Pre-Trained ReasoningBank Models

Welcome to the ReasoningBank model zoo! This directory contains 5 production-ready, pre-trained models with thousands of optimized patterns for immediate use.

🚀 Quick Start

bash

# Choose a model
cd safla  # or google-research, code-reasoning, problem-solving, domain-expert

# Install it
cp memory.db ~/.swarm/memory.db

# Try it!
npx claude-flow@alpha memory query "your question here" --reasoningbank

That's it! You now have expert-level patterns ready to use.

📦 Available Models

1. SAFLA (Self-Aware Feedback Loop Algorithm)

Best for: Self-learning systems that improve from experience

Patterns: 2,000
Size: 10.35 MB
Confidence: 83.8% average
Success Rate: 90.3% average
Specialties:
- Self-learning patterns
- Feedback loop optimization
- Bayesian confidence adjustment
- Success/failure distillation
- Recursive improvement cycles

Use when: Building agents that learn automatically, no retraining needed

bash

cp safla/memory.db ~/.swarm/memory.db

2. Google Research (Strategy-Level Memory)

Best for: Following latest AI research best practices

Patterns: 3,000
Size: 8.92 MB
Confidence: 88% average
Paper: arXiv:2509.25140
Specialties:
- Strategy-level memory (40% from failures!)
- MaTTS parallel & sequential scaling
- Closed-loop learning
- Success AND failure patterns
- Research-backed approaches

Use when: Implementing cutting-edge AI research

bash

cp google-research/memory.db ~/.swarm/memory.db

3. Code Reasoning (Programming Best Practices)

Best for: Software development and code generation

Patterns: 2,500
Size: 2.66 MB
Confidence: 91.5% average
Success Rate: 91.2% average
Specialties:
- Design patterns & architecture (SOLID, MVC, microservices)
- Algorithm optimization (O(n²) → O(n))
- Code quality & refactoring (DRY, KISS, clean code)
- Language-specific patterns (JS, Python, Go, Rust, Java)
- Debugging & error handling

Use when: Code generation, review, or refactoring

bash

cp code-reasoning/.swarm/memory.db ~/.swarm/memory.db

4. Problem Solving (Cognitive Diversity)

Best for: General reasoning and problem analysis

Patterns: 2,000
Size: 5.85 MB
Confidence: 83.7% average
Success Rate: 84.6% average
Specialties:
- Convergent thinking (logical, systematic)
- Divergent thinking (creative, exploratory)
- Lateral thinking (pattern-breaking, unconventional)
- Systems thinking (holistic, emergent behavior)
- Critical thinking (bias detection, validation)

Use when: Complex problems requiring multiple reasoning approaches

bash

cp problem-solving/memory.db ~/.swarm/memory.db

5. Domain Expert (Multi-Domain Expertise)

Best for: Specialized technical domains

Patterns: 1,500
Size: 2.39 MB
Confidence: 89.4% average
Success Rate: 88.5% average
Domains:
- DevOps & Infrastructure (CI/CD, Kubernetes, monitoring)
- Data Engineering & ML (ETL, MLOps, feature engineering)
- Security & Compliance (GDPR, SOC2, encryption)
- API Design & Integration (REST, GraphQL, webhooks)
- Performance & Scalability (caching, load balancing, CDN)

Use when: Domain-specific expertise needed

bash

cp domain-expert/memory.db ~/.swarm/memory.db

📊 Model Comparison

Model	Patterns	Size	Avg Confidence	Use Case
SAFLA	2,000	10.35 MB	83.8%	Self-learning systems
Google Research	3,000	8.92 MB	88.0%	Research best practices
Code Reasoning	2,500	2.66 MB	91.5%	Software development
Problem Solving	2,000	5.85 MB	83.7%	General reasoning
Domain Expert	1,500	2.39 MB	89.4%	Technical expertise

🎯 How to Choose

I want to...

✅ Build AI that learns from experience → SAFLA
✅ Follow latest research best practices → Google Research
✅ Generate/review code → Code Reasoning
✅ Solve complex problems → Problem Solving
✅ Get domain expertise → Domain Expert

My project is...

