commit 47ca0639dafeab230f3e0c637c2e3b3fff530538
Author: heimoshuiyu <heimoshuiyu@gmail.com>
Date:   Sat Oct 21 21:14:57 2023 +0800

    init

diff --git a/.dockerignore b/.dockerignore
new file mode 100644
index 0000000..d5e58ed
--- /dev/null
+++ b/.dockerignore
@@ -0,0 +1,4 @@
+__pycache__
+/venv
+/.env
+.git
diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..d5e58ed
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,4 @@
+__pycache__
+/venv
+/.env
+.git
diff --git a/Dockerfile b/Dockerfile
new file mode 100644
index 0000000..0716aff
--- /dev/null
+++ b/Dockerfile
@@ -0,0 +1,19 @@
+FROM python:3.11-slim-buster
+
+# Update the system
+RUN apt-get update -y
+
+# Install LibreOffice, Firefox ESR and pip
+RUN apt-get install -y libreoffice firefox-esr
+
+# Set the working directory in the container to /app
+WORKDIR /app
+
+COPY ./requirements_version.txt /app/requirements_version.txt
+RUN pip3 install --no-cache-dir -r requirements_version.txt
+
+# cache tiktoken dict
+RUN python3 -c 'import tiktoken; enc = tiktoken.get_encoding("cl100k_base")'
+
+# Add the current directory contents into the container at /app
+COPY . /app
diff --git a/bot.py b/bot.py
new file mode 100644
index 0000000..f69de6e
--- /dev/null
+++ b/bot.py
@@ -0,0 +1,315 @@
+import dotenv
+
+dotenv.load_dotenv()
+import os
+import traceback
+from functools import wraps
+from nio import AsyncClient, MatrixRoom, RoomMessageText, SyncError, InviteEvent
+import tiktoken
+import databases
+import builtins
+import sys
+
+
+def print(*args, **kwargs):
+    kwargs["file"] = sys.stderr
+    builtins.print(*args, **kwargs)
+
+
+class Bot(AsyncClient):
+    def __init__(self, homeserver: str, user: str, device_id: str, access_token: str):
+        super().__init__(homeserver)
+        self.access_token = access_token
+        self.user_id = self.user = user
+        self.device_id = device_id
+        self.welcome_message = ""
+        self._joined_rooms = []
+        self.db = databases.Database(os.environ["MATRIX_CHAIN_DB"])
+
+        self.enc = tiktoken.encoding_for_model("gpt-4")
+
+        # auto join
+        self.add_event_callback(self.auto_join, InviteEvent)
+
+    async def init_db(self):
+        db = self.db
+        await db.execute("CREATE EXTENSION IF NOT EXISTS vector")
+        await db.execute(
+            """
+        CREATE TABLE IF NOT EXISTS documents
+        (
+            md5 character(32) NOT NULL PRIMARY KEY,
+            content text,
+            token integer,
+            url text
+        )
+        """
+        )
+        await db.execute(
+            """
+        CREATE TABLE IF NOT EXISTS embeddings
+        (
+            document_md5 character(32) NOT NULL,
+            md5 character(32) NOT NULL,
+            content text NOT NULL,
+            token integer NOT NULL,
+            embedding vector(1536) NOT NULL,
+            PRIMARY KEY (document_md5, md5),
+            FOREIGN KEY (document_md5) REFERENCES documents(md5)
+        );
+        """
+        )
+        await db.execute(
+            """
+        CREATE TABLE IF NOT EXISTS event_document
+        (
+            event text NOT NULL PRIMARY KEY,
+            document_md5 character(32) NOT NULL,
+            FOREIGN KEY (document_md5)
+                REFERENCES documents (md5)
+        );
+        """
+        )
+        await db.execute(
+            """
+        CREATE TABLE IF NOT EXISTS memories
+        (
+            id SERIAL PRIMARY KEY,
+            root text NOT NULL,
+            role integer NOT NULL,
+            content text NOT NULL,
+            token integer NOT NULL
+        )
+            """
+        )
+        await db.execute(
+            """
+        CREATE TABLE IF NOT EXISTS room_configs
+        (
+            room text NOT NULL PRIMARY KEY,
+            model_name text,
+            temperature float NOT NULL DEFAULT 0,
+            system text,
+            embedding boolean NOT NULL DEFAULT false,
+            examples TEXT[] NOT NULL DEFAULT '{}'
+        )
+        """
+        )
+        await db.execute(
+            """
+        CREATE TABLE IF NOT EXISTS room_document (
+            room text NOT NULL,
+            document_md5 character(32) NOT NULL,
+            PRIMARY KEY (room, document_md5)
+        );
+            """
+        )
+
+    def get_token_length(self, text: str) -> int:
+        return len(self.enc.encode(text))
+
+    async def sync_forever(self):
+        # init
+        print("connecting to db")
+        await self.db.connect()
+
+        # init db hook
+        print("init db hook")
+        await self.init_db()
+
+        # remote callback to perform initial sync
+        callbacks = self.event_callbacks
+        self.event_callbacks = []
+        # skip intial sync
+        print("Perform initial sync")
+        resp = await self.sync(timeout=30000)
+        if isinstance(resp, SyncError):
+            raise BaseException(SyncError)
+        self.event_callbacks = callbacks
+        # set online
+        print("Set online status")
+        await self.set_presence("online")
+        # sync
+        print("Start forever sync")
+        return await super().sync_forever(300000, since=resp.next_batch)
+
+    async def auto_join(self, room: MatrixRoom, event: InviteEvent):
+        print("join", event.sender, room.room_id)
+        if room.room_id in self._joined_rooms:
+            return
+        await self.join(room.room_id)
+        self._joined_rooms.append(room.room_id)
+        if self.welcome_message:
+            await self.room_send(
+                room.room_id,
+                "m.room.message",
+                {
+                    "body": self.welcome_message,
+                    "msgtype": "m.text",
+                    "nogpt": True,
+                },
+            )
+
+    def ignore_self_message(self, func):
+        @wraps(func)
+        async def ret(room: MatrixRoom, event: RoomMessageText):
+            if event.sender == self.user:
+                return
+            return await func(room, event)
+
+        return ret
+
+    def log_message(self, func):
+        @wraps(func)
+        async def ret(room: MatrixRoom, event: RoomMessageText):
