chatluna Download - chatluna Source code download

chatluna

Other source code

1.0.0

Download

Chinese | English

Multi-platform model access, scalable, and multiple output formats, robot plug-in providing large language model chat services.

Project status: Steadily iterated to the official version of 1.0 (currently entered the Release Candidate stage)

screenshot

Presets	Plug-in Mode & Streaming Output	Image rendering output

characteristic

Highly scalable (based on LangChain and Koishi APIs)
? Custom dialogue presets
?️ Rate limiting and blacklisting system
? Multi-format output (text, voice, image, mixing)
? Contextual perception and long-term memory
? Three modes: chat, browse, plug-in
Content Review (passed Koishi Audit Service)

TODO (Pictures are right)

deploy

We can install this plug-in directly under Koishi to use basic functions without additional configuration.

Read this document to learn more.

Adaptation support

We currently support the following models/platforms:

Model/platform	Access method	characteristic	Things to note
OpenAI	Local Client, official API access	Customizable personality, support plug-in/browsing mode and other chat modes	API access is charged
Azure OpenAI	Local Client, official API access	Customizable personality, support plug-in/browsing mode and other chat modes	API access is charged
Google Gemini	Local Client, official API access	Fast speed, performance surpasses GPT-3.5	Need to have Gemini access account, and may charge
Claude API	Local Client, official API access	Extremely large context, in most cases, it can exceed GPT 3.5, requires API KEY, charges	May be expensive and does not support Function Call
Tongyi Qian Questions	Local Client, official API access	Ali produces domestic models with free quotas	The actual measurement effect is slightly better than that of Zhishu
Wisdom	Local Client, official API access	ChatGLM, newcomers can get free token quota	The actual test effect is slightly better than iFLYTEK Spark
iFLYTEK Spark	Local Client, official API access	Domestic model, newcomers can get free token quota	The actual measured effect is approximately equal to GPT 3.5
Wen Xin's words	Local Client, official API access	Baidu series model model	The actual test effect is slightly worse than iFLYTEK Spark
Hunyuan big model	Local Client, official API access	Tencent series of big models	The actual test effect is better than Wen Xinyiyan
Ollama	Local Client, built API access	Well-known open source model collection, supports CPU/GPU hybrid deployment, can be built locally	You need to build your own backend API and require certain configurations
GPT Free	Local Client, official API access	Local forwarding uses GPT models of other websites, project automatically configures websites, etc. without manual registration required	May fail at any time and be unstable
ChatGLM	Local Client, self-built backend API access	Can be built locally, rounding is free of money	You need to build your own backend API, which requires certain configurations. The model parameters are not large enough, resulting in insufficient chat effect.
RWKV	Local Client, built API access	Well-known open source model, can be built locally	You need to build your own backend API and require certain configurations

We support the model to provide network search capabilities:

Google (API)
Bing (API/Web)
DuckDuckGO (Lite)
Tavily (API)

Presets

Starting with version 1.0.0-alpha.10 , we use more customizable presets. The new personality preset uses yaml as the configuration file.

You can click here to view our personality file that comes with default: catgirl.yml

Our default preset folder path is你当前运行插件的koishi 目录的路径+/data/chathub/presets .

All preset files are loaded from the above folder. Therefore, you can freely add and edit preset files in this folder and then use the command to switch personality presets.

For more information, please check out this document.

Fork & Develop

Run the following instructions on any Koishi template project to clone ChatLuna:

 # yarn
yarn clone ChatLunaLab/chatluna
# npm
npm run clone ChatLunaLab/chatluna

You can replace ChatLunaLab/chatluna-koishi with your own project address after Fork.

Then edit the tsconfig.json file in the root directory of the template project and add the ChatLuna project path in compilerOptions.paths .

{
  "extends" : " ./tsconfig.base " ,
  "compilerOptions" : {
    "baseUrl" : " . " ,
    "paths" : {
      "koishi-plugin-chatluna-*" : [ " external/chatluna/packages/*/src " ]
    }
  }
}

Since the project itself is complex, the initial use must be built once.

 # yarn
yarn workspace @root/chatluna-koishi build
# npm
npm run build -w @root/chatluna-koishi

Finish! Now you can start the template project with yarn dev or npm run dev in the root project and develop ChatLuna twice.

Although Koishi supports module hot replacement (hmr), this project may not be fully compatible.
If you have a bug when you are using hmr to develop this project, please raise it in Issue and follow the steps above to rebuild the project and restart Koishi to try to fix it.

Help

Currently, the ChatLuna project team's production capacity is extremely scarce, and there is no more production capacity to achieve the following goals:

Web UI
Http Server
Project Documentation

Welcome to submit a Pull Request or discuss it, and your contribution is highly welcome!

List of contributors

Instructions for use

This project was developed by ChatLunaLab.

ChatLuna (hereinafter referred to as this project) is a dialogue robot framework based on large language models. We are committed to working with the open source community to promote the development of large-scale model technology. We strongly call on developers and other users to comply with the open source agreement to ensure that the framework and code of this project (and other derivatives based on this project promoted by the community) and related derivatives are not used for any purpose that may cause harm to the country and society, as well as services that have not been security-assessed and registered.

This project does not directly provide support for any generative artificial intelligence services, and users need to obtain the algorithm API used from organizations or individuals that provide production artificial intelligence services.

If you have used this project, please follow the laws and regulations of the local area and use the production artificial intelligence service algorithms available in the local area.

This project is not responsible for the results generated by the algorithm, and all results and operations are the responsibility of the user.

The relevant information storage of this project is all sourced by the user, and the project itself does not provide direct information storage.

This project does not assume any risks and responsibilities arising from data security, public opinion risks caused by users or the misleading, abuse, dissemination, or improper use of any model.