yyjeqhc/webcodex

What it solves

WebCodex provides a way for cloud-based AI agents (like ChatGPT or Claude) to interact with a real development environment on your local machine. This eliminates the need to copy codebases into a hosted workspace or chat interface, allowing agents to work directly with your local files, Git checkouts, and installed developer tools.

How it works

WebCodex uses a self-hosted architecture consisting of a Server and a Runner. The AI client connects via MCP (Model Context Protocol) or HTTPS to the WebCodex Server, which then routes requests to a Runner installed on the machine where the code resides. The Runner executes the commands, interacts with the Git repository, and returns the results back to the AI agent.

Who it’s for

Developers who want to use AI coding agents while keeping their source code on their own hardware and utilizing their own local toolchains (compilers, tests, and formatters) rather than relying on cloud-hosted environments.

Highlights

• Local Code Sovereignty: Your code stays on your machine; it is not copied into the cloud service.
• Real Toolchain Access: Agents can run actual local commands, tests, and compilers.
• Git Integration: Agents can inspect status and diffs, keeping operations visible and reviewable.
• Persistence: Long-running jobs remain observable and survive beyond a single AI request turn.
• Multi-platform Support: Provides installers for Windows, macOS, and Linux (Debian/Ubuntu).
• Crosstalk with MCP: Native support for the Model Context Protocol to connect with various AI clients.