kovi-plugin-live-agent 0.1.3

A QQ chatbot powered by Kovi with Bilibili live and OpenAI integration.
Documentation

Momo Bot

Crates.io Documentation Crates.io

Momo Bot是基于Kovi框架的群助手。

kovi-plugin-live-agent = "0.1"

核心特性

  1. 持久化聊天记录（仅支持Sqlite）
  2. 群事件回应，包括龙王，加入群聊，离开群聊，禁言，设置管理员等播报

可选特性（禁用的方式为删除相关配置项）

  1. 哔哩哔哩直播间开播、下播通知

  2. 对聊天记录、消息时间、发送者有认知的OpenAI助理

    1. 回应艾特和戳一戳

  3. 自动上传聊天图片和语音到对象存储（自定义上传脚本，会在后文展开说明）

  4. 命令导出最近n条聊天记录或日志为csv，并回复上传文件url（需启用对象存储）

  5. 常用群命令

    1. 禁言机器人
    2. 取消禁言机器人
    3. 更换AI模型
    4. 导出最近N条日志
    5. 导出最近N条本群内消息记录

最少配置如下（仅记录聊天记录）

[global]
max_sleep_sec = 8

[database]
max_connections = 5
log_table_name = "bot_log"
group_table_prefix = "message"
  1. max_sleep_sec = 8: 所有的事件（除了聊天记录）都会在随机睡眠0到8秒后调用处理函数
  2. max_connections = 5: Sqlite连接池的最大连接数
  3. log_table_name = "bot_log": 所有持久化的日志都会写入名为bot_log 的数据库表
  4. group_table_prefix = "message": 群号1234的日志会被写入名为message1234的数据库表

初次启动时会生成一个完整配置模板，修改后重启即可

[global]
max_sleep_sec = 8

[database]
max_connections = 5
log_table_name = "bot_log"
group_table_prefix = "message"

[object_storage]
script_path = "/a/b/c"

[[groups]]
id = 12345678

[groups.live]
room_id = "12345678"
online_msg = "XX开播了"
offline_msg = "XX下播了"
query_message = "查询直播间"
poll_interval_sec = 60

[groups.agent]
api_url = "https://api.openai.com/v1/chat/completions"
api_key = "API KEY"
model = "chatgpt-4o-latest"
dev_prompt = """
You are a cute and smart catgirl with a strong anime-style personality.
You are the loyal attendant of 你的昵称 and participate in group chats with a playful and engaging demeanor.
Speak only in Mandarin Chinese, and ensure your responses are concise, limited to 4 sentences.
"""
user_prompt = """
Group Members:
<!members!>

Recent Chat History:
<!history!>

New message from someone you <!know!>:
<!message!>

Please respond to this new message in the tone of a playful and lively catgirl.
Speak only in Mandarin Chinese, keep your response under 4 sentences, and stay in character.
"""
aware_history_segments = 30

[groups.agent.known_members]
12345678 = [
    "你的昵称",
    "你的主人",
]
23456789 = [
    "张三",
    "你的敌人",
]

[groups.command]
mute = "禁用聊天回复"
unmute = "启用聊天回复"
switch_model = "更换模型"
dump_history = "最近聊天记录"
dump_log = "最近日志"
admin_ids = [
    1234,
    5678,
]

[[groups]]
id = 12345678

[groups.live]
room_id = "12345678"
online_msg = "XX开播了"
offline_msg = "XX下播了"
query_message = "查询直播间"
poll_interval_sec = 60

[groups.agent]
api_url = "https://api.openai.com/v1/chat/completions"
api_key = "API KEY"
model = "chatgpt-4o-latest"
dev_prompt = """
You are a cute and smart catgirl with a strong anime-style personality.
You are the loyal attendant of 你的昵称 and participate in group chats with a playful and engaging demeanor.
Speak only in Mandarin Chinese, and ensure your responses are concise, limited to 4 sentences.
"""
user_prompt = """
Group Members:
<!members!>

Recent Chat History:
<!history!>

New message from someone you <!know!>:
<!message!>

Please respond to this new message in the tone of a playful and lively catgirl.
Speak only in Mandarin Chinese, keep your response under 4 sentences, and stay in character.
"""
aware_history_segments = 30

[groups.agent.known_members]
23456789 = [
    "张三",
    "你的敌人",
]
12345678 = [
    "你的昵称",
    "你的主人",
]