+            print(room.room_id, event.sender, event.body)
+            return await func(room, event)
+
+        return ret
+
+    def with_typing(self, func):
+        @wraps(func)
+        async def ret(room, *args, **kargs):
+            await self.room_typing(room.room_id, True, 60000 * 3)
+            resp = await func(room, *args, **kargs)
+            await self.room_typing(room.room_id, False)
+            return resp
+
+        return ret
+
+    def change_event_id_to_root_id(self, func):
+        @wraps(func)
+        async def ret(room, event, *args, **kargs):
+            root = event.event_id
+            if event.flattened().get("content.m.relates_to.rel_type") == "m.thread":
+                root = event.source["content"]["m.relates_to"]["event_id"]
+            event.event_id = root
+            return await func(room, event, *args, **kargs)
+
+        return ret
+
+    def ignore_not_mentioned(self, func):
+        @wraps(func)
+        async def ret(room, event, *args, **kargs):
+            flattened = event.flattened()
+            if not self.user in flattened.get(
+                "content.body", ""
+            ) and not self.user in flattened.get("content.formatted_body", ""):
+                return
+            return await func(room, event, *args, **kargs)
+
+        return ret
+
+    def replace_command_mark(self, func):
+        @wraps(func)
+        async def ret(room, event, *args, **kargs):
+            if event.body.startswith("！"):
+                event.body = "!" + event.body[1:]
+            return await func(room, event, *args, **kargs)
+
+        return ret
+
+    def handel_no_gpt(self, func):
+        @wraps(func)
+        async def ret(room, event, *args, **kargs):
+            if not event.flattened().get("content.nogpt") is None:
+                return
+            return await func(room, event, *args, **kargs)
+
+        return ret
+
+    def replace_reply_file_with_content(self, func):
+        @wraps(func)
+        async def ret(room, event, *args, **kargs):
+            flattened = event.flattened()
+            formatted_body = flattened.get("content.formatted_body", "")
+            if (
+                formatted_body.startswith("<mx-reply>")
+                and "</mx-reply>" in formatted_body
+            ):
+                print("replacing file content")
+                formatted_body = formatted_body[
+                    formatted_body.index("</mx-reply>") + len("</mx-reply>") :
+                ]
+                document_event_id = flattened.get(
+                    "content.m.relates_to.m.in_reply_to.event_id", ""
+                )
+                fetch = await self.db.fetch_all(
+                    query="""select d.content, d.token
+                    from documents d
+                    join event_document ed on d.md5 = ed.document_md5
+                    where ed.event = :event_id""",
+                    values={"event_id": document_event_id},
+                )
+                if len(fetch) > 0 and fetch[0][1] < 8192 + 4096:
+                    content = fetch[0][0]
+                    print(content)
+                    print("-----------")
+                    event.body = content + "\n\n---\n\n" + formatted_body
+                else:
+                    print("document not found or too large", event.event_id)
+
+            return await func(room, event, *args, **kargs)
+
+        return ret
+
+    def ignore_link(self, func):
+        @wraps(func)
+        async def ret(room, event, *args, **kargs):
+            if event.body.startswith("https://") or event.body.startswith("http://"):
+                return
+            return await func(room, event, *args, **kargs)
+
+        return ret
+
+    def safe_try(self, func):
+        @wraps(func)
+        async def ret(room, event, *args, **kargs):
+            try:
+                return await func(room, event, *args, **kargs)
+            except Exception as e:
+                print("--------------")
+                print("error:")
+                print(traceback.format_exc())
+                print("--------------")
+                await self.room_send(
+                    room.room_id,
+                    "m.reaction",
+                    {
+                        "m.relates_to": {
+                            "event_id": event.event_id,
+                            "key": "😵",
+                            "rel_type": "m.annotation",
+                        }
+                    },
+                )
+
+        return ret
+
+    def message_callback_common_wrapper(self, func):
+        @wraps(func)
+        @self.ignore_self_message
+        @self.handel_no_gpt
+        @self.log_message
+        @self.with_typing
+        @self.replace_reply_file_with_content
+        @self.change_event_id_to_root_id
+        @self.replace_command_mark
+        @self.safe_try
+        async def ret(*args, **kargs):
+            return await func(*args, **kargs)
+
+        return ret
+
+    async def react_ok(self, room_id: str, event_id: str):
+        await self.room_send(
+            room_id,
+            "m.reaction",
+            {
+                "m.relates_to": {
+                    "event_id": event_id,
+                    "key": "😘",
+                    "rel_type": "m.annotation",
+                }
+            },
+        )
diff --git a/bot_chatgpt.py b/bot_chatgpt.py
new file mode 100644
index 0000000..09f2e4f
--- /dev/null
+++ b/bot_chatgpt.py
@@ -0,0 +1,401 @@
+import os
+import dotenv
+
+dotenv.load_dotenv()
+import asyncio
+import jinja2
+import requests
+import datetime
+from langchain.schema import AIMessage, BaseMessage, HumanMessage, SystemMessage
+from nio import MatrixRoom, RoomMessageFile, RoomMessageText
+from langchain.chat_models import ChatOpenAI
+import json
+from langchain import LLMChain
+from langchain.prompts import ChatPromptTemplate
+
+from bot import Bot, print
+
+from langchain.embeddings import OpenAIEmbeddings, awa
+
+embeddings_model = OpenAIEmbeddings(
+    openai_api_key=os.environ["OPENAI_API_KEY"],
+    openai_api_base=os.environ["OPENAI_API_BASE"],
+    show_progress_bar=True,
+)
+
+client = Bot(
+    os.environ["BOT_CHATGPT_HOMESERVER"],
+    os.environ["BOT_CHATGPT_USER"],
+    os.environ["MATRIX_CHAIN_DEVICE"],
+    os.environ["BOT_CHATGPT_ACCESS_TOKEN"],
+)
+client.welcome_message = """你好👋，我是 matrix chain 中的大语言模型插件
+## 使用方式：
+- 直接在房间内发送消息，GPT 会在消息列中进行回复。GPT 会单独记住每个消息列中的所有内容，每个消息列单独存在互不干扰
+## 配置方式：
+- 发送 "!system + 系统消息" 配置大语言模型的角色，例如发送 "!system 你是一个专业英语翻译，你要把我说的话翻译成英语。你可以调整语序结构和用词让翻译更加通顺。"
+"""
+
+
+class SilentUndefined(jinja2.Undefined):
+    def _fail_with_undefined_error(self, *args, **kwargs):
+        print(f'jinja2.Undefined: "{self._undefined_name}" is undefined')
+        return ""
+
+
+def render(template: str, **kargs) -> str:
+    env = jinja2.Environment(undefined=SilentUndefined)
+    temp = env.from_string(template)
+
+    def now() -> str:
+        return datetime.datetime.now().strftime("%Y-%m-%d")
+
+    temp.globals["now"] = now
+    return temp.render(**kargs)
+
+
+async def get_reply_file_content(event):
+    """When user reply to a event, retrive the file content (document) of event
+
+    Return with the file content and token length
+    """
+    flattened = event.flattened()
+    formatted_body = flattened.get("content.formatted_body", "")
+    if not (
+        formatted_body.startswith("<mx-reply>") and "</mx-reply>" in formatted_body
+    ):
+        return "", 0
+
+    print("replacing file content")
+    formatted_body = formatted_body[
+        formatted_body.index("</mx-reply>") + len("</mx-reply>") :
+    ]
+    document_event_id = flattened.get("content.m.relates_to.m.in_reply_to.event_id", "")
+    fetch = await client.db.fetch_one(
+        query="""select d.content, d.token
+        from documents d
+        join event_document ed on d.md5 = ed.document_md5
+        where ed.event = :document_event_id""",
+        values={
+            "document_event_id": document_event_id,
+        },
+    )
+
+    if fetch and fetch[1] < 8192 + 4096:
+        content = fetch[0]
+        token = fetch[1]
+        print(content)
+        print(token)
+        print("-----------")
+        return content, token
+
+    print("document not found or too large", event.event_id)
+    return "", 0
+
+
+@client.ignore_link
+@client.message_callback_common_wrapper
+async def message_callback(room: MatrixRoom, event: RoomMessageText) -> None:
+    # handle set system message
+    if event.body.startswith("!"):
+        if event.body.startswith("!system"):
+            systemMessageContent = event.body.lstrip("!system").strip()
+            # save to db
+            await client.db.execute(
+                query="""
+                insert into room_configs (room, system, examples)
+                values (:room_id, :systemMessageContent, '{}')
+                on conflict (room)
+                do update set system = excluded.system, examples = '{}'
+                """,
+                values={
+                    "room_id": room.room_id,
+                    "systemMessageContent": systemMessageContent,
+                },
+            )
+            await client.react_ok(room.room_id, event.event_id)
+            return
+        if event.body.startswith("!model"):
+            model_name = event.body.lstrip("!model").strip()
+            # save to db
+            await client.db.execute(
+                query="""
+                insert into room_configs (room, model_name)
+                values (:room_id, :model_name)
+                on conflict (room)
+                do update set model_name = excluded.model_name
+                """,
+                values={"room_id": room.room_id, "model_name": model_name},
+            )
+            await client.react_ok(room.room_id, event.event_id)
+            return
+        if event.body.startswith("!temp"):
+            temperature = float(event.body.lstrip("!temp").strip())
+            # save to db
+            await client.db.execute(
+                query="""
+                insert into room_configs (room, temperature)
+                values (:room_id, :temperature)
+                on conflict (room)
+                do update set temperature = excluded.temperature
+                """,
+                values={"room_id": room.room_id, "temperature": temperature},
+            )
+            await client.react_ok(room.room_id, event.event_id)
+            return
+        return
+
+    messages: list[BaseMessage] = []
+    # query prompt from db
+    db_result = await client.db.fetch_one(
+        query="""
+        select system, examples, model_name, temperature
+        from room_configs
+        where room = :room_id
+        limit 1
+        """,
+        values={"room_id": room.room_id},
+    )
+    model_name: str = db_result[2] if db_result else ""
+    temperature: float = db_result[3] if db_result else 0
+
+    systemMessageContent: str = db_result[0] if db_result else ""
+    systemMessageContent = systemMessageContent or ""
+    if systemMessageContent:
+        messages.append(SystemMessage(content=systemMessageContent))
+
+    examples = db_result[1] if db_result else []
+    for i, m in enumerate(examples):
+        if not m:
+            print("Warning: message is empty", m)
+            continue
+        if i % 2 == 0:
+            messages.append(HumanMessage(content=m["content"], example=True))
+        else:
+            messages.append(AIMessage(content=m["content"], example=True))
+
+    exampleTokens = 0
+    exampleTokens += sum(client.get_token_length(m) for m in examples if m)
+
+    # get embedding
+    embedding_query = await client.db.fetch_all(
+        query="""
+        select content, distance, total_token from (
+            select
+                content,
+                document_md5,
+                distance,
+                sum(token) over (partition by room order by distance) as total_token
+            from (
+                select
+                    content,
+                    rd.room,
+                    e.document_md5,
+                    e.embedding <#> :embedding as distance,
+                    token
+                from embeddings e
+                join room_document rd on rd.document_md5 = e.document_md5
