Marc Rejohn Castillano 5cb6561924 added ruflo

2026-04-09 19:01:53 +08:00

9.5 KiB

Raw Blame History

agentic-flow v1.7.1 Release Notes

Release Date: October 24, 2025 Status: Full Implementation with Advanced Features

🎯 Overview

Version 1.7.1 delivers the complete implementation of advanced ReasoningBank features that were simplified in v1.7.0. This release includes full CausalRecall integration, NightlyLearner auto-consolidation, and comprehensive causal reasoning capabilities.

🚀 New Features

1. Full CausalRecall Integration

HybridReasoningBank now uses utility-based reranking for intelligent pattern retrieval:

// Utility formula: U = α·similarity + β·uplift − γ·latency
// Default weights: α=0.6, β=0.3, γ=0.1

import { HybridReasoningBank } from 'agentic-flow/reasoningbank';

const rb = new HybridReasoningBank({ preferWasm: true });

// Store patterns with causal tracking
await rb.storePattern({
  sessionId: 'session-1',
  task: 'API optimization',
  input: 'Slow endpoint',
  output: 'Cached with Redis',
  critique: 'Significant performance improvement',
  success: true,
  reward: 0.95,
  latencyMs: 120,
  tokensUsed: 1500
});

// Retrieve with causal ranking
const patterns = await rb.retrievePatterns('optimize API', {
  k: 5,
  minReward: 0.8,
  onlySuccesses: true
});

// Each pattern includes:
// - similarity: Vector similarity score
// - uplift: Causal improvement measure
// - utilityScore: Combined ranking metric

2. Causal Memory Graph

Automatic tracking of causal relationships between actions and outcomes:

// Causal edges are automatically recorded for successful patterns
// Tracks: p(y|do(x)) intervention-based causality

await rb.storePattern({
  sessionId: 'session-2',
  task: 'Add caching',
  success: true,
  reward: 0.92
  // ➜ Creates causal edge linking action → outcome
});

3. Strategy Learning with Task Statistics

Learn optimal strategies from historical patterns with statistical confidence:

const strategy = await rb.learnStrategy('Error handling');

console.log(strategy);
// {
//   patterns: [...],  // Successful historical patterns
//   causality: {
//     action: 'Error handling',
//     avgReward: 0.88,
//     avgUplift: 0.15,      // +15% improvement trend
//     confidence: 0.82,      // High statistical confidence
//     evidenceCount: 12,     // 12 historical attempts
//     recommendation: 'DO_IT'
//   },
//   confidence: 0.85,
//   recommendation: 'Strong evidence for success (12 patterns, +15.0% uplift)'
// }

4. Auto-Consolidation with NightlyLearner

Automatic discovery of causal patterns and skill consolidation:

import { AdvancedMemorySystem } from 'agentic-flow/reasoningbank';

const memory = new AdvancedMemorySystem();

// Run automated learning cycle
const result = await memory.autoConsolidate({
  minUses: 3,
  minSuccessRate: 0.7,
  lookbackDays: 30
});

console.log(result);
// {
//   skillsCreated: 5,
//   causalEdgesCreated: 12,
//   patternsAnalyzed: 45,
//   executionTimeMs: 1250,
//   recommendations: [
//     'High-performing pattern detected: API caching (92% success rate)',
//     'Causal relationship confirmed: caching → latency reduction',
//     '5 new skills consolidated from frequent patterns'
//   ]
// }

5. Episodic Replay - Learning from Failures

Analyze past failures to generate actionable recommendations:

const failures = await memory.replayFailures('database migration', 5);

failures.forEach(failure => {
  console.log('Critique:', failure.critique);
  console.log('What went wrong:', failure.whatWentWrong);
  // ['Low success rate observed', 'High latency detected']

  console.log('How to fix:', failure.howToFix);
  // ['Review similar successful patterns', 'Optimize for lower latency']

  console.log('Similar failures:', failure.similarFailures);
  // 3
});

6. What-If Causal Analysis

Predict outcomes before taking actions:

const analysis = await memory.whatIfAnalysis('Enable caching');

console.log(analysis);
// {
//   action: 'Enable caching',
//   avgReward: 0.93,
//   avgUplift: 0.22,             // +22% expected improvement
//   confidence: 0.88,
//   evidenceCount: 8,
//   recommendation: 'DO_IT',
//   expectedImpact: 'Highly beneficial: Expected +22.0% improvement'
// }

7. Skill Composition

Automatically compose multiple learned skills for complex tasks:

const composition = await memory.composeSkills('Build scalable API', 5);

console.log(composition);
// {
//   availableSkills: [
//     { name: 'api_caching', successRate: 0.95, uses: 12 },
//     { name: 'rate_limiting', successRate: 0.88, uses: 8 },
//     { name: 'auth_flow', successRate: 0.92, uses: 10 }
//   ],
//   compositionPlan: 'api_caching → rate_limiting → auth_flow',
//   expectedSuccessRate: 0.91
// }

