adun-denton/chinvat

Chinvat is a local server for Windows that sits between AI coding assistants like Claude Code, Claude Desktop, Codex, and Cursor, and the outside world. It speaks the Model Context Protocol, a standard way for these AI coordinators to call outside tools, and gives them one place to send work that needs to happen outside the conversation, such as running a local AI model, calling a remote AI model, controlling Windows itself, or posting to services like Telegram, WordPress, WhatsApp, and other social platforms. Rather than just passing tool calls straight through, Chinvat treats each request as a job. Jobs are queued in a local database, tracked as they run, and can be broken into parent and child tasks whose results get combined back together. A policy system controls what is allowed to happen automatically versus what needs a human to approve first, which matters because some actions, like writing files or posting publicly, carry real risk. Setting it up requires Node.js 20 or newer and Git. After cloning the project, installing dependencies, and building it, a local hub and web dashboard start at a local address on your machine. From there you connect your AI coordinator of choice, and Chinvat can often install the needed configuration for you automatically, backing up your existing settings first. The project ships with several ready made modules, including connections to local and remote AI models, Windows system access, WordPress site management, Telegram with approval buttons, and early support for tools like Blender, GIMP, and Orca slicer software. Some modules need extra setup like API keys or access tokens. New modules can be added by developers following the project's documented adapter format. The README stresses that its safety layer reduces risk but does not make running arbitrary code fully safe, so admin access and untrusted content should be handled carefully.