkB,oBAAoB,CAAC,CAAC;YAC7E,CAAC;YAED,qCAAqC;YACrC,IAAI,IAAI,CAAC,MAAM,CAAC,aAAa,EAAE,CAAC;gBAC9B,OAAO,CAAC,GAAG,CAAC,oCAAoC,CAAC,CAAC;gBAClD,MAAM,CAAC,WAAW,GAAG,MAAM,IAAI,CAAC,UAAU,EAAE,CAAC;gBAC7C,OAAO,CAAC,GAAG,CAAC,eAAe,MAAM,CAAC,WAAW,UAAU,CAAC,CAAC;YAC3D,CAAC;YAED,+BAA+B;YAC/B,MAAM,KAAK,GAAG,IAAI,CAAC,cAAc,EAAE,CAAC;YACpC,MAAM,CAAC,SAAS,GAAG,KAAK,CAAC,SAAS,CAAC;YACnC,MAAM,CAAC,aAAa,GAAG,KAAK,CAAC,aAAa,CAAC;YAE3C,mCAAmC;YACnC,MAAM,CAAC,eAAe,GAAG,IAAI,CAAC,uBAAuB,CAAC,MAAM,CAAC,CAAC;YAE9D,MAAM,CAAC,eAAe,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;YAEhD,OAAO,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;YAC7C,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC;YAEzB,OAAO,MAAM,CAAC;QAChB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,CAAC,KAAK,CAAC,2BAA2B,EAAE,KAAK,CAAC,CAAC;YAClD,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;IAED;;;;;;;;;;;OAWG;IACH,KAAK,CAAC,QAAQ,CAAC,MAKd;QACC,OAAO,IAAI,CAAC,mBAAmB,EAAE,CAAC;IACpC,CAAC;IAEO,KAAK,CAAC,mBAAmB;QAC/B,IAAI,UAAU,GAAG,CAAC,CAAC;QAEnB,gEAAgE;QAChE,MAAM,cAAc,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;;;;;;;;;;KAgBtC,CAAC,CAAC,GAAG,EAAW,CAAC;QAElB,KAAK,MAAM,IAAI,IAAI,cAAc,EAAE,CAAC;YAClC,+BAA+B;YAC/B,MAAM,QAAQ,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;OAGhC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC;YAEjC,IAAI,QAAQ;gBAAE,SAAS;YAEvB,6DAA6D;YAC7D,oDAAoD;YACpD,MAAM,UAAU,GAAG,IAAI,CAAC,mBAAmB,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YAE1D,2CAA2C;YAC3C,MAAM,GAAG,GAAG,IAAI,CAAC,qBAAqB,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,CAAE,iBAAiB;YAChF,MAAM,GAAG,GAAG,IAAI,CAAC,qBAAqB,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC,CAAC,oBAAoB;YAEnF,oCAAoC;YACpC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,gCAAgC;YAC7C,MAAM,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC;YACzB,MAAM,oBAAoB,GAAG,CAAC,GAAG,GAAG,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC,GAAG,UAAU,CAAC,CAAC;YAExE,yDAAyD;YACzD,MAAM,UAAU,GAAG,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YACtD,MAAM,UAAU,GAAG,IAAI,CAAC,mBAAmB,CAAC,UAAU,EAAE,oBAAoB,CAAC,CAAC;YAE9E,+BAA+B;YAC/B,IAAI,IAAI,CAAC,GAAG,CAAC,oBAAoB,CAAC,IAAI,IAAI,CAAC,MAAM,CAAC,eAAe,IAAI,UAAU,IAAI,IAAI,CAAC,MAAM,CAAC,mBAAmB,EAAE,CAAC;gBACnH,MAAM,IAAI,GAAe;oBACvB,YAAY,EAAE,IAAI,CAAC,OAAO;oBAC1B,cAAc,EAAE,SAAS;oBACzB,UAAU,EAAE,IAAI,CAAC,KAAK;oBACtB,YAAY,EAAE,SAAS;oBACvB,UAAU,EAAE,GAAG,EAAE,4DAA4D;oBAC7E,MAAM,EAAE,oBAAoB;oBAC5B,UAAU;oBACV,UAAU;oBACV,SAAS,EAAE,GAAG,IAAI,CAAC,SAAS,MAAM,IAAI,CAAC,OAAO,kBAAkB;oBAChE,QAAQ,EAAE;wBACR,UAAU;wBACV,GAAG;wBACH,GAAG;wBACH,YAAY,EAAE,IAAI,CAAC,GAAG,EAAE;qBACzB;iBACF,CAAC;gBAEF,IAAI,CAAC,WAAW,