General Introduction
HuixiangDou is a large-scale language model (LLM)-based group chat assistant designed to cope with group chat scenarios through a three-stage pre-processing, rejection and response process. It is able to answer user questions without causing information flooding. The project provides complete Web, Android, and pipeline source code that is industrial-grade and commercially viable.HuixiangDou supports multiple instant messengers, such as WeChat and Lark, and is capable of handling multimodal messages.
Function List
Intelligent Q&A for Group Chat Scenarios
Support for instant messengers such as WeChat and Lark
multimodal information processing
Web and Android versions available
Low memory requirements, minimum 2GB
Configuration example | VGA memory requirements | descriptive |
---|---|---|
config-cpu.ini | - | Using the siliconcloud API Search text only |
config-2G.ini | 2GB | Using the openai API (e.g. kimi, deepseek and stepfun) Search text only |
config-multimodal.ini | 10GB | LLM with openai API, graphical search |
[Standard version]config.ini | 19GB | Local Deployment LLM, Unimodal |
config-advanced.ini | 80GB | Local LLM, denotational disambiguation, unimodal, microtome utility |
Can be used without training
Supports multiple model access
LLM | file format | Search method | instant messaging (IM) | preprocessing |
|
|
|
Using Help
Visit the GitHub page for source code and documentation
Join the WeChat or Lark group chat experience
Check out the demo videos on BiliBili and YouTube!
Refer to the README file for installation and configuration
Our web version has been posted toOpenXLabYou can create knowledge bases, update positive and negative samples, enable web search, test chats, and integrate into Flybook/WeChat groups. ViewBiliBilicap (a poem)YouTube!