1001WillsStudio/AuroraCoder
An autonomous AI coding agent with novel innovations in tool state management an
AuroraCoder is an autonomous AI coding agent powered primarily by DeepSeek V4 Pro: it reads your codebase, writes code, runs commands in a Docker sandbox, searches the web, delegates to sub-agents, launches GUI applications visible through a built-in VNC desktop, and can run local LLMs with NVIDIA GPU passthrough. It is positioned as giving a frontier reasoning model a terminal, file editor, web browser, and sub-agent workforce in one isolated Linux container.
Install
Download the pre-built launcher binary from GitHub Releases, or npx aurora-coder (experimental)The one-click launcher (built by .github/workflows/release.yml) needs only Docker Desktop — no git clone, terminal, Node.js or Python; it embeds the project and builds the Docker image on first launch. The experimental npm path is npx aurora-coder (Node 18+, Python 3.10+, DEEPSEEK_API_KEY). Dev scripts (./dev-scripts/start.sh or start.bat) are the power-user path. After launch, open http://localhost:3000 and add API keys in Settings.
Compatibility
Standalone Docker-sandboxed agent; not a DeepSeek Harness plugin. Powered primarily by DeepSeek V4 Pro (GLM-5.1 and OpenCode Go also supported) with native OpenAI function calling. Optional NVIDIA GPU passthrough for PyTorch + vLLM (cu128). Web UI at localhost:3000 with VNC desktop.
Details
- Repo: 1001WillsStudio/AuroraCoder
- Category: Coding & Development
- Stars: 19
- Version: Python 3.10+ / Node 18+ / Docker required; MIT
- Last push: 2026-08-27
- First seen: 2026-05-30
Recent updates
The current README documents the one-click launcher, Docker-based dev scripts, GPU passthrough variant (gpu.sh / gpu.bat with separate AuroraCoder-GPU storage), experimental npx launcher, and the Settings flow for API keys.
FAQ
- Is AuroraCoder a DeepSeek Harness plugin?
- No — it is a standalone autonomous coding agent that runs in a Docker sandbox; it is powered by DeepSeek V4 Pro but is not a DSH plugin.
- What are the requirements?
- The easiest path needs only Docker Desktop; the experimental npm path needs Node 18+, Python 3.10+, and a DEEPSEEK_API_KEY.
- Does it support local GPU models?
- Yes — the README documents an NVIDIA passthrough variant (PyTorch + CUDA cu128 + vLLM) via dev-scripts/gpu.sh or gpu.bat.
Alternatives
morluto/flameox · cofy-x/axern · cpj-dev/dsh-plugin-cc