vinhnx/VTCode

What it solves

VT Code is a terminal-based coding agent designed to bridge the gap between simple AI chat and fully autonomous software engineering. It allows developers to move from a high-level question to a reviewed code change without leaving the terminal, providing a secure environment for both interactive exploration and long-running autonomous workflows.

How it works

Built in Rust, VT Code combines a responsive Terminal User Interface (TUI) with a suite of safe terminal tools. It integrates with a wide variety of LLM providers (including local inference via Ollama or llama.cpp) and utilizes open protocols like the Model Context Protocol (MCP) and Agent Client Protocol (ACP) to extend its capabilities. The agent can perform safe file operations, search code using ripgrep and ast-grep, and execute terminal commands within a restricted shell sandbox. It also features a "plan-build-evaluate" workflow where a primary agent creates a build plan that is reviewed by the user before being handed off to autonomous build agents.

Who it’s for

It is designed for developers who prefer the terminal over a GUI, those who want to integrate AI agents into their existing CLI workflows, and teams requiring strict security guardrails (such as shell sandboxing and provider whitelisting) for AI-driven code modifications.

Highlights

• Multi-Provider Support: Built-in support for over 30 cloud LLMs and local inference backends.
• Extensible Architecture: Supports Agent Skills, MCP, and Agent Plugins for adding new capabilities.
• Coded Safety: Includes a restricted shell sandbox, tool guardrails, and per-workspace approval for shell commands.
• WebMCP Bridge: An optional browser bridge that connects a browser-based editor to an active terminal session.
• Autonomous Workflows: Features worktree isolation for parallel agents and a structured planning-to-implementation pipeline.