Jay-Code: Enterprise AI Orchestration with Quality Control

Jay-Code is a production-grade AI CLI orchestration engine that enforces quality through hardware requirements and mathematical validation.

Quality Assurance Through Hardware Requirements

Local High-Performance Mode:

• Minimum: 40GB VRAM + 64GB RAM
• Model: Llama 3.3 70B Instruct
• Performance: Maximum quality, full privacy, deterministic results

API Fallback Mode:

• Requirement: Anthropic API key
• Model: Claude 3.5 Sonnet
• Performance: High quality, cloud-based processing

One-Command Installation

Automatic Quality Detection

curl -fsSL https://raw.githubusercontent.com/ruvnet/jay-code/main/install.sh | bash
The installer automatically:
- Detects system capabilities (VRAM/RAM)
- Installs Llama 3.3 70B if hardware supports it
- Falls back to Claude API configuration if insufficient resources
- Configures optimal settings for your system

## System Requirements

### For Local Model (Recommended)
- **VRAM**: 40GB+ (RTX A6000, RTX 6000 Ada, A100, H100)
- **RAM**: 64GB+ system memory
- **Storage**: 100GB+ for model files
- **OS**: Linux, macOS (Windows via WSL)

### For API Mode
- **Requirements**: ANTHROPIC_API_KEY environment variable
- **Resources**: Standard system (8GB+ RAM recommended)

## Architecture

**Single Model Intelligence** with mathematical agent coordination:

- **Model Manager**: Circuit breaker patterns, automatic fallback
- **Planning Agent**: Graph theory task decomposition (342 lines)
- **QA Agent**: Statistical analysis with 80%+ pass threshold (514 lines)
- **Configuration**: Type-safe validation with hardware detection

## Quick Start

```bash
Set API key (required for fallback or API-only mode)
export ANTHROPIC_API_KEY=your_api_keyGenerate enterprise-grade code
jay-code generate "Create a TypeScript REST API with authentication"With specific requirements
jay-code generate "Build GraphQL resolver" --requirements "TypeScript,validation,error-handling"Check system status
jay-code healthView performance metrics
jay-code metrics
## Quality Control Features

### Automatic Hardware Detection
```bash
Mathematical Validation

Complexity Analysis: Entropy-based text evaluation
Quality Scoring: Statistical code analysis (80%+ threshold)
Performance Bounds: Sub-5 second cold start guarantee
Reliability Patterns: Circuit breakers with failure analysis

Production Standards

Deterministic Operations: Formal verification requirements
Enterprise Compliance: Security, performance, maintainability validation
Mathematical Rigor: Proven complexity bounds throughout
Artifact Persistence: Cryptographic hashing for reproducibility

Configuration
Automatically generated based on system capabilities:
High-Performance Local:
{
  "models": {
    "primary": {
      "type": "ollama",
      "model": "llama3.3:70b-instruct",
      "endpoint": "http://localhost:11434"
    },
    "fallback": {
      "type": "anthropic",
      "model": "claude-3-5-sonnet-20241022"
    }
  },
  "mode": "local-primary"
}
API-Only Mode:
{
  "models": {
    "primary": {
      "type": "anthropic",
      "model": "claude-3-5-sonnet-20241022"
    }
  },
  "mode": "api-only"
}
Uninstall Options
Clean Uninstall (preserves Ollama/models):
curl -fsSL https://raw.githubusercontent.com/ruvnet/jay-code/main/scripts/uninstall-jay-code.js | node

Full Uninstall (removes everything):
curl -fsSL https://raw.githubusercontent.com/ruvnet/jay-code/main/scripts/uninstall-jay-code-full.js | node
Development Transformation
Built from claude-flow v2.0.0-alpha.90, delivering on actual capabilities:
Eliminated:

False claims (87 tools → actual implementations)
Neural network bloat (84% directory reduction)
Hive-mind complexity (452 directories removed)
Redundant systems (61 backup files removed)

Delivered:

1,482 lines of mathematically rigorous core logic
Deterministic algorithms with formal verification
Production-grade reliability patterns
Hardware-aware quality control

Support

Repository: https://github.com/ruvnet/jay-code
Issues: Hardware compatibility, model performance
Documentation: Installation guides, architecture specs

Jay-Code transforms AI development tooling from marketing promises to mathematical reality.