import re import threading import asyncio import time import util.unfit_words as uw import os import sys import traceback import util.agent.web_searcher as web_searcher import util.plugin_util as putil from nakuru.entities.components import Plain, At, Image from addons.baidu_aip_judge import BaiduJudge from model.provider.provider import Provider from model.command.command import Command from util import general_utils as gu from util.general_utils import upload, run_monitor from util.cmd_config import CmdConfig as cc from util.cmd_config import init_astrbot_config_items from type.types import GlobalObject from type.register import * from type.message import AstrBotMessage from type.config import * from addons.dashboard.helper import DashBoardHelper from addons.dashboard.server import DashBoardData from persist.session import dbConn from model.platform._message_result import MessageResult from SparkleLogging.utils.core import LogManager from logging import Logger logger: Logger = LogManager.GetLogger(log_name='astrbot-core') # 用户发言频率 user_frequency = {} # 时间默认值 frequency_time = 60 # 计数默认值 frequency_count = 10 # 语言模型 OPENAI_OFFICIAL = 'openai_official' NONE_LLM = 'none_llm' chosen_provider = None # 语言模型对象 llm_instance: dict[str, Provider] = {} llm_command_instance: dict[str, Command] = {} llm_wake_prefix = "" # 百度内容审核实例 baidu_judge = None # CLI PLATFORM_CLI = 'cli' # 全局对象 _global_object: GlobalObject = None def privider_chooser(cfg): l = [] if 'openai' in cfg and len(cfg['openai']['key']) > 0 and cfg['openai']['key'][0] is not None: l.append('openai_official') return l def init(): ''' 初始化机器人 ''' global llm_instance, llm_command_instance global baidu_judge, chosen_provider global frequency_count, frequency_time global _global_object init_astrbot_config_items() cfg = cc.get_all() _event_loop = asyncio.new_event_loop() asyncio.set_event_loop(_event_loop) # 初始化 global_object _global_object = GlobalObject() _global_object.version = VERSION _global_object.base_config = cfg _global_object.logger = logger logger.info("AstrBot v" + VERSION) if 'reply_prefix' in cfg: # 适配旧版配置 if isinstance(cfg['reply_prefix'], dict): _global_object.reply_prefix = "" cfg['reply_prefix'] = "" cc.put("reply_prefix", "") else: _global_object.reply_prefix = cfg['reply_prefix'] default_personality_str = cc.get("default_personality_str", "") if default_personality_str == "": _global_object.default_personality = None else: _global_object.default_personality = { "name": "default", "prompt": default_personality_str, } # 语言模型提供商 logger.info("正在载入语言模型...") prov = privider_chooser(cfg) if OPENAI_OFFICIAL in prov: logger.info("初始化：OpenAI官方") if cfg['openai']['key'] is not None and cfg['openai']['key'] != [None]: from model.provider.openai_official import ProviderOpenAIOfficial from model.command.openai_official import CommandOpenAIOfficial llm_instance[OPENAI_OFFICIAL] = ProviderOpenAIOfficial( cfg['openai']) llm_command_instance[OPENAI_OFFICIAL] = CommandOpenAIOfficial( llm_instance[OPENAI_OFFICIAL], _global_object) _global_object.llms.append(RegisteredLLM( llm_name=OPENAI_OFFICIAL, llm_instance=llm_instance[OPENAI_OFFICIAL], origin="internal")) chosen_provider = OPENAI_OFFICIAL instance = llm_instance[OPENAI_OFFICIAL] assert isinstance(instance, ProviderOpenAIOfficial) instance.DEFAULT_PERSONALITY = _global_object.default_personality instance.curr_personality = instance.DEFAULT_PERSONALITY # 检查provider设置偏好 p = cc.get("chosen_provider", None) if p is not None and p in llm_instance: chosen_provider = p # 百度内容审核 if 'baidu_aip' in cfg and 'enable' in cfg['baidu_aip'] and cfg['baidu_aip']['enable']: try: baidu_judge = BaiduJudge(cfg['baidu_aip']) logger.info("百度内容审核初始化成功") except BaseException as e: logger.info("百度内容审核初始化失败") threading.Thread(target=upload, args=( _global_object, ), daemon=True).start() # 得到发言频率配置 if 'limit' in cfg: if 'count' in cfg['limit']: frequency_count = cfg['limit']['count'] if 'time' in cfg['limit']: frequency_time = cfg['limit']['time'] try: if 'uniqueSessionMode' in cfg and cfg['uniqueSessionMode']: _global_object.unique_session = True else: _global_object.unique_session = False except BaseException as e: logger.info("独立会话配置错误: "+str(e)) nick_qq = cc.get("nick_qq", None) if nick_qq == None: nick_qq = ("ai", "!", "！") if isinstance(nick_qq, str): nick_qq = (nick_qq,) if isinstance(nick_qq, list): nick_qq = tuple(nick_qq) _global_object.nick = nick_qq # 语言模型唤醒词 global llm_wake_prefix llm_wake_prefix = cc.get("llm_wake_prefix", "") logger.info("正在载入插件...") # 加载插件 _command = Command(None, _global_object) ok, err = putil.plugin_reload(_global_object) if ok: logger.info( f"成功载入 {len(_global_object.cached_plugins)} 个插件") else: logger.info(err) if chosen_provider is None: llm_command_instance[NONE_LLM] = _command chosen_provider = NONE_LLM logger.info("正在载入机器人消息平台") # logger.info("提示：需要添加管理员 ID 才能使用 update/plugin 等指令)，可在可视化面板添加。（如已添加可忽略）") platform_str = "" # GOCQ if 'gocqbot' in cfg and cfg['gocqbot']['enable']: logger.info("启用 QQ_GOCQ 机器人消息平台") threading.Thread(target=run_gocq_bot, args=( cfg, _global_object), daemon=True).start() platform_str += "QQ_GOCQ," # QQ频道 if 'qqbot' in cfg and cfg['qqbot']['enable'] and cfg['qqbot']['appid'] != None: logger.info("启用 QQ_OFFICIAL 机器人消息平台") threading.Thread(target=run_qqchan_bot, args=( cfg, _global_object), daemon=True).start() platform_str += "QQ_OFFICIAL," # 初始化dashboard _global_object.dashboard_data = DashBoardData( stats={}, configs={}, logs={}, plugins=_global_object.cached_plugins, ) dashboard_helper = DashBoardHelper(_global_object, config=cc.get_all()) dashboard_thread = threading.Thread( target=dashboard_helper.run, daemon=True) dashboard_thread.start() # 运行 monitor threading.Thread(target=run_monitor, args=( _global_object,), daemon=True).start() logger.info( "如果有任何问题, 请在 https://github.com/Soulter/AstrBot 上提交 issue 或加群 322154837。") logger.info("请给 https://github.com/Soulter/AstrBot 点个 star。") if platform_str == '': platform_str = "(未启动任何平台，请前往面板添加)" logger.info(f"🎉 项目启动完成") dashboard_thread.join() ''' 运行 QQ_OFFICIAL 机器人 ''' def run_qqchan_bot(cfg: dict, global_object: GlobalObject): try: from model.platform.qq_official import QQOfficial qqchannel_bot = QQOfficial( cfg=cfg, message_handler=oper_msg, global_object=global_object) global_object.platforms.append(RegisteredPlatform( platform_name="qqchan", platform_instance=qqchannel_bot, origin="internal")) qqchannel_bot.run() except BaseException as e: logger.error("启动 QQ 频道机器人时出现错误, 原因如下: " + str(e)) logger.error(r"如果您是初次启动，请前往可视化面板填写配置。详情请看：https://astrbot.soulter.top/center/。") ''' 运行 QQ_GOCQ 机器人 ''' def run_gocq_bot(cfg: dict, _global_object: GlobalObject): from model.platform.qq_gocq import QQGOCQ noticed = False host = cc.get("gocq_host", "127.0.0.1") port = cc.get("gocq_websocket_port", 6700) http_port = cc.get("gocq_http_port", 5700) logger.info( f"正在检查连接...host: {host}, ws port: {port}, http port: {http_port}") while True: if not gu.port_checker(port=port, host=host) or not gu.port_checker(port=http_port, host=host): if not noticed: noticed = True logger.warning( f"连接到{host}:{port}（或{http_port}）失败。程序会每隔 5s 自动重试。") time.sleep(5) else: logger.info("已连接到 gocq。") break try: qq_gocq = QQGOCQ(cfg=cfg, message_handler=oper_msg, global_object=_global_object) _global_object.platforms.append(RegisteredPlatform( platform_name="gocq", platform_instance=qq_gocq, origin="internal")) qq_gocq.run() except BaseException as e: input("启动QQ机器人出现错误"+str(e)) ''' 检查发言频率 ''' def check_frequency(id) -> bool: ts = int(time.time()) if id in user_frequency: if ts-user_frequency[id]['time'] > frequency_time: user_frequency[id]['time'] = ts user_frequency[id]['count'] = 1 return True else: if user_frequency[id]['count'] >= frequency_count: return False else: user_frequency[id]['count'] += 1 return True else: t = {'time': ts, 'count': 1} user_frequency[id] = t return True async def record_message(platform: str, session_id: str): # TODO: 这里会非常吃资源。然而 sqlite3 不支持多线程，所以暂时这样写。 curr_ts = int(time.time()) db_inst = dbConn() db_inst.increment_stat_session(platform, session_id, 1) db_inst.increment_stat_message(curr_ts, 1) db_inst.increment_stat_platform(curr_ts, platform, 1) async def oper_msg(message: AstrBotMessage, session_id: str, role: str = 'member', platform: str = None, ) -> MessageResult: """ 处理消息。 