+                join room_configs rc on rc.room = rd.room
+                where rd.room = :room_id and rc.embedding
+                order by distance
+                limit 16
+            ) as sub
+        ) as sub2
+        where total_token < 6144
+        ;""",
+        values={
+            "embedding": str(await embeddings_model.aembed_query(event.body)),
+            "room_id": room.room_id,
+        },
+    )
+    print("emebdding_query", embedding_query)
+    embedding_token = 0
+    embedding_text = ""
+    if len(embedding_query) > 0:
+        embedding_query.reverse()
+        embedding_text = "\n\n".join([i[0] for i in embedding_query])
+        embedding_token = client.get_token_length(embedding_text)
+
+    filecontent, filetoken = await get_reply_file_content(event)
+
+    # query memory from db
+    max_token = 4096 * 4
+    token_margin = 4096
+    system_token = client.get_token_length(systemMessageContent) + exampleTokens
+    memory_token = max_token - token_margin - system_token - embedding_token - filetoken
+    print(
+        "system_token",
+        system_token,
+        "emebdding_token",
+        embedding_token,
+        "filetoken",
+        filetoken,
+    )
+    rows = await client.db.fetch_all(
+        query="""select role, content from (
+            select role, content, sum(token) over (partition by root order by id desc) as total_token
+            from memories
+            where root = :root
+            order by id
+        ) as sub
+        where total_token < :token
+        ;""",
+        values={"root": event.event_id, "token": memory_token},
+    )
+    for role, content in rows:
+        if role == 1:
+            messages.append(HumanMessage(content=content))
+        elif role == 2:
+            messages.append(AIMessage(content=content))
+        else:
+            print("Unknown message role", role, content)
+
+    temp = "{{input}}"
+    if filecontent and embedding_text:
+        temp = """## Reference information:
+
+{{embedding}}
+
+---
+
+## Query document:
+
+{{filecontent}}
+
+---
+
+{{input}}"""
+    elif embedding_text:
+        temp = """## Reference information:
+
+{{embedding}}
+
+---
+
+{{input}}"""
+    elif filecontent:
+        temp = """ ## Query document:
+
+{{filecontent}}
+
+---
+
+{{input}}"""
+    temp = render(
+        temp, input=event.body, embedding=embedding_text, filecontent=filecontent
+    )
+    messages.append(HumanMessage(content=temp))
+
+    total_token = (
+        sum(client.get_token_length(m.content) for m in messages) + len(messages) * 6
+    )
+    if not model_name:
+        model_name = "gpt-3.5-turbo" if total_token < 3939 else "gpt-3.5-turbo-16k"
+
+    print("messages", messages)
+    chat_model = ChatOpenAI(
+        openai_api_base=os.environ["OPENAI_API_BASE"],
+        openai_api_key=os.environ["OPENAI_API_KEY"],
+        model=model_name,
+        temperature=temperature,
+    )
+    chain = LLMChain(llm=chat_model, prompt=ChatPromptTemplate.from_messages(messages))
+
+    result = await chain.arun(
+        {
+            "input": event.body,
+            "embedding": embedding_text,
+            "filecontent": filecontent,
+        }
+    )
+    print(result)
+
+    await client.room_send(
+        room.room_id,
+        "m.room.message",
+        {
+            "body": result,
+            "msgtype": "m.text",
+            "m.relates_to": {
+                "rel_type": "m.thread",
+                "event_id": event.event_id,
+            },
+        },
+    )
+
+    # record query and result
+    await client.db.execute_many(
+        query="insert into memories(root, role, content, token) values (:root, :role, :content, :token)",
+        values=[
+            {
+                "root": event.event_id,
+                "role": 1,
+                "content": event.body,
+                "token": client.get_token_length(event.body),
+            },
+            {
+                "root": event.event_id,
+                "role": 2,
+                "content": result,
+                "token": client.get_token_length(result),
+            },
+        ],
+    )
+
+
+client.add_event_callback(message_callback, RoomMessageText)
+
+
+@client.ignore_self_message
+@client.handel_no_gpt
+@client.log_message
+@client.with_typing
+@client.replace_command_mark
+@client.safe_try
+async def message_file(room: MatrixRoom, event: RoomMessageFile):
+    if not event.flattened().get("content.info.mimetype") == "application/json":
+        print("not application/json")
+        return
+    size = event.flattened().get("content.info.size", 1024 * 1024 + 1)
+    if size > 1024 * 1024:
+        print("json file too large")
+        return
+    print("event url", event.url)
+    j = requests.get(
+        f'https://yongyuancv.cn/_matrix/media/r0/download/yongyuancv.cn/{event.url.rsplit("/", 1)[-1]}'
+    ).json()
+    if j.get("chatgpt_api_web_version") is None:
+        print("not chatgpt-api-web chatstore export file")
+        return
+    if j["chatgpt_api_web_version"] < "v1.5.0":
+        raise ValueError(j["chatgpt_api_web_version"])
+    examples = [m["content"] for m in j["history"] if m["example"]]
+    await client.db.execute(
+        query="""
+            insert into room_configs (room, system, examples)
+            values (:room_id, :system, :examples)
+            on conflict (room) do update set system = excluded.system, examples = excluded.examples
+        """,
+        values={
+            "room_id": room.room_id,
+            "system": j["systemMessageContent"],
+            "examples": str(examples),
+        },
+    )
+
+    await client.room_send(
+        room.room_id,
+        "m.reaction",
+        {
+            "m.relates_to": {
+                "event_id": event.event_id,
+                "key": "😘",
+                "rel_type": "m.annotation",
+            }
+        },
+    )
+
+
+client.add_event_callback(message_file, RoomMessageFile)
+
+asyncio.run(client.sync_forever())
diff --git a/bot_db.py b/bot_db.py
new file mode 100644
index 0000000..a8aa58c
--- /dev/null
+++ b/bot_db.py
@@ -0,0 +1,444 @@
+import PyPDF2
+import html2text
+import re
