Skip to content
aaron edited this page Mar 19, 2026 · 7 revisions

ConnectOnion Wiki - Python AI Agent Framework

Build powerful AI agents in Python with minimal code

ConnectOnion is a lightweight Python framework for creating AI agents that can use tools, make decisions, and interact with the world. This wiki provides tutorials, examples, and guides to help you build amazing AI-powered applications.

πŸš€ Quick Links

Getting Started

Features & Guides

Examples

Reference

πŸ“– What is ConnectOnion?

ConnectOnion helps you build AI agents that can:

  • Use tools - Search, send emails, browse the web, run code
  • Make decisions - Plan multi-step tasks and adapt to results
  • Debug easily - Interactive debugging with auto_debug()
  • Scale simply - From prototypes to production

Get Started in 2 Minutes

# Install
pip install connectonion

# Create your agent
co create my-agent
cd my-agent

# Run it!
python agent.py

That's it! The CLI handles API key setup automatically.

🎯 Why ConnectOnion?

Most frameworks give you a way to call LLMs. ConnectOnion gives you everything around it β€” so you only write prompt and tools.

Keep Simple Things Simple

  • 2-line agent creation: Agent("name").input("task")
  • Functions become tools automatically β€” no schema writing, no interface wiring
  • No complex configuration files

Make Complicated Things Possible

  • Interactive debugging with auto_debug()
  • Behavior tracking and automatic logging
  • Trust policies for multi-agent security
  • Plugin system with Claude Code-equivalent capabilities

What Other Frameworks Don't Have

Built-in AI Programmer β€” Run co ai to get an AI coding assistant that deeply understands ConnectOnion. It's built with ConnectOnion and fully open-source.

Built-in Frontend & Backend β€” Other frameworks: write agent β†’ build FastAPI β†’ build React frontend β†’ wire APIs β†’ deploy. ConnectOnion: write prompt and tools β†’ deploy. Frontend at chat.openonion.ai, all open-source.

Ready-to-Use Tool Ecosystem β€” bash, Shell, FileTools, BrowserAutomation, Gmail, Outlook, GoogleCalendar, Memory, TodoList β€” import and use. Need to customize? co copy Gmail copies the source into your project.

Built-in Approval System β€” Dangerous operations automatically trigger approval. Plugin-based: turn it off, customize it, or replace it.

Skills System β€” Reusable workflows with automatic permission scoping. Auto-discovers Claude Code skills from .claude/skills/ β€” no conversion needed.

Claude Code-Equivalent Plugins β€” auto_compact (context compression), subagents (sub-agent spawning), ulw (autonomous work mode), tool_approval (web approval UI) β€” these mirror Claude Code's internal capabilities, open to any agent.

Multi-Agent Trust System β€” Trust decisions happen before LLM involvement β€” zero token cost for 90% of cases. Three presets: open (dev), careful (staging), strict (production).

πŸ“š Learning Path

1. Beginner (Start here!)

  1. Quick Start Guide - 5 minutes
  2. Building Your First Agent - 15 minutes
  3. Email Agent Example - Working code to study

2. Intermediate

  1. Creating Custom Tools
  2. Interactive Debugging Guide
  3. Debug Agent Errors

3. Advanced

  1. Deploy to Production
  2. Web Scraping Agent
  3. Official Documentation - Full API reference

πŸ”— Official Resources

πŸ†˜ Need Help?

πŸ“ Contributing

Found an error? Want to add a tutorial?


Ready to build your first AI agent? β†’ Start with the Quick Start Guide

Clone this wiki locally