Skip to main content


There is not yet a straightforward way to export personal WeChat messages. However if you just need no more than few hundreds of messages for model fine-tuning or few-shot examples, this notebook shows how to create your own chat loader that works on copy-pasted WeChat messages to a list of LangChain messages.

Highly inspired by

The process has five steps:

  1. Open your chat in the WeChat desktop app. Select messages you need by mouse-dragging or right-click. Due to restrictions, you can select up to 100 messages once a time. CMD/Ctrl + C to copy.
  2. Create the chat .txt file by pasting selected messages in a file on your local computer.
  3. Copy the chat loader definition from below to a local file.
  4. Initialize the WeChatChatLoader with the file path pointed to the text file.
  5. Call loader.load() (or loader.lazy_load()) to perform the conversion.

1. Create message dump​

This loader only supports .txt files in the format generated by copying messages in the app to your clipboard and pasting in a file. Below is an example.

%%writefile wechat_chats.txt
ε₯³ζœ‹ε‹ 2023/09/16 2:51 PM

η”·ζœ‹ε‹ 2023/09/16 2:51 PM

ε₯³ζœ‹ε‹ 2023/09/16 3:06 PM

η”·ζœ‹ε‹ 2023/09/16 3:06 PM
今倩εͺεΉ²ζˆδΊ†δΈ€δ»Άεƒζ ·ηš„δΊ‹

ε₯³ζœ‹ε‹ 2023/09/16 3:06 PM
Overwriting wechat_chats.txt

2. Define chat loader​

LangChain currently does not support

import logging
import re
from typing import Iterator, List

from langchain_community.chat_loaders import base as chat_loaders
from langchain_core.messages import BaseMessage, HumanMessage

logger = logging.getLogger()

class WeChatChatLoader(chat_loaders.BaseChatLoader):
def __init__(self, path: str):
Initialize the Discord chat loader.

path: Path to the exported Discord chat text file.
self.path = path
self._message_line_regex = re.compile(
r"(?P<sender>.+?) (?P<timestamp>\d{4}/\d{2}/\d{2} \d{1,2}:\d{2} (?:AM|PM))", # noqa
# flags=re.DOTALL,

def _append_message_to_results(
results: List,
current_sender: str,
current_timestamp: str,
current_content: List[str],
content = "\n".join(current_content).strip()
# skip non-text messages like stickers, images, etc.
if not re.match(r"\[.*\]", content):
"sender": current_sender,
"events": [{"message_time": current_timestamp}],
return results

def _load_single_chat_session_from_txt(
self, file_path: str
) -> chat_loaders.ChatSession:
Load a single chat session from a text file.

file_path: Path to the text file containing the chat messages.

A `ChatSession` object containing the loaded chat messages.
with open(file_path, "r", encoding="utf-8") as file:
lines = file.readlines()

results: List[BaseMessage] = []
current_sender = None
current_timestamp = None
current_content = []
for line in lines:
if re.match(self._message_line_regex, line):
if current_sender and current_content:
results = self._append_message_to_results(
results, current_sender, current_timestamp, current_content
current_sender, current_timestamp = re.match(
self._message_line_regex, line
current_content = []

if current_sender and current_content:
results = self._append_message_to_results(
results, current_sender, current_timestamp, current_content

return chat_loaders.ChatSession(messages=results)

def lazy_load(self) -> Iterator[chat_loaders.ChatSession]:
Lazy load the messages from the chat file and yield them in the required format.

A `ChatSession` object containing the loaded chat messages.
yield self._load_single_chat_session_from_txt(self.path)

2. Create loader​

We will point to the file we just wrote to disk.

loader = WeChatChatLoader(

3. Load Messages​

Assuming the format is correct, the loader will convert the chats to langchain messages.

from typing import List

from langchain_community.chat_loaders.utils import (
from langchain_core.chat_sessions import ChatSession

raw_messages = loader.lazy_load()
# Merge consecutive messages from the same sender into a single message
merged_messages = merge_chat_runs(raw_messages)
# Convert messages from "η”·ζœ‹ε‹" to AI messages
messages: List[ChatSession] = list(map_ai_messages(merged_messages, sender="η”·ζœ‹ε‹"))
[{'messages': [HumanMessage(content='ε€©ζ°”ζœ‰η‚Ήε‡‰', additional_kwargs={'sender': 'ε₯³ζœ‹ε‹', 'events': [{'message_time': '2023/09/16 2:51 PM'}]}, example=False),
AIMessage(content='ηη°Ÿε‡‰ι£Žθ‘—οΌŒη‘Άη΄ε―„ζ¨η”Ÿγ€‚ε΅‡ε›ζ‡’δΉ¦ζœ­οΌŒεΊ•η‰©ζ…°η§‹ζƒ…γ€‚', additional_kwargs={'sender': 'η”·ζœ‹ε‹', 'events': [{'message_time': '2023/09/16 2:51 PM'}]}, example=False),
HumanMessage(content='εΏ™δ»€δΉˆε‘’', additional_kwargs={'sender': 'ε₯³ζœ‹ε‹', 'events': [{'message_time': '2023/09/16 3:06 PM'}]}, example=False),
AIMessage(content='今倩εͺεΉ²ζˆδΊ†δΈ€δ»Άεƒζ ·ηš„δΊ‹\nι‚£ε°±ζ˜―ζƒ³δ½ ', additional_kwargs={'sender': 'η”·ζœ‹ε‹', 'events': [{'message_time': '2023/09/16 3:06 PM'}]}, example=False)]}]

Next Steps​

You can then use these messages how you see fit, such as fine-tuning a model, few-shot example selection, or directly make predictions for the next message

from langchain_openai import ChatOpenAI

llm = ChatOpenAI()

for chunk in[0]["messages"]):
print(chunk.content, end="", flush=True)

API Reference:

Help us out by providing feedback on this documentation page: