Back up and export AI conversations: Codex, ChatGPT and Claude

Save AI conversations before you lose access

A missing chat is a recovery problem. A backup is a copy made beforehand. Copying a conversation to the same browser does not protect it from a broken computer, cleared storage or a lost profile.

Open the private archive tool →

1. Export conversations or select local session files.
2. Import the files, check the message count and read the conversation.
3. Download a source-preserving JSON backup. Add a password if you want encryption.
4. Also download the offline HTML reader. It opens without this website and offers Print / Save PDF.
5. Keep a copy on another drive or in a private cloud folder. Reopen the backup to verify it.

ChatGPT export and PDF

Request a data export through ChatGPT’s data controls or the Privacy Portal. Follow OpenAI’s export instructions. Extract the downloaded ZIP, then select conversations.json in the archive tool.

Download Markdown for a selected conversation, or the offline reader for your whole archive. Open the HTML file and choose Print / Save PDF. Text from alternate conversation branches is shown in time order; images, audio and external files are not included in the reading copy.

Claude export

Follow Anthropic’s data export instructions, extract the ZIP and select its conversation JSON file. The reader accepts conversation objects containing chat_messages. An exported archive is a copy to read, not a promise that Claude can import it into a different account.

Codex conversation backup

For Codex versions that store JSONL sessions, choose files from sessions or archived_sessions inside your Codex state directory. The default state location is ~/.codex, or your configured CODEX_HOME. Official state-location documentation and the official rollout source describe these locations.

Choose the session files you want to preserve. Do not select credential files such as auth.json. The tool keeps the exact UTF-8 source file alongside its readable text. Newer formats, compressed sessions and missing external files need their own backup method.

Optional: repeated Codex JSONL snapshots

Download the local backup script. It uses Node.js 22 or later, with no packages or network calls. Choose a private destination outside your Codex state directory, preferably a folder already synced to another device.

node codex-backup.mjs --destination "/path/to/private-synced-folder"

# Repeat while this process and computer stay running
node codex-backup.mjs --destination "/path/to/private-synced-folder" --interval 60

The script verifies each snapshot, keeps earlier versions, and copies only text-session JSONL files. It fails visibly if it finds compressed sessions or exceeds the reader’s size limits. It does not install a system service, manage cloud sync, encrypt snapshots, or copy authentication, project files or SQLite databases. Session files written during a snapshot may contain an incomplete last record; the original is retained and the reader reports unreadable lines.

Recovery has limits

Keeping a readable archive protects your own copy of past answers. It cannot recover content you never exported, restore a suspended account, or resume a Codex execution environment. For missing local sessions, use the Codex recovery checklist before repeating commands.

Privacy and passwords

Conversation files are processed in your browser and are not sent to CodexAnywhere, OpenAI or Anthropic. The archive page disables website analytics and recording. Local browser storage is unencrypted. JSON backups can optionally use AES-256-GCM password encryption with PBKDF2-SHA256; keep the password separately, because it cannot be reset. Markdown, offline HTML and helper snapshots are unencrypted. Read the data policy.

More Codex guides

Browse cloud, mobile and session-continuity guides.