+import hashlib
+from nio import (
+    DownloadError,
+    MatrixRoom,
+    RoomMessageAudio,
+    RoomMessageFile,
+    RoomMessageText,
+)
+from langchain.text_splitter import MarkdownTextSplitter
+from bot import Bot, print
+import asyncio
+import io
+import yt_dlp
+import os
+import subprocess
+from langchain.embeddings import OpenAIEmbeddings
+
+from selenium import webdriver
+
+print("lanuching driver")
+options = webdriver.FirefoxOptions()
+options.add_argument("-headless")
+driver = webdriver.Firefox(options=options)
+
+
+async def get_html(url: str) -> str:
+    driver.get(url)
+    await asyncio.sleep(3)
+    return driver.page_source or ""
+
+
+import openai
+
+embeddings_model = OpenAIEmbeddings(
+    openai_api_key=os.environ["OPENAI_API_KEY"],
+    openai_api_base=os.environ["OPENAI_API_BASE"],
+    show_progress_bar=True,
+)
+
+client = Bot(
+    os.environ["BOT_DB_HOMESERVER"],
+    os.environ["BOT_DB_USER"],
+    os.environ["MATRIX_CHAIN_DEVICE"],
+    os.environ["BOT_DB_ACCESS_TOKEN"],
+)
+client.welcome_message = """欢迎使用 matrix chain db 插件，我能将房间中的所有文件添加进embedding数据库，并为gpt提供支持
+## 使用方式
+- 发送文件或视频链接
+  目前支持文件格式：txt / pdf / md / doc / docx / ppt / pptx
+  目前支持视频链接：Bilibili / Youtube
+## 配置选项
+- !clean 或 !clear 清除该房间中所有的embedding信息
+- !embedding on 或 !embedding off 开启或关闭房间内embedding功能 （默认关闭）"""
+
+spliter = MarkdownTextSplitter(
+    chunk_size=400,
+    chunk_overlap=100,
+    length_function=client.get_token_length,
+)
+
+offices_mimetypes = [
+    "application/wps-office.docx",
+    "application/wps-office.doc",
+    "application/wps-office.pptx",
+    "application/wps-office.ppt",
+    "application/msword",
+    "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+    "application/vnd.openxmlformats-officedocument.wordprocessingml.template",
+    "application/vnd.ms-powerpoint",
+    "application/vnd.openxmlformats-officedocument.presentationml.presentation",
+    "application/vnd.oasis.opendocument.text",
+    "application/vnd.oasis.opendocument.presentation",
+]
+mimetypes = [
+    "text/plain",
+    "application/pdf",
+    "text/markdown",
+    "text/html",
+] + offices_mimetypes
+
+
+def allowed_file(mimetype: str):
+    return mimetype.lower() in mimetypes
+
+
+async def create_embedding(room, event, md5, content, url):
+    transaction = await client.db.transaction()
+    await client.db.execute(
+        query="""insert into documents (md5, content, token, url)
+        values (:md5, :content, :token, :url)
+        on conflict (md5) do nothing
+        ;""",
+        values={
+            "md5": md5,
+            "content": content,
+            "token": client.get_token_length(content),
+            "url": url,
+        },
+    )
+
+    rows = await client.db.fetch_all(
+        query="select document_md5 from room_document where room = :room and document_md5 = :md5 limit 1;",
+        values={"room": room.room_id, "md5": md5},
+    )
+    if len(rows) > 0:
+        await transaction.rollback()
+        print("document alreadly insert in room", md5, room.room_id)
+        await client.room_send(
+            room.room_id,
+            "m.reaction",
+            {
+                "m.relates_to": {
+                    "event_id": event.event_id,
+                    "key": "👍",
+                    "rel_type": "m.annotation",
+                }
+            },
+        )
+        return
+
+    await client.db.execute(
+        query="""
+            insert into room_document (room, document_md5)
+            values (:room_id, :md5)
+            on conflict (room, document_md5) do nothing
+        ;""",
+        values={"room_id": room.room_id, "md5": md5},
+    )
+
+    # start embedding
+    chunks = spliter.split_text(content)
+    print("chunks", len(chunks))
+    embeddings = await embeddings_model.aembed_documents(chunks, chunk_size=1600)
+    print("embedding finished", len(embeddings))
+    if len(chunks) != len(embeddings):
+        raise ValueError("asdf")
+    insert_data: list[dict] = []
+    for chunk, embedding in zip(chunks, embeddings):
+        insert_data.append(
+            {
+                "document_md5": md5,
+                "md5": hashlib.md5(chunk.encode()).hexdigest(),
+                "content": chunk,
+                "token": client.get_token_length(chunk),
+                "embedding": str(embedding),
+            }
+        )
+    await client.db.execute_many(
+        query="""insert into embeddings (document_md5, md5, content, token, embedding)
+        values (:document_md5, :md5, :content, :token, :embedding)
+        on conflict (document_md5, md5) do nothing
+        ;""",
+        values=insert_data,
+    )
+    print("insert", len(insert_data), "embedding data")
+
+    await client.db.execute(
+        query="""
+            insert into event_document (event, document_md5)
+            values (:event_id, :md5)
+            on conflict (event) do nothing
+        ;""",
+        values={"event_id": event.event_id, "md5": md5},
+    )
+
+    await transaction.commit()
+
+    await client.room_send(
+        room.room_id,
+        "m.reaction",
+        {
+            "m.relates_to": {
+                "event_id": event.event_id,
+                "key": "😘",
+                "rel_type": "m.annotation",
+            }
+        },
+    )
+
+
+def clean_html(html: str) -> str:
+    h2t = html2text.HTML2Text()
+    h2t.ignore_emphasis = True
+    h2t.ignore_images = True
+    h2t.ignore_links = True
+    h2t.body_width = 0
+    content = h2t.handle(html)
+    return content
+
+
+def clean_content(content: str, mimetype: str, document_md5: str) -> str:
+    # clean 0x00
+    content = content.replace("\x00", "")
+    # clean links
+    content = re.sub(r"\[.*?\]\(.*?\)", "", content)
+    content = re.sub(r"!\[.*?\]\(.*?\)", "", content)
+    # clean lines
+    lines = [i.strip() for i in content.split("\n\n")]
+    while "" in lines:
+        lines.remove("")
+
+    content = "\n\n".join(lines)
+    content = "\n".join([i.strip() for i in content.split("\n")])
+
+    return content
+
+
+def pdf_to_text(f) -> str:
+    pdf_reader = PyPDF2.PdfReader(f)
+    num_pages = len(pdf_reader.pages)
+
+    content = ""