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC;gBACrC,UAAU,EAAE,CAAC;YACf,CAAC;QACH,CAAC;QAED,OAAO,UAAU,CAAC;IACpB,CAAC;IAED;;OAEG;IACK,mBAAmB,CAAC,SAAiB;QAC3C,MAAM,OAAO,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,oDAAoD,CAAC,CAAC,GAAG,CAAC,SAAS,CAAQ,CAAC;QAE5G,iDAAiD;QACjD,MAAM,MAAM,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;KAM9B,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,UAAU,CAAQ,CAAC;QAEhD,MAAM,UAAU,GAAG,MAAM,CAAC,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;QAEjE,iCAAiC;QACjC,OAAO,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC,CAAC;IACpD,CAAC;IAED;;OAEG;IACK,qBAAqB,CAAC,IAAY,EAAE,OAAgB;QAC1D,0EAA0E;QAC1E,MAAM,SAAS,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;cAGxB,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK;;;;;;KAM7B,CAAC,CAAC,GAAG,CAAC,IAAI,CAAQ,CAAC;QAEpB,OAAO,SAAS,EAAE,UAAU,IAAI,GAAG,CAAC;IACtC,CAAC;IAED;;OAEG;IACK,aAAa,CAAC,IAAY;QAChC,MAAM,KAAK,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;KAI7B,CAAC,CAAC,GAAG,CAAC,IAAI,CAAQ,CAAC;QAEpB,OAAO,KAAK,CAAC,KAAK,CAAC;IACrB,CAAC;IAED;;OAEG;IACK,mBAAmB,CAAC,UAAkB,EAAE,MAAc;QAC5D,oCAAoC;QACpC,kEAAkE;QAElE,MAAM,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,UAAU,GAAG,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC,qBAAqB;QAC3E,MAAM,gBAAgB,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC,oBAAoB;QAEpF,OAAO,YAAY,GAAG,gBAAgB,CAAC;IACzC,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,mBAAmB;QAC/B,MAAM,kBAAkB,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;KAK1C,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,aAAa,CAAU,CAAC;QAE3C,IAAI,SAAS,GAAG,CAAC,CAAC;QAElB,KAAK,MAAM,GAAG,IAAI,kBAAkB,EAAE,CAAC;YACrC,IAAI,CAAC;gBACH,IAAI,CAAC,WAAW,CAAC,eAAe,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;gBACzC,SAAS,EAAE,CAAC;YACd,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,OAAO,CAAC,KAAK,CAAC,kDAAkD,GAAG,CAAC,EAAE,GAAG,EAAE,KAAK,CAAC,CAAC;YACpF,CAAC;QACH,CAAC;QAED,OAAO,SAAS,CAAC;IACnB,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,iBAAiB;QAC7B,MAAM,kBAAkB,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;KAI1C,CAAC,CAAC,GAAG,EAAS,CAAC;QAEhB,MAAM,SAAS,GAAG,IAAI,CAAC,MAAM,CAAC,gBAAgB,GAAG,kBAAkB,CAAC,KAAK,CAAC;QAC1E,IAAI,SAAS,IAAI,CAAC,EAAE,CAAC;YACnB,OAAO,CAAC,CAAC;QACX,CAAC;QAED,4DAA4D;QAC5D,MAAM,UAAU,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;;;;;;;;;;;KAgBlC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,aAAa,EAAE,SAAS,CAAU,CAAC;QAEtD,IAAI,OAAO,GAAG,CAAC,CAAC;QAEhB,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;YACnC,MAAM,KAAK,GAAG,IAAI,CAAC,WAAW,CAAC,gBAAgB,CAAC;gBAC9C,IAAI,EAAE,SAAS,SAAS,CAAC,cAAc,SAAS;gBAChD,UAAU