# AuroraCoder

An autonomous AI coding agent that runs inside a Docker sandbox with a persistent shell, a VNC desktop, and parallel tool use, reading a codebase, writing code, running commands, browsing the web, and delegating to sub-agents.

AuroraCoder is an autonomous coding agent, powered primarily by DeepSeek V4 Pro with GLM and OpenCode Go also supported, that executes real tasks inside a Docker sandbox rather than only holding a chat conversation: it reads a codebase, writes code, runs shell commands, searches the web, delegates to sub-agents, and can launch GUI applications visible through a built-in VNC desktop. Its web interface shows a streaming chat panel, a collapsible file tree with a diff viewer, and a visualization of the model's reasoning tokens.

A prebuilt launcher binary embeds the whole project and builds the Docker image on first launch, needing only Docker Desktop with no separate git clone or language runtime installation, while a developer path clones the repository and runs setup scripts directly. An NVIDIA GPU passthrough path adds PyTorch and vLLM acceleration for locally run models alongside the cloud-hosted DeepSeek and GLM options.

Website: https://github.com/1001WillsStudio/AuroraCoder