+    for page_number in range(num_pages):
+        page = pdf_reader.pages[page_number]
+        content += page.extract_text()
+    return content
+
+
+@client.ignore_self_message
+@client.handel_no_gpt
+@client.log_message
+@client.with_typing
+@client.replace_command_mark
+@client.safe_try
+async def message_file(room: MatrixRoom, event: RoomMessageFile):
+    print("received file")
+    mimetype = event.flattened().get("content.info.mimetype", "")
+    if not allowed_file(mimetype):
+        print("not allowed file", event.body)
+        raise ValueError("not allowed file")
+    resp = await client.download(event.url)
+    if isinstance(resp, DownloadError):
+        raise ValueError("file donwload error")
+
+    assert isinstance(resp.body, bytes)
+    md5 = hashlib.md5(resp.body).hexdigest()
+
+    document_fetch_result = await client.db.execute(
+        query="select content from documents where md5 = :md5;", values={"md5": md5}
+    )
+    document_alreadly_exists = len(document_fetch_result) == 0
+
+    # get content
+    content = ""
+    # document not exists
+    if not document_alreadly_exists:
+        print("document", md5, "alreadly exists")
+        content = document_fetch_result[0][0]
+    else:
+        if mimetype == "text/plain" or mimetype == "text/markdown":
+            content = resp.body.decode()
+        elif mimetype == "text/html":
+            content = clean_html(resp.body.decode())
+        elif mimetype == "application/pdf":
+            f = io.BytesIO(resp.body)
+            content = pdf_to_text(f)
+        elif mimetype in offices_mimetypes:
+            # save file to temp dir
+            base = event.body.rsplit(".", 1)[0]
+            ext = event.body.rsplit(".", 1)[1]
+            print("base", base)
+            source_filepath = os.path.join("./cache/office", event.body)
+            txt_filename = base + ".txt"
+            txt_filepath = os.path.join("./cache/office", txt_filename)
+            print("source_filepath", source_filepath)
+            with open(source_filepath, "wb") as f:
+                f.write(resp.body)
+            if ext in ["doc", "docx", "odt"]:
+                process = subprocess.Popen(
+                    [
+                        "soffice",
+                        "--headless",
+                        "--convert-to",
+                        "txt:Text",
+                        "--outdir",
+                        "./cache/office",
+                        source_filepath,
+                    ]
+                )
+                process.wait()
+                with open(txt_filepath, "r") as f:
+                    content = f.read()
+            elif ext in ["ppt", "pptx", "odp"]:
+                pdf_filename = base + ".pdf"
+                pdf_filepath = os.path.join("./cache/office", pdf_filename)
+                process = subprocess.Popen(
+                    [
+                        "soffice",
+                        "--headless",
+                        "--convert-to",
+                        "pdf",
+                        "--outdir",
+                        "./cache/office",
+                        source_filepath,
+                    ]
+                )
+                process.wait()
+                with open(pdf_filepath, "rb") as f:
+                    content = pdf_to_text(f)
+            else:
+                raise ValueError("unknown ext: ", ext)
+            print("converted txt", content)
+        else:
+            raise ValueError("unknown mimetype", mimetype)
+
+    content = clean_content(content, mimetype, md5)
+
+    print("content length", len(content))
+
+    await create_embedding(room, event, md5, content, event.url)
+
+
+client.add_event_callback(message_file, RoomMessageFile)
+
+yt_dlp_support = ["b23.tv/", "www.bilibili.com/video/", "youtube.com/"]
+
+
+def allow_yt_dlp(link: str) -> bool:
+    if not link.startswith("http://") and not link.startswith("https://"):
+        return False
+    allow = False
+    for u in yt_dlp_support:
+        if u in link:
+            allow = True
+            break
+    return allow
+
+
+def allow_web(link: str) -> bool:
+    print("checking web url", link)
+    if not link.startswith("http://") and not link.startswith("https://"):
+        return False
+    return True
+
+
+@client.message_callback_common_wrapper
+async def message_text(room: MatrixRoom, event: RoomMessageText) -> None:
+    if event.body.startswith("!"):
+        should_react = True
+        if event.body.startswith("!clear") or event.body.startswith("!clean"):
+            # save to db
+            async with client.db.transaction():
+                await client.db.execute(
+                    query="""
+                        delete from embeddings e
+                        using room_document rd
+                        where e.document_md5 = rd.document_md5 and
+                            rd.room = :room_id;
+                    """,
+                    values={"room_id": room.room_id},
+                )
+                await client.db.execute(
+                    query="delete from room_document where room = :room_id;",
+                    values={"room_id": room.room_id},
+                )
+        elif event.body.startswith("!embedding"):
+            sp = event.body.split()
+            if len(sp) < 2:
+                return
+            if not sp[1].lower() in ["on", "off"]:
+                return
+            status = sp[1].lower() == "on"
+            await client.db.execute(
+                query="""
+                insert into room_configs (room, embedding)
+                values (:room_id, :status)
+                on conflict (room) do update set embedding = excluded.embedding
+            ;""",
+                values={"room_id": room.room_id, "status": status},
+            )
+        else:
+            should_react = False
+        if should_react:
+            await client.room_send(
+                room.room_id,
+                "m.reaction",
+                {
+                    "m.relates_to": {
+                        "event_id": event.event_id,
+                        "key": "😘",
