Uses the OpenAI-compatible chat API. Set model to any model your endpoint supports (e.g. gpt-5.5, gpt-4.1), or point api_base to a compatible gateway to use other providers. See the official docs for details.

Install dependencies

pip3 install "openai<1.0.0"

Note: this project uses the legacy openai SDK (0.27.x+ but below 1.0.0) — requirements.txt already pins a compatible version. If installation fails, upgrade pip first with pip3 install --upgrade pip.

Configuration

{
  "model": {
    "type" : "chatgpt",
    "openai": {
      "api_key": "YOUR API KEY",
      "api_base": "",                                   # optional, an OpenAI-compatible endpoint
      "model": "gpt-5.5",                               # model name
      "proxy": "http://127.0.0.1:7890",                 # proxy address
      "character_desc": "You are ChatGPT, a large language model trained by OpenAI...",
      "conversation_max_tokens": 1000,                  # max reply length, total of input and output
      "temperature":0.75,     # entropy in [0,1]; higher means more random word choices
      "top_p":0.7,            # candidate word list; 0.7 means only the top 70% of candidates are considered
      "frequency_penalty":0.0,            # in [-2,2]; higher reduces word repetition in a line
      "presence_penalty":1.0,             # in [-2,2]; higher is less constrained by the input
    }
}

• api_key: the OpenAI API KEY created when you registered your account
• api_base (optional): an OpenAI-compatible endpoint; leave empty for the official API, or point it to a compatible gateway to use other providers
• model: any model your endpoint supports, e.g. gpt-5.5, gpt-4.1, gpt-4o (GPT-5 / o-series only accept default sampling params; this project skips them automatically)
• proxy: proxy client address, see #56
• character_desc: the bot's persona; the model plays this role, feel free to customize it
• max_history_num (optional): max length of conversation memory; older memory is cleared beyond this

Configuration

{
  "model": {
    "type" : "linkai",
    "linkai": {
      "api_key": "",
      "api_base": "https://api.link-ai.tech",
      "app_code":  "",
      "model": "",
      "conversation_max_tokens": 1000,
      "temperature":0.75,
      "top_p":0.7,
      "frequency_penalty":0.0,
      "presence_penalty":1.0,
      "character_desc": "You are an intelligent assistant."
    },
}

• api_key: the key for calling LinkAI, created in the console
• app_code: the code of a LinkAI app or workflow, optional, see Creating an App
• model: one key gives access to 100+ models (DeepSeek, Claude, Gemini, Qwen, GLM, GPT, etc.), see the model list; can be left empty and set the app's default model on the LinkAI platform
• Other parameters have the same meaning as in the OpenAI model

Based on Baidu ERNIE Bot's web version, needs a Cookie obtained manually.

{
  "model": {
    "type" : "baidu",
    "baidu": {
      "acs_token": "YOUR ACS TOKEN",
      "cookie": "YOUR COOKIE"
    }
  }
}

• cookie: after logging into ERNIE Bot in the browser, grab it from the developer tools
• acs_token: same as above, grab it from the request parameters; search for a tutorial if needed

Based on Bing chat, depends on the EdgeGPT library, needs a Cookie after logging into Bing.

{
  "model": {
    "type" : "bing",
    "bing":{
      "jailbreak": true,
      "jailbreak_prompt": "...",
      "cookies": []
    }
  }
}

• cookies: the array of cookies exported from the browser after logging into Bing
• jailbreak: whether to enable jailbreak (Sydney) mode, which bypasses some of the official restrictions

Based on Google Bard's web version, needs a Cookie after logging in.

{
  "model": {
    "type" : "bard",
    "bard": {
      "cookie": "YOUR COOKIE"
    }
  }
}

The config template starts the terminal by default — no extra config needed. Run python3 app.py in the project directory to start it. Type right in the terminal to chat with the model, with streaming output supported.

Contributor: RegimenArsenic

Dependencies

pip3 install PyJWT flask flask_socketio

Configuration

"channel": {
    "type": "http",
    "http": {
      "http_auth_secret_key": "6d25a684-9558-11e9-aa94-efccd7a0659b",    // JWT auth secret key
      "http_auth_password": "6.67428e-11",        // auth password, for personal use, a basic defense against port scanning and DDOS wasting tokens
      "port": "80"       // port
    }
  }

Run locally: after python3 app.py, visit http://127.0.0.1:80.

Run on a server: after deploying, visit http://your-domain-or-IP:port.

Requirements: a server and a subscription account.

1. Install dependencies

Install werobot:

pip3 install werobot

2. Configuration

"channel": {
    "type": "wechat_mp",
    "wechat_mp": {
      "token": "YOUR TOKEN",           # token value
      "port": "8088"                   # port the program listens on
    }
}

3. Run the program

Run python3 app.py in the project directory. If the terminal shows the following, it started successfully:

[INFO][2023-02-16 01:39:53][app.py:12] - [INIT] load config: ...
[INFO][2023-02-16 01:39:53][wechat_mp_channel.py:25] - [WX_Public] Wechat Public account service start!
Bottle v0.12.23 server starting up (using AutoServer())...
Listening on http://127.0.0.1:8088/
Hit Ctrl-C to quit.

4. Set the callback URL

Go to your subscription account in the WeChat Official Platform and enable server configuration:

Server address (URL): if you can reach the program on your server through this URL in a browser (default port 8088), the config is valid. Since subscription accounts only allow ports 80/443, either make the program listen on port 80 directly (needs sudo) or forward it with a reverse proxy like nginx. A public IP or a domain both work here.

Token: must match the token in config.json.

For the detailed process, see the official docs.

Note: after a user sends a message, WeChat pushes it to the configured URL, but if there's no reply within 5 seconds it disconnects and retries 3 times, while model requests often take longer than 5s. This project uses async and caching to stretch the limit to 15s, but beyond that it still can't reply in time. For time-sensitive scenarios, use the "Service Account" instead.

Requirements: a server and a WeChat-verified service account.

The service account calls the model asynchronously first, then pushes the result to the user via the customer-service API, which gets around the subscription account's 15s timeout. Its developer-mode config is similar to the subscription account's — see the official docs.

In the config, just change type to wechat_mp_service, keep reusing the wechat_mp block, and add app_id and app_secret:

"channel": {
    "type": "wechat_mp_service",
    "wechat_mp": {
      "token": "YOUR TOKEN",            # token value
      "port": "8088",                   # port the program listens on
      "app_id": "YOUR APP ID",          # app ID
      "app_secret": "YOUR APP SECRET"   # app secret
    }
}

Note: add the server IP to the "IP whitelist", otherwise users won't receive pushed messages.

Requirements: a PC or server (mainland China network) and a QQ account.

Running a QQ bot also needs a separate go-cqhttp process, which handles sending and receiving QQ messages, while this project requests the model and generates replies.

1. Download go-cqhttp

Download the build for your system from the go-cqhttp Release, unzip it, and put the go-cqhttp binary in bot-on-anything/channel/qq. There's already a config.yml there — just fill in your QQ account (account-uin).

2. Install aiocqhttp

Use aiocqhttp to talk to go-cqhttp:

pip3 install aiocqhttp

3. Configuration

Just change the channel type in config.json to qq:

"channel": {
    "type": "qq"
}

4. Running

Terminal 1, in the project root (listens on port 8080):

python3 app.py

Terminal 2, in the go-cqhttp directory:

cd channel/qq
./go-cqhttp

Note: there's no keyword matching or group whitelist yet, so all private chats get auto-replies, and in group chats it replies whenever it's @-mentioned. If you hit issues like a frozen account, change protocol in device.json from 5 to 2, see this Issue.

Contributor: brucelt1993

1. Get the token

You can search for how to create a Telegram bot — the key thing is getting the bot's token id.

