openonion/connectonion

What it solves

ConnectOnion is a toolkit for developers to build, debug, and deploy production-ready AI agents. It removes the boilerplate of setting up backends, frontends, and API layers, allowing developers to focus solely on writing prompts and defining tools.

How it works

The framework uses a template-first approach where agents are created via a CLI (co create). It converts standard Python functions into LLM tools automatically using type hints and docstrings. To move from development to production, the host() function turns an agent into a remotely callable service with an integrated HTTP server and P2P relay. It also includes a built-in chat interface for interaction.

Who it’s for

It is designed for Forward Deployment Engineers (FDEs) and developers who want to move quickly from a working prototype to a deployed agent without building a full application stack from scratch.

Highlights

• Integrated Tooling: Built-in tools for bash, browser automation, Gmail, Outlook, Google Drive, and SMS.
• Interactive Debugging: The @xray decorator allows developers to pause execution at breakpoints, inspect variables, and edit state in real-time.
• Plugin System: A lifecycle hook system with 12 hooks to inject logic (e.g., reflection, context compression) at any point in the agent's execution.
• Skills System: Reusable workflows that support auto-discovery and are compatible with Claude Code skills.
• Multi-Agent Trust: A pre-LLM trust system (open, careful, strict) that manages how agents call each other without consuming tokens.
• Built-in AI Programmer: Includes co ai, an AI coding assistant built using the framework itself to help users write agent code.