+                        "rel_type": "m.annotation",
+                    }
+                },
+            )
+        return
+
+    if allow_yt_dlp(event.body.split()[0]):
+        # handle yt-dlp
+        ydl_opts = {
+            "format": "wa*",
+            # ℹ️ See help(yt_dlp.postprocessor) for a list of available Postprocessors and their arguments
+            "postprocessors": [
+                {  # Extract audio using ffmpeg
+                    "key": "FFmpegExtractAudio",
+                    #'preferredcodec': 'opus',
+                    #'preferredquality': 64,
+                }
+            ],
+        }
+
+        url = event.body.split()[0]
+
+        info = None
+        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+            info = ydl.extract_info(url, download=True)
+        filepath = info["requested_downloads"][0]["filepath"]
+        filename = info["requested_downloads"][0]["filename"]
+        title = info["title"]
+        realfilepath = os.path.join("./cache/yt-dlp", filename)
+        os.rename(filepath, realfilepath)
+
+        result = openai.Audio.transcribe(
+            file=open(realfilepath, "rb"),
+            model="large-v2",
+            prompt=title,
+        )
+        result = "\n".join([i.text for i in result["segments"]])
+        print(event.sender, result)
+
+        md5 = hashlib.md5(result.encode()).hexdigest()
+
+        await create_embedding(room, event, md5, result, url)
+        return
+
+    if allow_web(event.body.split()[0]):
+        url = event.body.split()[0]
+        print("downloading", url)
+        html = await get_html(url)
+        md5 = hashlib.md5(html.encode()).hexdigest()
+        content = clean_html(html)
+        content = clean_content(content, "text/markdown", md5)
+        if not content:
+            raise ValueError("Empty content")
+        print(content)
+        await create_embedding(room, event, md5, content, url)
+        return
+
+
+client.add_event_callback(message_text, RoomMessageText)
+
+asyncio.run(client.sync_forever())
diff --git a/bot_tts.py b/bot_tts.py
new file mode 100644
index 0000000..b796ee5
--- /dev/null
+++ b/bot_tts.py
@@ -0,0 +1,75 @@
+import asyncio
+import os
+import aiohttp
+import json
+import base64
+from langdetect import detect
+from bot import Bot, print
+from nio import MatrixRoom, RoomMessageText
+from io import BytesIO
+
+url = f"https://texttospeech.googleapis.com/v1/text:synthesize?key={os.environ['GOOGLE_TTS_API_KEY']}"
+
+client = Bot(
+    os.environ["BOT_TTS_HOMESERVER"],
+    os.environ["BOT_TTS_USER"],
+    os.environ["MATRIX_CHAIN_DEVICE"],
+    os.environ["BOT_TTS_ACCESS_TOKEN"],
+)
+
+
+async def tts(text: str):
+    lang = detect(text)
+    langMap = {
+        "zh-cn": {
+            "languageCode": "cmn-cn",
+            "name": "cmn-CN-Wavenet-B",
+        },
+        "en": {"languageCode": "en-US", "name": "en-US-Neural2-F"},
+        "ja": {"languageCode": "ja-JP", "name": "ja-JP-Neural2-B"},
+    }
+    voice = langMap.get(lang, langMap["en"])
+    async with aiohttp.ClientSession() as session:
+        payload = {
+            "input": {"text": text},
+            "voice": voice,
+            "audioConfig": {"audioEncoding": "OGG_OPUS", "speakingRate": 1.39},
+        }
+        headers = {"content-type": "application/json"}
+        async with session.post(url, data=json.dumps(payload), headers=headers) as resp:
+            data = await resp.json()
+            audio_content = data.get("audioContent")
+            decoded = base64.b64decode(audio_content)
+            return decoded
+
+
+@client.ignore_self_message
+@client.handel_no_gpt
+@client.log_message
+@client.with_typing
+@client.change_event_id_to_root_id
+@client.replace_command_mark
+@client.safe_try
+async def message_callback(room: MatrixRoom, event: RoomMessageText) -> None:
+    if not event.sender.startswith("@chatgpt-bot"):
+        return
+
+    audio = await tts(event.body)
+    # convert
+    resp, upload = await client.upload(BytesIO(audio), "audio/ogg", filesize=len(audio))
+    content = {
+        "msgtype": "m.audio",
+        "body": event.body if len(event.body) < 20 else event.body[16] + "...",
+        "info": {"mimetype": "audio/ogg", "size": len(audio)},
+        "url": resp.content_uri,
+        "m.relates_to": {
+            "rel_type": "m.thread",
+            "event_id": event.event_id,
+        },
+    }
+    await client.room_send(room.room_id, message_type="m.room.message", content=content)
+
+
+client.add_event_callback(message_callback, RoomMessageText)
+
+asyncio.run(client.sync_forever())
diff --git a/bot_whisper.py b/bot_whisper.py
new file mode 100644
index 0000000..120afb8
--- /dev/null
+++ b/bot_whisper.py
@@ -0,0 +1,169 @@
+import os
+from nio import DownloadError, MatrixRoom, RoomMessageAudio, RoomMessageFile
+import asyncio
+import openai
+import io
+from bot import Bot, print
+
+
+client = Bot(
+    os.environ["BOT_WHISPER_HOMESERVER"],
+    os.environ["BOT_WHISPER_USER"],
+    os.environ["MATRIX_CHAIN_DEVICE"],
+    os.environ["BOT_WHISPER_ACCESS_TOKEN"],
+)
+client.welcome_message = (
+    """欢迎使用 matrix chain whisper 插件，我能将房间中的语音消息转换成文字发出，如果语音过长，我会用文件形式发出"""
+)
+
+
+@client.message_callback_common_wrapper
+async def message_callback(room: MatrixRoom, event: RoomMessageAudio):
+    print("received message")
+    print(event.flattened())
+    if event.flattened().get("content.info.duration", 0) > 1000 * 60 * 5:
+        return await message_file(room, event)
+    if event.source.get("content", {}).get("org.matrix.msc1767.audio") is None:
+        # handle audio file
+        return await message_file(room, event)
+    resp = await client.download(event.url)
+    if isinstance(resp, DownloadError):
+        return
+
+    filelikeobj = io.BytesIO(resp.body)
+    filelikeobj.name = "matrixaudio.ogg"
+    # get prompt
+    rows = await client.db.execute(
+        query="""select content from (
+        select role, content, sum(token) over (partition by root order by id desc) as total_token
+        from memories
+        where root = :event_id
+        order by id
+    ) as sub
+    where total_token < 3039
+    ;""",
