ComputerUseAgent

A sophisticated Deno-based CLI tool that provides AI-powered file editing and bash command execution capabilities using Claude API with configurable model selection.

Features

• Dual Mode Operation:
  • Editor mode for AI-assisted text file manipulation
  • Bash mode for intelligent command execution
  • Hybrid mode for combined capabilities
• Configurable AI Models:
  • Support for multiple Claude models (3.5 Sonnet, 3.7 Sonnet, 4 Sonnet, 4 Opus)
  • Smart planner model selection based on main model choice
  • Model-specific cost tracking and pricing
  • Dynamic tool version selection based on selected model
  • Automatic beta header configuration for optimal compatibility
  • Easy model switching via settings commands
• Enhanced Export System:
  • Prompt ID-based export functionality
  • Dedicated export paths
  • Rich session data export
  • Comprehensive session logging
• Improved User Management:
  • Streamlined settings management
  • Enhanced session handling
  • Efficient tool configuration
• Memory Management
• Comprehensive Logging
  • Text response content tracking
  • Debug logging for history entries
  • Multi-model token usage tracking
• Advanced Cost Calculation
  • Model-specific pricing
  • Per-model cost breakdown
  • Cost warnings for expensive models
• History Management with database integration
• Clipboard Management with cross-platform support
• Tool Configuration Management
• User Settings Management
• Jina API Integration for enhanced search capabilities

Demo

ComputerUseDemo.mp4

This is a trimmed demo, Watch full demo on Youtube

Prerequisites

• Deno installed
• Anthropic API Key

Setup

1. Clone the repository:

git clone <repository-url>
cd ComputerUseAgent

2. Set up your environment:

# Set your Anthropic API key
export ANTHROPIC_API_KEY="your-api-key"

3. Build the project:

deno task build

Usage

Direct Execution

# Editor mode
deno run -A src/main.ts --mode=editor "your prompt"

# Bash mode
deno run -A src/main.ts --mode=bash "your command"

# Bash mode with mock execution
deno run -A src/main.ts --mode=bash --no-agi "your command"

Using Built Binary

# After building
./build/ComputerUseAgent --mode=editor "your prompt"
./build/ComputerUseAgent --mode=bash "your command"
./build/ComputerUseAgent --export "prompt-id" # Export session data

Model Configuration

Configure which Claude model to use for AI operations:

# Set model to Claude 4 Sonnet
deno run -A src/main.ts settings --set-model "4-sonnet"

# List available models
deno run -A src/main.ts settings --list-models

# View current settings including selected model
deno run -A src/main.ts settings --list

Available Models:

• 3.5-sonnet - Claude 3.5 Sonnet (Default, most cost-effective)
• 3.7-sonnet - Claude 3.7 Sonnet (Enhanced reasoning capabilities)
• 4-sonnet - Claude 4 Sonnet (Latest generation)
• 4-opus - Claude 4 Opus (Most capable, higher cost)

Smart Features:

• Automatic planner model selection based on your chosen model
• Model-specific cost tracking and warnings
• Thinking capabilities automatically enabled for supported models
• Dynamic tool version selection ensuring compatibility with each model
• Automatic beta header configuration for optimal API compatibility

Project Structure

• src/: Source code directory
  • config/: Configuration files
  • modules/: Core functionality modules
    • bash/: Bash command execution
    • editor/: Text editor operations
    • memory/: Memory management
  • types/: TypeScript interfaces
  • utils/: Utility functions

Configuration

Key configuration files:

• src/config/constants.ts: System-wide constants and API settings
• src/config/logging.ts: Logging configuration
• src/config/tool_config.ts: Tool configuration
• deno.json: Deno project configuration

Documentation

• See CHANGELOG.md for detailed version history and latest changes
• CONVENTIONS.md for development patterns and best practices

Development

# Run in development mode with file watching
deno task dev

# Build for local use
deno task buildLocal

Logging

Logs are stored in app.log with both console and file output. The logging system tracks:

• User inputs
• API usage
• Command execution
• Errors
• Token usage and costs

Contributing

1. Fork the repository
2. Create your feature branch
3. Commit your changes
4. Push to the branch
5. Create a new Pull Request

License

See LICENSE file.

Releases

Installation

You can download pre-built binaries for your platform from:

• Latest stable release: Releases page
• Nightly builds: Nightly Release

Binaries are automatically built for Windows, Linux, and macOS.