• Disclaimer
• Introduction
• ToDo
• Prerequisites
  • Node
  • PNPM
  • Openai Api Key
• Install dependencies
  • Backend/Service
  • Frontend/Web
• Development environment setup
  • Backend/Service
  • Frontend/Web
• Environment Variables
• Packaging
  • Docker
    • Docker parameter example
    • Docker build & Run
    • Docker compose
  • Railway Deployment
    • Railway Environment Variables
  • Manual/CI
    • Backend/Service
    • Frontend/Web
• FAQ

Disclaimer: This project is currently published on GitHub, based on the MIT license, free and used as open source learning. And there will be no sales of accounts, paid services, discussion groups, discussion groups, etc. in any form. Beware of being deceived.

Supports dual models, providing two unofficial ChatGPT API methods

Compared:

1. ChatGPTAPI uses gpt-3.5-turbo to call ChatGPT through OpenAI official API
2. ChatGPTUnofficialProxyAPI uses an unofficial proxy server to access ChatGPT’s backend API, bypassing Cloudflare (depends on third-party servers, and has a rate limit)

warn:

1. You should use the API method first
2. When using API, if the network is not available, it means that the country is blocked, and you need to build your own proxy. Never use other people's public proxy, it is dangerous.
3. When using the accessToken method, the reverse proxy will expose your access token to third parties, which should not have any adverse effects, but please consider the risks before using this method.
4. When using accessToken, no matter whether you are a domestic or foreign machine, you will use a proxy. The default proxy is acheong08 https://bypass.churchless.tech/api/conversation, which is not a backdoor or monitoring, unless you have the ability to turn over by yourself CF verification, please know before using. Community Proxy (Note: only these two are recommended, other third-party sources, please identify by yourself)
5. When publishing the project to the public network, you should set the AUTH_SECRET_KEY variable to add your password access rights, and you should also modify the title in index.html to prevent it from being searched by keywords.

Switching method:

1. Enter the service/.env.example file and copy the content to the service/.env file
2. To use OpenAI API Key, please fill in OPENAI_API_KEY field (get apiKey)
3. To use Web API, please fill in OPENAI_ACCESS_TOKEN field (get accessToken)
4. When both exist, OpenAI API Key takes precedence

Environment variables:

Please check all parameter variables or [here](#environment variable)

/service/.env.example

[✓] Dual Model

[✓] Multi-session storage and context logic

[✓] Formatting and beautifying processing of message types such as codes

[✓] Access Control

[✓] Data import, export

[✓] Save message to local picture

[✓] Multilingual interface

[✓] Interface themes

[✗] More...

node requires ^16 || ^18 || ^19 version (node >= 14 needs to install fetch polyfill), use nvm to manage multiple local node versions

node -v

If you don't have pnpm installed

npm install pnpm -g

Get Openai Api Key or accessToken and fill in local environment variables Jump

# service/.env file

# OpenAI API Key - https://platform.openai.com/overview
OPENAI_API_KEY=

# change this to an `accessToken` extracted from the ChatGPT site's `https://chat.openai.com/api/auth/session` response
OPENAI_ACCESS_TOKEN=

In order to simplify the understanding burden of back-end developers, the front-end workspace mode is not adopted, but stored in folders. If you only need to do secondary development on the front-end page, just delete the service folder.

Go to the folder /service and run the following command

pnpm install

Run the following command in the root directory

pnpm bootstrap

Go to the folder /service and run the following command

pnpm start

Run the following command in the root directory

pnpm dev

The API is available:

• Choose one of OPENAI_API_KEY and OPENAI_ACCESS_TOKEN
• OPENAI_API_MODEL set model, optional, default: gpt-3.5-turbo
• OPENAI_API_BASE_URL set interface address, optional, default: https://api.openai.com
• OPENAI_API_DISABLE_DEBUG set the interface to close the debug log, optional, default: empty is not closed

ACCESS_TOKEN is available:

• You can choose between OPENAI_ACCESS_TOKEN and OPENAI_API_KEY. When they exist at the same time, OPENAI_API_KEY takes precedence
• API_REVERSE_PROXY set reverse proxy, optional, default: https://bypass.churchless.tech/api/conversation, [Community](https://github.com/transitive-bullshit/chatgpt-api# reverse-proxy) (Note: only these two are recommended, other third-party sources, please identify by yourself)

General:

• AUTH_SECRET_KEY access authorization key, optional
• MAX_REQUEST_PER_HOUR maximum number of requests per hour, optional, default unlimited
• TIMEOUT_MS timeout in milliseconds, optional
• SOCKS_PROXY_HOST works with SOCKS_PROXY_PORT, optional
• SOCKS_PROXY_PORT works with SOCKS_PROXY_HOST, optional
• HTTPS_PROXY supports http, https, socks5, optional
• ALL_PROXY supports http, https, socks5, optional

docker build -t chatgpt-web .

# run in the foreground
docker run --name chatgpt-web --rm -it -p 127.0.0.1:3002:3002 --env OPENAI_API_KEY=your_api_key chatgpt-web

# Background process
docker run --name chatgpt-web -d -p 127.0.0.1:3002:3002 --env OPENAI_API_KEY=your_api_key chatgpt-web

# run address
https://localhost:3002/

Hub address

version: '3'

services:
  app:
    image: chenzhaoyu94/chatgpt-web # Always use latest, and re-pull the tag image when updating
    ports:
      - 127.0.0.1:3002:3002
    environment:
      # pick one of two
      OPENAI_API_KEY: sk-xxx
      # pick one of two
      OPENAI_ACCESS_TOKEN: xxx
      # API interface address, optional, available when OPENAI_API_KEY is set
      OPENAI_API_BASE_URL: xxx
      # API model, optional, available when OPENAI_API_KEY is set, https://platform.openai.com/docs/models
      # gpt-4, gpt-4-0314, gpt-4-32k, gpt-4-32k-0314, gpt-3.5-turbo, gpt-3.5-turbo-0301, text-davinci-003, text-davinci-002, code-davinci-002
      OPENAI_API_MODEL: xxx
      # reverse proxy, optional
      API_REVERSE_PROXY: xxx
      # access key, optional
      AUTH_SECRET_KEY: xxx
      # The maximum number of requests per hour, optional, default unlimited
      MAX_REQUEST_PER_HOUR: 0
      # Timeout in milliseconds, optional
      TIMEOUT_MS: 60000
      # Socks proxy, optional, works with SOCKS_PROXY_PORT
      SOCKS_PROXY_HOST: xxx
      # Socks proxy port, optional, effective when combined with SOCKS_PROXY_HOST
      SOCKS_PROXY_PORT: xxx
      # HTTPS proxy, optional, supports http, https, socks5
      HTTPS_PROXY: https://xxx:7890

• OPENAI_API_BASE_URL optional, available when OPENAI_API_KEY is set
• OPENAI_API_MODEL optional, available when OPENAI_API_KEY is set

Note: Railway modifying environment variables will restart Deploy

If you do not need the node interface of this project, you can omit the following operations

Copy the service folder to the server where you have the node service environment.

# Install
pnpm install

# Pack
pnpm build

# run
pnpm prod

PS: You can run pnpm start directly on the server without packaging

1. Modify VITE_GLOB_API_URL in the .env file in the root directory to your actual backend interface address

2. Run the following command in the root directory, and then copy the files in the dist folder to the root directory of your website service

Reference Information

pnpm build

Q: Why do Git commits always report errors?

A: Because there is commit information verification, please follow the Commit Guide

Q: If only the front-end page is used, where should I change the request interface?

A: The VITE_GLOB_API_URL field in the .env file in the root directory.

Q: When the file is saved, it all explodes?

A: vscode Please install the project recommended plugin, or manually install the Eslint plugin.

Q: Is there no typewriter effect on the front end?

A: One possible reason is that after the Nginx reverse proxy is enabled and the buffer is enabled, Nginx will try to buffer a certain size of data from the backend and then send it to the browser. Please try to add proxy_buffering off; after the reverse parameter, and then reload Nginx. The same goes for other web server configurations.