# AntNest

A desktop AI agent that runs entirely on the user's own machine, splitting a task between a queen process that only plans and decides and disposable worker processes that execute commands in an isolated temporary directory.

AntNest is a local desktop AI agent built around a queen-and-worker architecture: the queen process only thinks, plans and decides and never executes a command itself, and when work needs doing it spawns a worker that copies the agent's own code into an isolated temporary directory, executes there, reports the result back, and is then destroyed.

It connects to an OpenAI-compatible endpoint such as DeepSeek, MiniMax, Kimi or a local Ollama server, streams the queen's reasoning live rather than hiding it, and keeps a memory layer combining a tree-based index with retrieval-augmented recall over saved notes. Up to ten workers can run per task layer with automatic retry on failure, and dangerous commands are filtered before execution. It ships as a Windows installer or can be run from source with a Python package manager.

Website: https://github.com/llxpy/AntNest