Skip to content

A modular AI agent framework with secure CLI and toolkit architecture

Notifications You must be signed in to change notification settings

WeismannS/Proto-agent

Repository files navigation

Proto-Agent

An educational AI agent framework demonstrating capability-based security and modular toolkit architecture. Built for learning secure AI agent patterns with human oversight and permission controls.

Features

  • Capability-based security with granular permission controls
  • CLI tool with human-in-the-loop approval for dangerous operations
  • Python framework for building custom agents with programmatic control
  • Modular toolkits for file operations, system monitoring, and version control
  • Educational focus - clear, readable code demonstrating AI agent security patterns

Quick Start

Installation

pip install proto-agent 
# or if you prefer the cli to be used from anywhere
uv tool install proto-agent  # Recommended
# or using pipx
pipx install proto-agent

Configuration

proto-agent --help # View CLI options, which include your config path for your OS 
# Example config path for Linux: ~/.config/proto-agent/ will have .env file and config.toml 

For Model configuration, please refer to the Litellm documentation for your exact name of the model you want to use.

CLI Usage

# Safe read-only analysis
proto-agent "Analyze this codebase structure" ./my_project --read-only

# Interactive execution with approval prompts
proto-agent "Run the test suite" ./my_project
# Prompts: "Allow execution of function 'run_python_file'? (y/N):"

Framework Usage

from proto_agent import Agent, AgentConfig
from proto_agent.tool_kits import FileOperationToolkit

# Autonomous mode - no human approval needed
agent = Agent(AgentConfig(
    api_key="your_api_key",
    working_directory="./my_project",
    tools=[FileOperationToolkit(
        enable_read=True,
        enable_write=False,    # Disable risky operations
        enable_execute=False
    ).tool]
))

response = agent.generate_content("Analyze this project's structure")
print(response.text)

Key Concepts

  • CLI Mode: Interactive approval prompts for dangerous operations
  • Framework Mode: Full programmatic control over permissions
  • Capability Flags: Enable/disable specific operations per toolkit
  • Human Oversight: Configurable approval gates for security

Available Toolkits

  • 📁 FileOperationToolkit: File reading, writing, and execution
  • 💻 SystemInfoToolkit: System monitoring and resource information
  • 🔧 GitToolkit: Version control operations with safety controls

Documentation

Educational Goals

Proto-Agent prioritizes learning and security over performance:

  • Demonstrates secure AI agent design patterns
  • Shows human-in-the-loop safety controls
  • Provides clear, modular architecture examples
  • Emphasizes permission-based security models

Perfect for developers learning about AI agents, security patterns, or needing a foundation for experimentation.

License

MIT License - see LICENSE file for details.


⭐ Star the repo to support educational AI agent development!

About

A modular AI agent framework with secure CLI and toolkit architecture

Topics

Resources

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages