Orcas

Transform natural language into safe, executable bash commands using local AI models.

Features

• Local AI model execution with no cloud dependencies
• Security-first design with command validation
• No command chaining to prevent dangerous operations
• Interactive approval for multistep commands
• Smart sudo privilege handling

Installation

git clone https://github.com/Mogalina/orcas.git
cd orcas
./scripts/install.sh

Usage

# Single command mode
orcas find all python files larger than 1MB
orcas create backup of documents folder
orcas show disk usage sorted by size

# Interactive mode
orcas --interactive

# Preview commands without executing
orcas --dry-run compress logs folder

Configuration

Configuration file: ~/.config/orcas/config.yaml

model:
  name: "gemma-3"
  path: "~/.orcas/models"
  
security:
  require_confirmation: true
  allow_sudo: true
  
execution:
  timeout: 300

Supported Models

Model	Parameters (B)	Size (GB)
Gemma3	12.2	7.3
Phi-2	2.7	1.2
TinyLlama	1.1	0.7

Security

• Command validation before execution
• No automatic command chaining
• Sudo privilege confirmation
• Dangerous command blacklist
• Command preview before execution

Documentation

• Installation Guide
• Usage Guide
• Architecture

Requirements

• Python 3.8 or higher
• 4GB RAM minimum
• 8GB disk space for models

License

MIT License