2. Install dependencies

pip install pyTelegramBotAPI

3. Configuration

"channel": {
    "type": "telegram",
    "telegram":{
      "bot_token": "YOUR BOT TOKEN ID"
    }
}

Requirements: a server and a Gmail account.

Contributor: Simon

Follow the official docs to create an APP password for your Google account, then configure it like below:

"channel": {
    "type": "gmail",
    "gmail": {
      "subject_keyword": ["bot", "@bot"],
      "host_email": "xxxx@gmail.com",
      "host_password": "GMAIL ACCESS KEY"
    }
  }

❉ No longer needs a server or public IP

Contributor: amaoo

Dependencies

pip3 install slack_bolt

Configuration

"channel": {
    "type": "slack",
    "slack": {
      "slack_bot_token": "xoxb-xxxx",
      "slack_app_token": "xapp-xxxx"
    }
  }

Set bot token scopes - OAuth & Permission:

app_mentions:read
chat:write

Enable Socket Mode - Socket Mode: if you don't have an app-level token yet, you'll be prompted to create one; put it in slack_app_token.

Event Subscriptions - Subscribe to bot events:

app_mention

Reference: Slack Bolt for Python

Requirements: an enterprise internal development bot.

Dependencies

pip3 install requests flask

Configuration

"channel": {
    "type": "dingtalk",
    "dingtalk": {
      "image_create_prefix": ["draw", "draw", "Draw"],
      "port": "8081",                  # external port
      "dingtalk_token": "xx",          # access_token of the webhook URL
      "dingtalk_post_token": "xx",     # verification token in the header when DingTalk posts back
      "dingtalk_secret": "xx"          # security signing secret for the group bot
    }
  }

Create the bot

At https://open-dev.dingtalk.com/fe/app#/corp/robot , add a bot, then in the development settings fill in the server's outbound IP (run curl ifconfig.me on the host to get it) and the message-receiving address (the external address in your config, e.g. https://xx.xx.com:8081).

Reference: DingTalk internal bot tutorial · Enterprise internal bot tutorial

Dependencies

pip3 install requests flask

Configuration

"channel": {
    "type": "feishu",
    "feishu": {
        "image_create_prefix": ["draw", "draw", "Draw"],
        "port": "8082",                  # external port
        "app_id": "xxx",                 # app_id
        "app_secret": "xxx",             # app secret
        "verification_token": "xxx"      # event subscription verification token
    }
}

Create the bot

At https://open.feishu.cn/app/ :

1. Add an enterprise self-built app
2. Grant permissions: im:message, im:message.group_at_msg, im:message.group_at_msg:readonly, im:message.p2p_msg, im:message.p2p_msg:readonly, im:message:send_as_bot
3. In the event subscription menu, add the event (Receive messages v2.0) and set the request URL (the external address in your config, e.g. https://xx.xx.com:8081)
4. Publish the app in version management; once approved, add the self-built app to your group

Requirements: a server and a verified Enterprise WeChat.

Just change type in config.json to wechat_com; the default message-receiving URL is http://ip:8888/wechat:

"channel": {
    "type": "wechat_com",
    "wechat_com": {
      "wechat_token": "YOUR TOKEN",            # token value
      "port": "8888",                          # port the program listens on
      "app_id": "YOUR APP ID",                 # app ID
      "app_secret": "YOUR APP SECRET",         # app secret
      "wechat_corp_id": "YOUR CORP ID",
      "wechat_encoding_aes_key": "YOUR AES KEY"
    }
}

Note: add the server IP to the "Enterprise trusted IP" list, otherwise users won't receive pushed messages.

Reference: Enterprise WeChat setup tutorial

Depends on discord.py:

pip3 install "discord.py>=2.0.0"

Configuration

"channel": {
    "type": "discord",
    "discord": {
        "app_token": "xxx",
        "channel_name": "xxx",
        "channel_session": "xxx"
    }
}

• app_token: the Discord bot's Bot Token
• channel_name: restrict the bot to a specific channel; leave empty to listen to all channels
• channel_session: session granularity, author (per user) or thread (per thread)