Skip to content

Latest commit

 

History

History
170 lines (123 loc) · 6.14 KB

README.en.md

File metadata and controls

170 lines (123 loc) · 6.14 KB
Raw

First and foremost

I apologize for not updating for several months. The reason is that I have been working on the PLUS version. In the future, I will gradually transfer the features of the PLUS version to the open-source version. I also hope that everyone can provide new suggestions.


Preview


Introduction

This project comes with a built-in server for local development and testing, and can also be directly deployed online for personal learning use. Want to use your own API? No problem, the client can connect to the server of this project as well as any official standard API.

We recommend connecting to AIOK, which is free for the first registration and provides 100,000 points for free.

The following models are currently supported:

Model Description Max Tokens Training Data Pros Cons
gpt-3.5-turbo One of the most capable GPT-3.5 models, optimized for chat at 1/10th the cost of text-davinci-003. 4096 tokens Up to Sep 2021 Faster response, slower token usage Slightly less intelligent than GPT-4
gpt4 More capable than any GPT-3.5 model, able to do more complex tasks and optimized for chat. 8192 tokens Up to Sep 2021 More intelligent, more accurate problem solving Slower response time and higher token usage compared to GPT-3.5

You can choose the model for each conversation on the conversation settings page or when creating a new conversation.

For the GPT4 model, it is important to select it in the configuration page and enter an API key available for that model.

Getting Started

Node.js v18.x.0 is recommended.

If you have nvm installed locally, you can use nvm use to switch to the recommended version of Node.js.

npm run inst # Installation

npm run dev # Running

npm run dev:env # Generating environment variable files during runtime

npm run dev:vite # Run only the frontend if you don't want to use the local API.

After using the npm run dev:env command, if you need a local server, you need to fill in the correct and available API KEY in the .env.development file generated under the /api directory.

If you use a local server, Chinese users may need to proxy the interface themselves.

Using the official interface directly

Modify your local VITE_API_HOST, and enter your own API KEY in the settings (shown below) and click Save.

VITE_API_HOST="https://api.openai.com/v1/chat/completions"

Don't want to set environment variables?

No problem, just need to fill in the official API endpoint directly in the settings as shown in the figure. https://api.openai.com/v1/chat/completions

Then enter the correct and available API KEY.

Environment Variables

Please refer to the .env.example file.

# Prompt for the input box
VITE_DEFAULT_PLACEHOLDER='send message to AI'

# Bottom message of the input box
VITE_DEFAULT_BOTTOM_TIPS=''

# API_HOST
VITE_API_HOST='http://localhost:3000/aios-chat'

# GPT4 API_HOST
VITE_API_HOST_GPT4='https://api.openai.com/v1/chat/completions'

# Whether to return only text content, which means whether the stream needs to be parsed by the frontend.
VITE_ONLY_TEXT=true

# The maximum number of dialogues to be cached excluding system information
VITE_CACHE_TIMES=10

# BASE_URL
VITE_BASE_URL='/' # BASE_URL

# AI avatar url
VITE_AI_AVATOR_URL=''

# User avatar url
VITE_USER_AVATOR_URL=''

# Home page logo url
VITE_LOGO_URL=''

# Display information below the homepage logo
VITE_INFO=''

Deployment

To be updated later.

Features

  • Contextual conversation
  • Chat history
  • Markdown rendering
  • Katex mathematical formula support
  • Share current conversation with one click
  • Streaming transmission
  • Dark mode
  • Supports GPT-4

Upcoming Features

  • github action（eslint）
  • custom API key
  • Pre-set scenario-based robot system applications
  • prompt store
  • Adding a BFF layer to the backend
  • Adding http and socks proxy configuration or providing some pre-installed openai interface proxies
  • Access control
  • Material modification of openai model (3.5, 3.0) and modification of interface parameters temperature, frequency_penalty, presence_penalty
  • Multi-language support
  • One-click deployment document
  • docker-compose files

FAQ

Q: Why is it that after using the official interface, a large character string is returned without being parsed?

A: Because the VITE_ONLY_TEXT in the environment variables is set to true by default, there is no parsing on the client side. Change it to false or delete it.

The default value is now set to false.

License

MIT License © AIOS-club