+        values={"event_id": event.event_id},
+    )
+    prompt = "".join([i[0] for i in rows])
+    # no memory
+    if not prompt:
+        db_result = await client.db.fetch_all(
+            query="select system, examples from room_configs where room = :room_id;",
+            values={"room_id": room.room_id},
+        )
+        if len(db_result) > 0:
+            systemMessageContent = db_result[0][0]
+            examples = [
+                m.get("content", "") for m in db_result[0][1] if m.get("example")
+            ]
+            while "" in examples:
+                examples.remove("")
+            if systemMessageContent:
+                prompt += systemMessageContent + "\n\n"
+            if len(examples) > 0:
+                prompt += "\n\n".join(examples)
+
+    print("initial_prompt", prompt)
+
+    result = openai.Audio.transcribe(file=filelikeobj, model="large-v2", prompt=prompt)
+    result = "\n".join([i.text for i in result["segments"]])
+    print(event.sender, result)
+
+    await client.room_send(
+        room.room_id,
+        "m.room.message",
+        {
+            "body": result,
+            "msgtype": "m.text",
+            "m.relates_to": {
+                "rel_type": "m.thread",
+                "event_id": event.event_id,
+            },
+        },
+    )
+
+
+client.add_event_callback(message_callback, RoomMessageAudio)
+
+ALLOWED_EXTENSIONS = {
+    "mp3",
+    "mp4",
+    "mpeg",
+    "mpga",
+    "m4a",
+    "wav",
+    "webm",
+    "3gp",
+    "flac",
+    "ogg",
+    "mkv",
+}
+
+
+def allowed_file(mimetype):
+    return "/" in mimetype and mimetype.rsplit("/", 1)[1].lower() in ALLOWED_EXTENSIONS
+
+
+def get_txt_filename(filename):
+    return filename + ".txt"
+
+
+async def message_file(room: MatrixRoom, event: RoomMessageFile):
+    print("received file")
+    if not allowed_file(event.flattened().get("content.info.mimetype")):
+        print("not allowed file", event.body)
+        raise Exception("not allowed file")
+    resp = await client.download(event.url)
+    if isinstance(resp, DownloadError):
+        return
+    filelikeobj = io.BytesIO(resp.body)
+    filelikeobj.name = event.body
+
+    # get prompt
+    rows = await client.db.execute(
+        query="""
+            select content from (
+                select role, content, sum(token) over (partition by root order by id desc) as total_token
+                from memories
+                where root = :event_id
+                order by id
+            ) as sub
+            where total_token < 3039
+            ;""",
+        values={"event_id": event.event_id},
+    )
+    prompt = "".join([i[0] for i in rows])
+    print("initial_prompt", prompt)
+
+    result = openai.Audio.transcribe(file=filelikeobj, model="large-v2", prompt=prompt)
+    result = "\n".join([i.text for i in result["segments"]])
+    print(event.sender, result)
+    resultfilelike = io.BytesIO(result.encode())
+    resultfilelike.name = get_txt_filename(event.body)
+    resultfileSize = len(result.encode())
+    uploadResp, _ = await client.upload(
+        resultfilelike, content_type="text/plain", filesize=resultfileSize
+    )
+    print("uri", uploadResp.content_uri)
+
+    await client.room_send(
+        room.room_id,
+        "m.room.message",
+        {
+            "body": resultfilelike.name,
+            "filename": resultfilelike.name,
+            "msgtype": "m.file",
+            "info": {
+                "mimetype": "text/plain",
+                "size": resultfileSize,
+            },
+            "m.relates_to": {
+                "rel_type": "m.thread",
+                "event_id": event.event_id,
+            },
+            "url": uploadResp.content_uri,
+        },
+    )
+
+
+asyncio.run(client.sync_forever())
diff --git a/docker-compose.yaml b/docker-compose.yaml
new file mode 100644
index 0000000..55ca815
--- /dev/null
+++ b/docker-compose.yaml
@@ -0,0 +1,25 @@
+version: '3'
+services:
+  bot-chatgpt:
+    image: matrix-chain
+    env_file:
+      - ./.env
+    command: python3 bot_chatgpt.py
+
+  bot-db:
+    image: matrix-chain
+    env_file:
+      - ./.env
+    command: python3 bot_db.py
+
+  bot-whisper:
+    image: matrix-chain
+    env_file:
+      - ./.env
+    command: python3 bot_whisper.py
+
+  bot-tts:
+    image: matrix-chain
+    env_file:
+      - ./.env
+    command: python3 bot_tts.py
\ No newline at end of file
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000..31038d5
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,13 @@
+databases
+aiopg
+matrix-nio
+langchain
+python-dotenv
+tiktoken
+PyPDF2
+html2text
+yt_dlp
+selenium
+openai
+langdetect
+jinja2
diff --git a/requirements_version.txt b/requirements_version.txt
new file mode 100644
index 0000000..cc2e13f
--- /dev/null
+++ b/requirements_version.txt
@@ -0,0 +1,72 @@
+aiofiles==23.2.1
+aiohttp==3.8.6
+aiohttp-socks==0.7.1
+aiopg==1.4.0
+aiosignal==1.3.1
+annotated-types==0.6.0
+anyio==3.7.1
+async-timeout==4.0.3
+attrs==23.1.0
+Brotli==1.1.0
+certifi==2023.7.22
+charset-normalizer==3.3.0
+databases==0.8.0
+dataclasses-json==0.6.1
+frozenlist==1.4.0
+greenlet==3.0.0
+h11==0.14.0
+h2==4.1.0
+hpack==4.0.0
+html2text==2020.1.16
+hyperframe==6.0.1
+idna==3.4
+Jinja2==3.1.2
+jsonpatch==1.33
+jsonpointer==2.4
+jsonschema==4.19.1
+jsonschema-specifications==2023.7.1
+langchain==0.0.319
+langdetect==1.0.9
+langsmith==0.0.49
+MarkupSafe==2.1.3
+marshmallow==3.20.1
+matrix-nio==0.22.1
+multidict==6.0.4
+mutagen==1.47.0
+mypy-extensions==1.0.0
+numpy==1.26.1
+openai==0.28.1
+outcome==1.3.0
+packaging==23.2
+psycopg2-binary==2.9.9
+pycryptodome==3.19.0
+pycryptodomex==3.19.0
+pydantic==2.4.2
+pydantic_core==2.10.1
+PyPDF2==3.0.1
+PySocks==1.7.1
+python-dotenv==1.0.0
+python-socks==2.4.3
+PyYAML==6.0.1
+referencing==0.30.2
+regex==2023.10.3
+requests==2.31.0
+rpds-py==0.10.6
+selenium==4.14.0
+six==1.16.0
+sniffio==1.3.0
+sortedcontainers==2.4.0
+SQLAlchemy==1.4.49
+tenacity==8.2.3
+tiktoken==0.5.1
+tqdm==4.66.1
+trio==0.22.2
+trio-websocket==0.11.1
+typing-inspect==0.9.0
+typing_extensions==4.8.0
+unpaddedbase64==2.1.0
+urllib3==2.0.7
+websockets==11.0.3
+wsproto==1.2.0
+yarl==1.9.2
+yt-dlp==2023.10.13