📊 Performance Improvements

116x faster vector search with WASM acceleration (vs TypeScript fallback)
56% memory reduction with SharedMemoryPool singleton pattern
Intelligent caching with 60-second TTL for frequent queries
Lazy WASM loading with graceful fallback to TypeScript

🔧 Technical Implementation

API Alignment

All implementations now use agentdb v1.3.9 API correctly:

✅ ReflexionMemory.getTaskStats() for strategy learning
✅ CausalRecall.recall() with utility-based reranking
✅ CausalMemoryGraph.addCausalEdge() for causal tracking
✅ NightlyLearner.run() for automated discovery
✅ Direct imports from agentdb/controllers/* (with patch)

Breaking Changes from v1.7.0

None - v1.7.1 is fully backwards compatible with v1.7.0. All simplified implementations have been replaced with full versions while maintaining the same API surface.

Module Structure

src/reasoningbank/
├── HybridBackend.ts          # Full CausalRecall integration
├── AdvancedMemory.ts         # NightlyLearner + high-level ops
├── backend-selector.ts        # Automatic backend selection
├── agentdb-adapter.ts         # AgentDB integration layer
└── index.ts                   # Public exports

🐛 Known Issues & Workarounds

AgentDB Import Resolution

Issue: agentdb v1.3.9 has missing .js extensions in controllers/index.js

Workaround: Apply patch automatically during npm install:

# Post-install patch applied via patches/agentdb-fix-imports.patch
# Adds .js extensions to:
# - export { ReflexionMemory } from './ReflexionMemory.js';
# - export { SkillLibrary } from './SkillLibrary.js';
# - export { EmbeddingService } from './EmbeddingService.js';

Status: Reported to agentdb maintainers. Patch is non-invasive and safe.

📚 Migration Guide

From v1.7.0 to v1.7.1

No code changes required! v1.7.1 replaces simplified implementations with full versions while maintaining the exact same API.

Example: Strategy Learning

// v1.7.0 (simplified)
const strategy = await rb.learnStrategy('task');
// { patterns: [...], causality: {...}, confidence: 0.6, recommendation: '...' }

// v1.7.1 (full implementation)
const strategy = await rb.learnStrategy('task');
// Same API! But now includes:
// - Real causal analysis from ReflexionMemory.getTaskStats()
// - Improvement trends and confidence scores
// - Evidence-based recommendations

Example: Auto-Consolidation

// v1.7.0 (basic consolidation)
const result = await rb.autoConsolidate(3, 0.7, 30);
// { skillsCreated: X }

// v1.7.1 (with NightlyLearner)
const result = await memory.autoConsolidate({ minUses: 3, minSuccessRate: 0.7 });
// {
//   skillsCreated: X,
//   causalEdgesCreated: Y,        // NEW: Causal discovery
//   patternsAnalyzed: Z,           // NEW: Pattern analysis
//   executionTimeMs: T,
//   recommendations: [...]         // NEW: Actionable insights
// }

🧪 Testing

Comprehensive test suite added in tests/reasoningbank/:

integration.test.ts - 20 integration tests covering all features
hybrid-backend.test.ts - Unit tests for HybridReasoningBank (vitest-ready)
advanced-memory.test.ts - Unit tests for AdvancedMemorySystem (vitest-ready)

# Run tests
npm test

# Run specific test suite
npx vitest run tests/reasoningbank/integration.test.ts

📖 Documentation

HybridBackend API - Full source with JSDoc
AdvancedMemory API - Full source with JSDoc
TESTING.md - Test results and validation
CHANGELOG.md - Detailed version history

🔮 Future Enhancements (v1.8.0)

WASM SIMD Optimization: Full Rust implementation with SIMD acceleration
Distributed Causal Discovery: Multi-node causal inference
Explainable Recall: Provenance certificates with Merkle proofs
Streaming Patterns: Real-time pattern updates and notifications
Cross-Session Learning: Persistent learning across multiple sessions

🙏 Credits

AgentDB v1.3.9: Frontier memory systems integration
ReasoningBank: Self-learning AI with experience replay
agentic-flow community: Testing and feedback

📦 Installation

npm install agentic-flow@1.7.1
# or
npm install agentic-flow@latest

# Post-install: agentdb patch applied automatically

🔗 Resources

GitHub: https://github.com/ruvnet/agentic-flow
npm: https://www.npmjs.com/package/agentic-flow
AgentDB: https://www.npmjs.com/package/agentdb
Issues: https://github.com/ruvnet/agentic-flow/issues

Status: ✅ Production Ready Compatibility: Node.js 18+, TypeScript 5+ License: MIT

9.5 KiB Raw Blame History Unescape Escape