Skip to main content


This notebook shows how to use the iMessage chat loader. This class helps convert iMessage conversations to LangChain chat messages.

On MacOS, iMessage stores conversations in a sqlite database at ~/Library/Messages/chat.db (at least for macOS Ventura 13.4). The IMessageChatLoader loads from this database file.

  1. Create the IMessageChatLoader with the file path pointed to chat.db database you’d like to process.
  2. Call loader.load() (or loader.lazy_load()) to perform the conversion. Optionally use merge_chat_runs to combine message from the same sender in sequence, and/or map_ai_messages to convert messages from the specified sender to the “AIMessage” class.

1. Access Chat DB

It’s likely that your terminal is denied access to ~/Library/Messages. To use this class, you can copy the DB to an accessible directory (e.g., Documents) and load from there. Alternatively (and not recommended), you can grant full disk access for your terminal emulator in System Settings > Security and Privacy > Full Disk Access.

We have created an example database you can use at this linked drive file.

# This uses some example data
import requests

def download_drive_file(url: str, output_path: str = "chat.db") -> None:
file_id = url.split("/")[-2]
download_url = f"{file_id}"

response = requests.get(download_url)
if response.status_code != 200:
print("Failed to download the file.")

with open(output_path, "wb") as file:
print(f"File {output_path} downloaded.")

url = (

# Download file to chat.db
File chat.db downloaded.

2. Create the Chat Loader

Provide the loader with the file path to the zip directory. You can optionally specify the user id that maps to an ai message as well an configure whether to merge message runs.

from langchain_community.chat_loaders.imessage import IMessageChatLoader
loader = IMessageChatLoader(

3. Load messages

The load() (or lazy_load) methods return a list of “ChatSessions” that currently just contain a list of messages per loaded conversation. All messages are mapped to “HumanMessage” objects to start.

You can optionally choose to merge message “runs” (consecutive messages from the same sender) and select a sender to represent the “AI”. The fine-tuned LLM will learn to generate these AI messages.

from typing import List

from langchain_community.chat_loaders.base import ChatSession
from langchain_community.chat_loaders.utils import (

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 "Tortoise" to AI messages. Do you have a guess who these conversations are between?
chat_sessions: List[ChatSession] = list(
map_ai_messages(merged_messages, sender="Tortoise")
# Now all of the Tortoise's messages will take the AI message class
# which maps to the 'assistant' role in OpenAI's training format
[AIMessage(content="Slow and steady, that's my motto.", additional_kwargs={'message_time': 1693182723, 'sender': 'Tortoise'}, example=False),
HumanMessage(content='Speed is key!', additional_kwargs={'message_time': 1693182753, 'sender': 'Hare'}, example=False),
AIMessage(content='A balanced approach is more reliable.', additional_kwargs={'message_time': 1693182783, 'sender': 'Tortoise'}, example=False)]

3. Prepare for fine-tuning

Now it’s time to convert our chat messages to OpenAI dictionaries. We can use the convert_messages_for_finetuning utility to do so.

from langchain.adapters.openai import convert_messages_for_finetuning
training_data = convert_messages_for_finetuning(chat_sessions)
print(f"Prepared {len(training_data)} dialogues for training")
Prepared 10 dialogues for training

4. Fine-tune the model

It’s time to fine-tune the model. Make sure you have openai installed and have set your OPENAI_API_KEY appropriately

%pip install --upgrade --quiet  langchain-openai
import json
import time
from io import BytesIO

import openai

# We will write the jsonl file in memory
my_file = BytesIO()
for m in training_data:
my_file.write((json.dumps({"messages": m}) + "\n").encode("utf-8"))
training_file = openai.files.create(file=my_file, purpose="fine-tune")

# OpenAI audits each training file for compliance reasons.
# This make take a few minutes
status = openai.files.retrieve(
start_time = time.time()
while status != "processed":
print(f"Status=[{status}]... {time.time() - start_time:.2f}s", end="\r", flush=True)
status = openai.files.retrieve(
print(f"File {} ready after {time.time() - start_time:.2f} seconds.")
File file-zHIgf4r8LltZG3RFpkGd4Sjf ready after 10.19 seconds.

With the file ready, it’s time to kick off a training job.

job =,

Grab a cup of tea while your model is being prepared. This may take some time!

status =
start_time = time.time()
while status != "succeeded":
print(f"Status=[{status}]... {time.time() - start_time:.2f}s", end="\r", flush=True)
job =
status = job.status
Status=[running]... 524.95s

5. Use in LangChain

You can use the resulting model ID directly the ChatOpenAI model class.

from langchain_openai import ChatOpenAI

model = ChatOpenAI(
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate.from_messages(
("system", "You are speaking to hare."),
("human", "{input}"),

chain = prompt | model | StrOutputParser()
for tok in{"input": "What's the golden thread?"}):
print(tok, end="", flush=True)
A symbol of interconnectedness.