🤖 AI agent development → SAFLA or Google Research
💻 Software development → Code Reasoning
🧩 Problem-solving system → Problem Solving
🏗️ Infrastructure/DevOps → Domain Expert

📖 Documentation

Each model directory contains:

README.md - Model overview and usage guide
memory.db - Pre-trained database (ready to use!)
train-*.js - Training script (see how it was made)
validation-report.md - Quality validation results
TRAINING_SUMMARY.md - Detailed training information

General Guides:

HOW-TO-USE.md - Installation and usage guide
HOW-TO-TRAIN.md - Train your own models
INDEX.md - Complete navigation index

Technical Documentation:

_docs/ - Technical references and completion reports
_scripts/ - Utility scripts for validation and training

🔧 Advanced Usage

Merge Multiple Models

bash

# Combine patterns from multiple models
cp safla/memory.db ~/.swarm/memory.db

# Merge Google Research patterns
sqlite3 ~/.swarm/memory.db << SQL
ATTACH DATABASE 'google-research/memory.db' AS source;
INSERT OR IGNORE INTO patterns SELECT * FROM source.patterns;
INSERT OR IGNORE INTO pattern_embeddings SELECT * FROM source.pattern_embeddings;
DETACH DATABASE source;
SQL

Project-Specific Models

bash

# Use different models per project
mkdir ./my-project/.swarm
cp code-reasoning/.swarm/memory.db ./my-project/.swarm/

# Set environment variable
export CLAUDE_FLOW_DB_PATH=./my-project/.swarm/memory.db

# Or use --db-path flag
npx claude-flow@alpha memory query "test" --db-path ./my-project/.swarm/memory.db

Query Examples

bash

# Find patterns by domain
npx claude-flow@alpha memory query "API authentication" --namespace security

# High confidence only
npx claude-flow@alpha memory query "database optimization" --min-confidence 0.8

# Specific domain
sqlite3 ~/.swarm/memory.db "SELECT * FROM patterns WHERE domain = 'api-development' LIMIT 5"

🎓 Training Your Own Models

Want to create custom models? See HOW-TO-TRAIN.md for:

Pattern generation strategies
Embedding creation
Relationship mapping
Benchmarking & validation
Parallel training with agents

Training Scripts Provided:

safla/train-safla.js - 2,000 self-learning patterns
google-research/train-google.js - 3,000 strategy patterns
code-reasoning/train-code.js - 2,500 programming patterns
problem-solving/train-problem.js - 2,000 reasoning patterns
domain-expert/train-domain.js - 1,500 domain patterns

✅ Quality Assurance

All models have been:

✅ Validated for schema compliance
✅ Benchmarked for performance (<5ms queries)
✅ Tested for data quality (>70% confidence)
✅ Optimized for storage efficiency (<10 KB/pattern)
✅ Verified for production readiness

Run validation yourself:

bash

cd safla  # or any model directory
node ../_scripts/validation-suite.cjs . safla

🚀 Integration

With agentic-flow

javascript

import { AgenticFlow } from 'agentic-flow';

const agent = new AgenticFlow('coder', {
  reasoningBank: {
    enabled: true,
    dbPath: process.env.HOME + '/.swarm/memory.db',
    minConfidence: 0.7
  }
});

// Agent automatically uses ReasoningBank patterns
await agent.execute({ task: 'Implement JWT auth' });

With Claude Code

bash

# Load patterns as context
npx claude-flow@alpha memory query "authentication patterns" > context.json

# Use in Claude Code
claude code --context context.json "Implement auth"

Direct SQL

javascript

const Database = require('better-sqlite3');
const db = new Database(process.env.HOME + '/.swarm/memory.db');

const patterns = db.prepare(`
  SELECT * FROM patterns
  WHERE domain = ? AND confidence > 0.8
  ORDER BY success_rate DESC
  LIMIT 10
`).all('api-development');

🏆 Performance Benchmarks

All models meet or exceed these criteria:

Metric	Target	All Models
Query Latency	<5ms	✅ 0.05-2ms
Storage	<10 KB/pattern	✅ 2-6 KB/pattern
Confidence	>70%	✅ 83-91%
Embedding Coverage	100%	✅ 100%

📝 License

All models are MIT licensed and free to use in commercial and non-commercial projects.

🤝 Contributing

Want to contribute a model? See HOW-TO-TRAIN.md and submit a PR!

Model submission requirements:

Minimum 1,000 patterns
100% embedding coverage
70% average confidence
<10 KB per pattern
Comprehensive README
Validation report

💡 Support

Documentation: See HOW-TO-USE.md and HOW-TO-TRAIN.md
Issues: GitHub Issues
Examples: Check each model's README.md

Happy reasoning! 🧠✨

"The best AI doesn't just answer questions - it learns from experience."