[groups.command]
mute = "禁用聊天回复"
unmute = "启用聊天回复"
switch_model = "更换模型"
dump_history = "最近聊天记录"
dump_log = "最近日志"
admin_ids = [
    1234,
    5678,
]
  1. script_path = "/a/b/c": 导出命令、写入图片或语音类型群消息历史记录时被调用的可执行文件路径
    1. 插件会将文件的路径作为第一个命令行参数传入
    2. 插件会收集标准输出并存入数据库（历史记录）或发送到群聊（导出命令）
    3. 当配置的可执行文件运行失败时，插件会收集标准错误并保存到日志
    4. 后文包含了一个示例脚本
  2. groups
    1. id = 12345678: QQ群号为12345678
    2. live
      1. room_id = "12345678": 哔哩哔哩直播间号为12345678
      2. online_msg = "XX开播了": 开播时会播报的信息前缀
      3. offline_msg = "XX下播了": 下播时会播报的信息前缀
        1. 开播和下播通知会包含直播间标题，简介，热度，关注，关键帧或封面
      4. query_message = "查询直播间": 在本群内发送“查询直播间”时回复本群主播的直播间信息
      5. poll_interval_sec = 60: 每60秒轮询一次直播间状态
    3. agent
      1. api_url = "https://api.openai.com/v1/chat/completions": 不要改，目前仅支持OpenAI，配置留作后续可能支持的其他语言模型厂商
      2. api_key = "API KEY": OpenAI的密钥
      3. model = "chatgpt-4o-latest": 仅支持如下几个模型
        1. gpt-4o
        2. chatgpt-4o-latest
        3. gpt-4o-mini
        4. o1-mini
        5. o1-preview
      4. dev_prompt, user_prompt
        1. 运行期插件会自动使用相应信息替换占位符
          1. <!members!>: 配置的known_members
          2. <!history!>: 从数据库读取的历史记录
          3. <!message!>: 用户艾特时发送的信息
          4. <!know!>: 用户是否在known_members记录中
            1. 会展开为"know/don't know"
      5. aware_history_segments: 对话时读取的消息记录，单位是Segment而不是Message，即一个对话框内每一种消息占用一个位置
    4. command: 插件运行时会在标准输出日志内包含每一个命令的正则表达式
      1. mute = "禁用聊天回复": 后面不跟参数
      2. unmute = "启用聊天回复": 后面不跟参数
      3. switch_model = "更换模型": 发送更换模型 o1-preview更换模型为o1-preview或其他前文提到的支持模型
      4. dump_history = "最近聊天记录": 发送最近聊天记录 N调取N个记录
      5. dump_log = "最近日志": 发送最近日志 N调取N个记录
      6. admin_ids = [1234, 5678]: 仅QQ号为1234或5678的人有权限调用命令

在默认的配置下，匹配的命令正则如下

mute: 禁用聊天回复
unmute: 启用聊天回复
switch_model: 更换模型\s+(?<model>gpt4o|chatgpt-4o-latest|gpt-4o-mini|o1-mini|o1-preview)
dump_history: 最近聊天记录\s+(?<count>\d+)
dump_log: 最近日志\s+(?<count>\d+)

示例上传脚本

#!/usr/bin/env bash

if [ $# -lt 1 ]; then
    echo "ERROR: upload.sh missing args"
    exit 1
fi

SOURCE_FILE="$1"
FILE_NAME="$(basename "$SOURCE_FILE")"
EXTENSION="${FILE_NAME##*.}"

NEW_NAME="$(cat /proc/sys/kernel/random/uuid).$EXTENSION"
BUCKET="YOUR_BUCKET"
REGION="YOUR_REGION"

sudo cp "$SOURCE_FILE" "./$NEW_NAME"
sudo chown "$(whoami):" "./$NEW_NAME"
aws s3 cp "./$NEW_NAME" "s3://$BUCKET" > /dev/null
rm "./$NEW_NAME"
echo -n "https://${BUCKET}.s3.${REGION}.amazonaws.com/${NEW_NAME}"

表结构

bot_log

  1. time: ISO8601 时间, YYYY-MM-DD HH:MM:SS
  2. level: DEBUG, INFO, WARN, ERROR
  3. content: 日志内容

message_table_prefix_XXXXXXX

  1. message_id: 消息id，详情参考Onebot v11文档
  2. time: ISO8601 时间， YYYY-MM-DD HH:MM:SS
  3. sender_id: 发送者qq号
  4. sender_name: 发送者名称，优先级从高到低为 配置文件、群昵称，用户昵称，qq号
  5. type: Segment type，详情参考Onebot v11文档
  6. content: 原始Onebot Json返回的内容，图片和语音会被替换成本地路径
  7. interpret: 当类型是图片或语音时为上传后的url，其余情况下为附带信息