Skip to main content

​
Overview

Local Models support allows you to run AI models directly on your machine, ensuring complete privacy and security of your code while providing fast, offline AI assistance.
Coming Soon: Local Models support is currently in development and will be available to beta users first.

​
Why Use Local Models?

​
Privacy and Security

  • No Data Transmission: Your code never leaves your machine
  • No Internet Required: Work completely offline
  • No Third-Party Access: No external services can access your code
  • Corporate Compliance: Meet strict security requirements
  • Full Ownership: You control all your data
  • No Logging: No usage logs or analytics
  • Custom Policies: Set your own data retention rules
  • Audit Trail: Complete control over data access

​
Performance Benefits

  • No Network Latency: Instant responses
  • No Rate Limits: Use as much as you need
  • Consistent Performance: No network-related slowdowns
  • Predictable Costs: No per-request charges
  • Always Available: No internet connection required
  • No Service Outages: Independent of external services
  • Consistent Quality: Same model performance every time
  • Custom Tuning: Optimize for your specific use cases

​
Supported Platforms

​
Ollama Integration

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# Pull a model
ollama pull codellama

# Start the model
ollama serve
  • Code Llama: Specialized for code generation
  • Llama 2: General-purpose language model
  • Mistral: Fast and efficient coding assistant
  • Custom Models: Your own fine-tuned models

​
LM Studio Integration

  • Model Management: Easy model switching
  • Performance Tuning: Optimize for your hardware
  • Custom Prompts: Fine-tune model behavior
  • Resource Monitoring: Track GPU/CPU usage
  • GGML: Optimized for CPU inference
  • GGUF: Next-generation format
  • ONNX: Cross-platform compatibility
  • Custom Formats: Your preferred model format

​
Hardware Requirements

​
Minimum Requirements

  • RAM: 16GB minimum (32GB recommended)
  • CPU: 8-core processor (Intel i7/AMD Ryzen 7 or better)
  • Storage: 50GB available space
  • OS: macOS 12.0 or later
  • GPU: Apple Silicon M1/M2/M3 or NVIDIA GPU with 8GB+ VRAM
  • RAM: 32GB minimum (64GB recommended)
  • Storage: 100GB available space
  • Performance: 10x faster than CPU-only

Basic Setup

  • Apple Silicon M1/M2
  • 16GB RAM
  • 256GB SSD
  • Good for small to medium projects

Professional Setup

  • Apple Silicon M2 Pro/Max
  • 32GB RAM
  • 512GB SSD
  • Ideal for large codebases

Enterprise Setup

  • Apple Silicon M3 Max
  • 64GB+ RAM
  • 1TB+ SSD
  • Maximum performance and capability

Custom Setup

  • Custom hardware configuration
  • Optimized for specific use cases
  • Maximum flexibility and control

​
Model Selection Guide

​
Code-Specific Models

Best for: General code generation and completion
  • Size: 7B, 13B, 34B parameters
  • Specialization: Code understanding and generation
  • Performance: Excellent for Swift and iOS development
  • Resource Usage: Moderate to high
Best for: Fast responses and efficient resource usage
  • Size: 7B parameters
  • Specialization: Balanced performance and efficiency
  • Performance: Good for most coding tasks
  • Resource Usage: Low to moderate
Best for: Specialized use cases and company-specific needs
  • Size: Variable
  • Specialization: Trained on your specific codebase
  • Performance: Optimized for your patterns
  • Resource Usage: Depends on model size

​
Performance vs. Resource Trade-offs

  • Speed Priority
  • Quality Priority
  • Resource Constrained
Recommended: Code Llama 7B with GPU acceleration
  • Fast responses (1-3 seconds)
  • Good code quality
  • Moderate resource usage
  • Suitable for most developers

​
Setup and Configuration

​
Initial Setup

1

Install Ollama

# macOS installation
curl -fsSL https://ollama.ai/install.sh | sh

# Verify installation
ollama --version
2

Download Models

# Download Code Llama (recommended)
ollama pull codellama:7b

# Or download Mistral for efficiency
ollama pull mistral:7b
3

Configure Parsaa

  1. Open Parsaa settings in Xcode
  2. Go to “Local Models” section
  3. Select your preferred model
  4. Configure performance settings
4

Test Setup

// Test with simple code
func hello() {
    print("Hello, Parsaa!")
}
If you get AI suggestions for this code, your setup is working!

​
Advanced Configuration

{
  "model": "codellama:7b",
  "gpu_layers": 32,
  "context_length": 4096,
  "batch_size": 512,
  "threads": 8
}

{
  "low_vram": true,
  "mmap": true,
  "mlock": false,
  "f16_kv": true
}

{
  "system_prompt": "You are a Swift expert specializing in iOS development...",
  "user_prompt": "Help me with this Swift code:",
  "assistant_prompt": "Here's how I can help you:"
}

​
Hybrid Mode

​
Cloud + Local Processing

  • Simple Tasks: Process locally for speed and privacy
  • Complex Tasks: Use cloud for better quality
  • Fallback: Cloud processing when local model is unavailable
  • User Choice: Override automatic routing
{
  "hybrid_mode": true,
  "local_threshold": 0.7,
  "cloud_models": ["gpt-4", "claude-3"],
  "local_models": ["codellama:7b"],
  "fallback_to_cloud": true
}

​
Privacy Levels

Maximum Privacy

  • All processing local
  • No data transmission
  • Complete offline operation
  • Corporate compliance

Balanced Approach

  • Sensitive code local
  • General queries cloud
  • User-controlled routing
  • Best of both worlds

Performance Priority

  • Cloud for complex tasks
  • Local for simple tasks
  • Automatic optimization
  • Maximum speed

Custom Rules

  • Define your own rules
  • Project-specific settings
  • Team-wide policies
  • Flexible configuration

​
Troubleshooting

​
Common Issues

Solutions:
  1. Check available disk space
  2. Verify model download completed
  3. Restart Ollama service
  4. Check system resources
Solutions:
  1. Reduce model size
  2. Enable GPU acceleration
  3. Increase system RAM
  4. Optimize model parameters
Solutions:
  1. Reduce context length
  2. Enable memory mapping
  3. Close other applications
  4. Use smaller models
Solutions:
  1. Check Ollama is running
  2. Verify port configuration
  3. Restart Parsaa extension
  4. Check firewall settings

​
Performance Optimization

  • GPU Acceleration: Use Apple Silicon or NVIDIA GPU
  • Memory: Increase RAM for larger models
  • Storage: Use SSD for faster model loading
  • Cooling: Ensure adequate cooling for sustained performance
  • Model Quantization: Use quantized models for efficiency
  • Context Length: Optimize context window size
  • Batch Processing: Process multiple requests together
  • Caching: Cache frequent responses

​
Best Practices

​
Model Management

  • Model Selection
  • Resource Monitoring
  • Backup and Recovery
  • Start with smaller models
  • Test performance on your hardware
  • Upgrade to larger models as needed
  • Keep multiple models for different tasks

​
Security Considerations

Data Isolation

  • Keep models in secure directories
  • Use encrypted storage if needed
  • Restrict access permissions
  • Regular security audits

Network Security

  • Disable unnecessary network access
  • Use VPN for model downloads
  • Monitor network traffic
  • Keep software updated
Beta Feature: Local Models support is currently in development. Join our waitlist to get early access and help shape this privacy-focused feature.