List all conversation lines from exported data of ChatGPT.

export_chatgpt.py

#!/usr/bin/env python3
# Usage: ./export_chatgpt.py conversations.json ./path/to/export/dir

from dataclasses import dataclass

import itertools
import json
import logging
from logging import getLogger, StreamHandler
import sys

__author__ = "Amane Katagiri"
__author_email__ = "[email protected]"
__url__ = ""
__copyright__ = "Copyright (C) 2023 Amane Katagiri"
__credits__ = [
    "Amane Katagiri",
]
__date__ = "2023-07-31"
__license__ = "MIT License"
__version__ = "0.0.1"

logger = getLogger(__name__)
stream_handler = StreamHandler()
stream_handler.setLevel(logging.DEBUG)
logger.setLevel(logging.DEBUG)
logger.addHandler(stream_handler)


@dataclass
class Event:
    author: str
    content: str


@dataclass
class Conversation:
    event: Event
    parent: str | None
    children: list[str]


@dataclass
class Session:
    title: str
    mapping: dict[str, Conversation]


def new_session(session: dict) -> Session:
    return Session(
        session.get("title", "untitled"),
        dict(
            map(
                lambda conversation: [
                    conversation["id"],
                    Conversation(
                        Event(
                            (conversation.get("message") or {})
                            .get("author", {})
                            .get("role", "someone"),
                            "\n".join(
                                (conversation.get("message") or {})
                                .get("content", {})
                                .get("parts", [])
                            ),
                        ),
                        conversation.get("parent"),
                        conversation["children"],
                    ),
                ],
                session["mapping"].values(),
            )
        ),
    )


def parse(cursor: str, conversations: dict[str, Conversation]) -> list[list[Event]]:
    if cursor not in conversations:
        return [[]]

    children = conversations[cursor].children
    cursor_event = conversations[cursor].event

    if len(children) == 0:
        return [[cursor_event]]

    parsed = itertools.chain.from_iterable(
        map(lambda child: parse(child, conversations), children)
    )
    return [
        *map(
            lambda event: [
                *filter(lambda event: len(event.content) != 0, [cursor_event, *event])
            ],
            parsed,
        )
    ]


def main(infile: str, outdir: str):
    with open(infile, encoding="utf-8") as file:
        sessions: list[Session] = [
            *map(
                new_session,
                json.load(file),
            )
        ]
    for i, session in enumerate(sessions):
        title = session.title
        conversations = session.mapping
        root_id = next(
            filter(
                lambda conversation: conversation[1].parent is None,
                conversations.items(),
            )
        )[0]
        flattened = map(
            lambda conversation: "\n\n".join(
                map(lambda event: f"**{event.author}**: {event.content}", conversation)
            ),
            parse(root_id, conversations),
        )
        for j, conversation in enumerate(flattened):
            with open(
                f"{outdir}/{i:03d}_{title}_{j:03d}.md", "w", encoding="utf-8"
            ) as file:
                file.write(f"# {title} {i}\n\n" + conversation + "\n")


if __name__ == "__main__":
    main(sys.argv[1], sys.argv[2])