Janus是一款用Go语言编写的开源项目，以单一二进制文件的形式提供本地大语言模型推理能力1。该工具支持GGUF格式模型通过Vulkan在AMD、Intel和Nvidia等GPU上运行，同时也可进行CPU推理1。Janus暴露OpenAI兼容的API接口（包括/v1/chat/completions等端点）和内置Web UI，用户可通过本地地址http://127.0.0.1:8990访问[1](#source-1)。
该项目内置了多项实用工具，包括文件读写、命令执行、数学计算、DOCX及PDF处理、OCR等功能1。Janus无需依赖Python、Docker或Ollama环境即可独立运行，尽管也支持将Ollama作为后端选项1。从资源需求看，模型占用磁盘空间通常为2-8GB，而Janus本身仅需约50MB空间1。首次启动时模型加载预期耗时10-60秒1。
该项目采用MIT许可证开源发布，代码可在https://github.com/Vibra-Ingenn/Janus.git获取[1](#source-1)。
Janus is an open-source Go project that provides a single binary for running large language models in GGUF format locally on consumer hardware 1. The tool supports GPU acceleration through Vulkan on AMD, Intel, and Nvidia devices, as well as CPU-based inference, without requiring Python, Docker, or Ollama dependencies 1.
The binary exposes an OpenAI-compatible API with standard endpoints such as /v1/chat/completions and includes a built-in web interface 1. It also incorporates a suite of integrated tools covering file operations, command execution, mathematical calculations, document processing for DOCX and PDF formats, and optical character recognition 1. Model loading on first startup typically takes between 10 to 60 seconds, with the API accessible by default at http://127.0.0.1:8990 1.
Storage requirements are modest: individual models range from 2 to 8 GB, while Janus itself occupies approximately 50 MB 1. The project is distributed under the MIT license and can be found at https://github.com/Vibra-Ingenn/Janus.git 1. Though designed as a standalone tool, it also supports Ollama as an alternative backend 1.
评论
还没有评论，欢迎留下第一条。