I built Codey to explore how an AI coding agent should work beyond a simple chatbot. Instead of only answering questions about code, Codey is designed to work directly with a codebase — understanding project context, editing files, running commands, and using tools to complete development tasks from the terminal.
Architecture
Under the hood, Codey is written in Rust and built around five core components: sessions, context management, model providers, agents, and tools. Each component has a separate responsibility, making the system easier to extend without tightly coupling the agent logic to a specific model or interface.
The session layer manages conversations and agent state, while the context system gathers and maintains information about the codebase. Model providers abstract LLM integrations, allowing the agent to work with different providers without changing the core architecture. Agents coordinate reasoning and task execution, while tools give the model the ability to interact with the development environment.
Agent Capabilities
Codey can analyze codebases, read and edit files, execute shell commands, and support development workflows directly from an interactive terminal interface.
The agent is designed around tool use rather than treating the model as a simple text generator. This allows the model to take actions in the environment and use the results of those actions as additional context for subsequent steps.
Extensibility
A major focus of the project is extensibility. Codey supports Skills, Model Context Protocol (MCP) servers, and plugins, allowing new capabilities and tools to be added without rewriting the core agent architecture.
This makes the system closer to a modular coding-agent platform rather than a fixed single-purpose application.
Terminal Interface
The interactive terminal interface is built around a dedicated TUI architecture, allowing users to interact with the agent while keeping conversations, tool execution, and development workflows inside the terminal.
The project uses a Rust-based terminal interface built with Ratatui and Crossterm, separating the user interface from the underlying agent and tool system.
Tech Stack
Rust, Ratatui, Crossterm, LLM Providers, AI Agents, Model Context Protocol (MCP), Skills, Plugins, Terminal User Interfaces (TUI).