Skip to content

Introduction

OpenZess is an open-source, self-hosted autonomous agent platform that gives AI models direct access to your local system through a secure, sandboxed environment.

The Vision

Most AI chat interfaces are isolated text boxes — they can generate code but can't run it, they can describe files but can't create them. OpenZess breaks this barrier entirely.

When you chat with an OpenZess agent, it can:

  • Execute terminal commands inside a sandboxed Linux environment
  • Create, read, and edit files on your system
  • Search the web and scrape URLs for live information
  • Control a virtual desktop via mouse and keyboard
  • Schedule background tasks that run autonomously
  • Connect to external services via the Model Context Protocol (MCP)

Core Philosophy

1. Safety First

Every command executes inside a Debian WSL sandbox. The agent physically cannot touch your Windows host. Even GUI interactions are trapped inside an invisible X11 virtual display.

2. Provider Agnostic

OpenZess doesn't lock you into a single AI provider. Through LiteLLM, it routes to 10+ providers including Google Gemini, OpenAI, Anthropic, Groq, DeepSeek, Qwen, and local Ollama models.

3. Extensible by Design

The plugin system supports two extension mechanisms:

  • MCP Servers: Industry-standard protocol for connecting external tools
  • Python Plugins: Drop a .py file into backend/plugins/ and it's instantly available to your agent

System Requirements

ComponentMinimumRecommended
OSWindows 10 + WSL2Windows 11 + WSL2
RAM4 GB8 GB+
Node.jsv18+v20+
Python3.10+3.12+
BrowserChrome/EdgeChrome

Released under the MIT License.