,EAAE,GAAG,SAAS,CAAC,cAAc,8BAA8B;gBACrE,WAAW,EAAE,SAAS,CAAC,YAAY;gBACnC,aAAa,EAAE,SAAS;gBACxB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;gBACrB,UAAU,EAAE,CAAC;gBACb,MAAM,EAAE,SAAS;gBACjB,QAAQ,EAAE;oBACR,aAAa,EAAE,IAAI;oBACnB,iBAAiB,EAAE,SAAS,CAAC,kBAAkB;iBAChD;aACF,CAAC,CAAC;YAEH,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;OAEG;IACK,KAAK,CAAC,UAAU;QACtB,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,CAAC,cAAc,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;QAClE,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,GAAG,QAAQ,GAAG,IAAI,CAAC;QAEvD,MAAM,MAAM,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;KAI9B,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,mBAAmB,EAAE,UAAU,CAAC,CAAC;QAEpD,OAAO,MAAM,CAAC,OAAO,CAAC;IACxB,CAAC;IAED;;OAEG;IACK,cAAc;QACpB,MAAM,KAAK,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC;;;;;;KAM7B,CAAC,CAAC,GAAG,EAAS,CAAC;QAEhB,OAAO;YACL,SAAS,EAAE,KAAK,EAAE,UAAU,IAAI,CAAC;YACjC,aAAa,EAAE,KAAK,EAAE,cAAc,IAAI,CAAC;SAC1C,CAAC;IACJ,CAAC;IAED;;OAEG;IACK,uBAAuB,CAAC,MAAqB;QACnD,MAAM,eAAe,GAAa,EAAE,CAAC;QAErC,IAAI,MAAM,CAAC,eAAe,KAAK,CAAC,EAAE,CAAC;YACjC,eAAe,CAAC,IAAI,CAAC,gFAAgF,CAAC,CAAC;QACzG,CAAC;QAED,IAAI,MAAM,CAAC,SAAS,GAAG,GAAG,EAAE,CAAC;YAC3B,eAAe,CAAC,IAAI,CAAC,8EAA8E,CAAC,CAAC;QACvG,CAAC;QAED,IAAI,MAAM,CAAC,aAAa,GAAG,GAAG,EAAE,CAAC;YAC/B,eAAe,CAAC,IAAI,CAAC,2FAA2F,CAAC,CAAC;QACpH,CAAC;QAED,IAAI,MAAM,CAAC,oBAAoB,GAAG,CAAC,EAAE,CAAC;YACpC,eAAe,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,oBAAoB,iEAAiE,CAAC,CAAC;QACxH,CAAC;QAED,IAAI,MAAM,CAAC,WAAW,GAAG,MAAM,CAAC,eAAe,EAAE,CAAC;YAChD,eAAe,CAAC,IAAI,CAAC,8EAA8E,CAAC,CAAC;QACvG,CAAC;QAED,OAAO,eAAe,CAAC;IACzB,CAAC;IAED;;OAEG;IACK,WAAW,CAAC,MAAqB;QACvC,OAAO,CAAC,GAAG,CAAC,6DAA6D,CAAC,CAAC;QAC3E,OAAO,CAAC,GAAG,CAAC,0BAA0B,CAAC,CAAC;QACxC,OAAO,CAAC,GAAG,CAAC,+DAA+D,CAAC,CAAC;QAC7E,OAAO,CAAC,GAAG,CAAC,qBAAqB,MAAM,CAAC,eAAe,IAAI,CAAC,CAAC;QAC7D,OAAO,CAAC,GAAG,CAAC,gBAAgB,IAAI,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC;QAC1E,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;QAC1B,OAAO,CAAC,GAAG,CAAC,2BAA2B,MAAM,CAAC,eAAe,EAAE,CAAC,CAAC;QACjE,OAAO,CAAC,GAAG,CAAC,uBAAuB,MAAM,CAAC,WAAW,EAAE,CAAC,CAAC;QACzD,OAAO,CAAC,GAAG,CAAC,gCAAgC,MAAM,CAAC,oBAAoB,EAAE,CAAC,CAAC;QAC3E,OAAO,CAAC,GAAG,CAAC,8BAA8B,MAAM,CAAC,kBAAkB,IAAI,CAAC,CAAC;QACzE,OAAO,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC;QAC7B,OAAO,CAAC,GAAG,CAAC,qBAAqB,MAAM,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QAChE,OAAO,CAAC,GAAG,CAAC,yBAAyB,MAAM,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;QAE1E,IAAI,MAAM,CAAC,eAAe,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtC,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,CAAC;YAClC,MAAM,CAAC,eAAe,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,SAAS,GAAG,EAAE,CAAC,CAAC,CAAC;YACnE,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,+DAA+D,CAAC,CAAC;IAC/E,CAAC;IAED;;OAEG;IACH,YAAY,CAAC,MAA8B;QACzC,IAAI,CAAC,MAAM,GAAG,EAAE,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,MAAM,EAAE,CAAC;IAC9C,CAAC;CACF"}

