HuixiangDou

What it is

HuixiangDou is a GraphRAG-based solution designed for group chat scenarios, acting as a knowledge assistant built on LLMs. It offers a two-stage pipeline involving preprocessing, rejection, and response to cope with group chat dynamics. It provides a Web, Android, and pipeline source code setup, with CPU-only and multimodal configurations, and integrations for Feishu/WeChat, Lark, OpenXLab Web, Gradio, and HTTP server usage.

How it works

The project describes a three-stage/process-oriented approach including preprocess, rejection, and response. It supports dense/document and knowledge-graph retrieval, as well as image and text retrieval, with multiple LLM backends via a configurable API (kimi, deepseek, vllm, siliconcloud, ppio, etc.). Retrieval methods include Knowledge Graph, Internet Search, and SourceGraph, plus image-text modalities.

Getting started

Recommended steps in the README include:

• Install dependencies and login to Hugging Face:

huggingface-cli login

• Install dependencies:

# parsing `word` format requirements
apt update
apt install python-dev libxml2-dev libxslt1-dev antiword unrtf poppler-utils pstotext tesseract-ocr flac ffmpeg lame libmad0 libsox-fmt-mp3 sox libjpeg-dev swig libpulse-dev
# python requirements
pip install -r requirements.txt

• Create knowledge base and build features:

cd HuixiangDou
mkdir repodir
cp -rf resource/data* repodir/

mkdir workdir
python3 -m huixiangdou.services.store

• Configure and run LLM API, then test:

# Example for vllm
vllm serve /path/to/Qwen-2.5-7B-Instruct --served-model-name vllm --enable-prefix-caching --served-model-name Qwen-2.5-7B-Instruct

• Run a simple web UI or server:

python3 -m huixiangdou.gradio_ui

python3 -m huixiangdou.api_server

• Update documents and QA datasets under repodir as directed in the README.

Getting started (config and tests)

The README provides a sample config section showing remote_llm_model options and remote_type choices (e.g., vllm, deepseek, kimi, siliconcloud, ppio). It also includes example QA responses and a test harness using python3 -m huixiangdou.main. It also notes a Gradio UI and HTTP server interfaces for testing.

Recent releases

Latest releases include 20251117 with changes such as support for kimi k2 and step, refactored knowledge base building, and a v0.1.0 bump for the next major release. Earlier releases include 20240722 and 20240415 with frontend additions and feature improvements. The release notes list feature changes and contributors.

Traction

Stars: 2496 Forks: 178 Open issues: 37

Behind the repo

The project is linked with multiple integrations and deployment targets, including WeChat, Lark, OpenXLab Web, and Read the Docs. It references multiple LLMs and retrieval methods and provides web/frontend/backend source code and a gradio UI.

Caveats

License: BSD-3-Clause. Created 2023-12-28; last_push 2025-11-24. Hardware notes indicate CPU-only edition and 10G multimodal edition with specific setup steps and model paths. The README includes external links for WeChat and other integrations and notes the need to download model weights and Visualized_m3.pth for multimodal configurations.