message: 消息对象 session_id: 该消息源的唯一识别号 role: member | admin platform: str 所注册的平台的名称。如果没有注册，将抛出一个异常。 """ global chosen_provider, _global_object message_str = '' session_id = session_id role = role hit = False # 是否命中指令 command_result = () # 调用指令返回的结果 # 获取平台实例 reg_platform: RegisteredPlatform = None for p in _global_object.platforms: if p.platform_name == platform: reg_platform = p break if not reg_platform: raise Exception(f"未找到平台 {platform} 的实例。") # 统计数据，如频道消息量 await record_message(platform, session_id) for i in message.message: if isinstance(i, Plain): message_str += i.text.strip() if message_str == "": return MessageResult("Hi~") # 检查发言频率 if not check_frequency(message.sender.user_id): return MessageResult(f'你的发言超过频率限制(╯▔皿▔)╯。\n管理员设置{frequency_time}秒内只能提问{frequency_count}次。') # 检查是否是更换语言模型的请求 temp_switch = "" if message_str.startswith('/gpt'): target = chosen_provider if message_str.startswith('/gpt'): target = OPENAI_OFFICIAL l = message_str.split(' ') if len(l) > 1 and l[1] != "": # 临时对话模式，先记录下之前的语言模型，回答完毕后再切回 temp_switch = chosen_provider chosen_provider = target message_str = l[1] else: chosen_provider = target cc.put("chosen_provider", chosen_provider) return MessageResult(f"已切换至【{chosen_provider}】") llm_result_str = "" # check commands and plugins message_str_no_wake_prefix = message_str for wake_prefix in _global_object.nick: # nick: tuple if message_str.startswith(wake_prefix): message_str_no_wake_prefix = message_str.removeprefix(wake_prefix) break hit, command_result = await llm_command_instance[chosen_provider].check_command( message_str_no_wake_prefix, session_id, role, reg_platform, message, ) # 没触发指令 if not hit: # 关键词拦截 for i in uw.unfit_words_q: matches = re.match(i, message_str.strip(), re.I | re.M) if matches: return MessageResult(f"你的提问得到的回复未通过【默认关键词拦截】服务, 不予回复。") if baidu_judge != None: check, msg = await asyncio.to_thread(baidu_judge.judge, message_str) if not check: return MessageResult(f"你的提问得到的回复未通过【百度AI内容审核】服务, 不予回复。\n\n{msg}") if chosen_provider == NONE_LLM: logger.info("一条消息由于 Bot 未启动任何语言模型并且未触发指令而将被忽略。") return try: if llm_wake_prefix != "" and not message_str.startswith(llm_wake_prefix): return # check image url image_url = None for comp in message.message: if isinstance(comp, Image): if comp.url is None: image_url = comp.file break else: image_url = comp.url break # web search keyword web_sch_flag = False if message_str.startswith("ws ") and message_str != "ws ": message_str = message_str[3:] web_sch_flag = True else: message_str += " " + cc.get("llm_env_prompt", "") if chosen_provider == OPENAI_OFFICIAL: if _global_object.web_search or web_sch_flag: official_fc = chosen_provider == OPENAI_OFFICIAL llm_result_str = await web_searcher.web_search(message_str, llm_instance[chosen_provider], session_id, official_fc) else: llm_result_str = await llm_instance[chosen_provider].text_chat(message_str, session_id, image_url) llm_result_str = _global_object.reply_prefix + llm_result_str except BaseException as e: logger.error(f"调用异常：{traceback.format_exc()}") return MessageResult(f"调用异常。详细原因：{str(e)}") # 切换回原来的语言模型 if temp_switch != "": chosen_provider = temp_switch if hit: # 有指令或者插件触发 # command_result 是一个元组：(指令调用是否成功, 指令返回的文本结果, 指令类型) if command_result == None: return command = command_result[2] if not command_result[0]: return MessageResult(f"指令调用错误: \n{str(command_result[1])}") # 画图指令 if command == 'draw': # 保存到本地 path = await gu.download_image_by_url(command_result[1]) return MessageResult([Image.fromFileSystem(path)]) # 其他指令 else: try: return MessageResult(command_result[1]) except BaseException as e: return MessageResult(f"回复消息出错: {str(e)}") return # 敏感过滤 # 过滤不合适的词 for i in uw.unfit_words: llm_result_str = re.sub(i, "***", llm_result_str) # 百度内容审核服务二次审核 if baidu_judge != None: check, msg = await asyncio.to_thread(baidu_judge.judge, llm_result_str) if not check: return MessageResult(f"你的提问得到的回复【百度内容审核】未通过，不予回复。\n\n{msg}") # 发送信息 try: return MessageResult(llm_result_str) except BaseException as e: logger.info("回复消息错误: \n"+str(e))