package/dist/controllers/ReflexionMemory.d.ts

@@ -0,0 +1,84 @@
+/**
+ * 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
+ */
+import { Database } from 'better-sqlite3';
+import { EmbeddingService } from './EmbeddingService.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;
+    constructor(db: Database, embedder: EmbeddingService);
+    /**
+     * 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>;
+    /**
+     * 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;
+    private cosineSimilarity;
+}
+//# sourceMappingURL=ReflexionMemory.d.ts.map

package/dist/controllers/ReflexionMemory.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ReflexionMemory.d.ts","sourceRoot":"","sources":["../../src/controllers/ReflexionMemory.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,gBAAgB,CAAC;AAC1C,OAAO,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AAEzD,MAAM,WAAW,OAAO;IACtB,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,SAAS,EAAE,MAAM,CAAC;IAClB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,OAAO,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;CAChC;AAED,MAAM,WAAW,oBAAqB,SAAQ,OAAO;IACnD,SAAS,CAAC,EAAE,YAAY,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,CAAC,CAAC,EAAE,MAAM,CAAC;IACX,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB;AAED,qBAAa,eAAe;IAC1B,OAAO,CAAC,EAAE,CAAW;IACrB,OAAO,CAAC,QAAQ,CAAmB;gBAEvB,EAAE,EAAE,QAAQ,EAAE,QAAQ,EAAE,gBAAgB;IAKpD;;OAEG;IACG,YAAY,CAAC,OAAO,EAAE,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC;IAoCrD;;OAEG;IACG,gBAAgB,CAAC,KAAK,EAAE,cAAc,GAAG,OAAO,CAAC,oBAAoB,EAAE,CAAC;IAiF9E;;OAEG;IACH,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,cAAc,CAAC,EAAE,MAAM,GAAG;QACnD,aAAa,EAAE,MAAM,CAAC;QACtB,WAAW,EAAE,MAAM,CAAC;QACpB,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;QACnB,gBAAgB,EAAE,MAAM,CAAC;KAC1B;IA4CD;;OAEG;IACG,kBAAkB,CAAC,KAAK,EAAE,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC;IAmBhE;;OAEG;IACG,oBAAoB,CAAC,KAAK,EAAE,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC;IAsBlE;;OAEG;IACH,aAAa,CAAC,MAAM,EAAE;QACpB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB,cAAc,CAAC,EAAE,MAAM,CAAC;KACzB,GAAG,MAAM;IA4BV,OAAO,CAAC,gBAAgB;IAOxB,OAAO,CAAC,cAAc;IAStB,OAAO,CAAC,kBAAkB;IAI1B,OAAO,CAAC,oBAAoB;IAI5B,OAAO,CAAC,gBAAgB;CAazB"}