ChatGPT Export Viewer

Turn your ChatGPT data export into a local static site you can actually read: full-text search, images, LaTeX math, per-conversation contents. No upload, no server, no account.

Open the live demo: a generated archive of invented conversations, so you can see the result before running anything.

The problem

You asked ChatGPT for your data, waited a few days, and got a ZIP. Inside there is a conversations.json written as one endless line, a chat.html that dumps every conversation onto a single page with no search, and a pile of files called file_0a1b2c3d.dat with no extension.

If your export is recent, conversations.json may not even exist. It arrives split into conversations-000.json, conversations-001.json and so on, and an export requested through the Privacy Portal buries those shards inside a second ZIP. Tools that only look for the single file quietly show you a fraction of your history, or nothing at all.

What you get

• Search across everything you and the model wrote, with the matching sentence shown under each hit
• Images inline, including the extension-less .dat attachments from 2026-era exports
• LaTeX rendered with KaTeX bundled locally, not fetched from a CDN
• The thread as it was, reconstructed from current_node back through parent links, so edits and regenerations do not scramble the order
• Reasoning and tool steps folded into one collapsible block per turn instead of being dropped
• A missing-files report: how many attachments your conversations reference, how many the export actually contains, and which conversations lost the most
• Year and month navigation with a sticky table of contents, dark and light themes, three reading widths
• Everything is static. Copy the folder to a USB stick and it still works.

Quick start

Ask ChatGPT for the export first: Settings, Data controls, Export data, Confirm export. The download link arrives by email and expires 24 hours later.

You need Python 3.8 or newer. Pillow is optional and only used to reserve image dimensions.

git clone https://github.com/Gusarovv/chatgpt-export-viewer.git
cd chatgpt-export-viewer
python3 build_site.py ~/Downloads/your-export.zip

The site lands next to the export, in your-export-site/. Open your-export-site/site/index.html in any browser.

An unpacked export works too:

python3 build_site.py ~/Downloads/your-export/ ~/my-archive

Never used a terminal?

1. Install Python from python.org. On Windows, tick "Add Python to PATH" during setup.
2. Download this project with the green Code button, then Download ZIP, and unpack it.
3. Open the unpacked folder, then open a terminal there: on Windows, type cmd in the address bar of the folder window and press Enter; on macOS, right-click the folder and choose New Terminal at Folder.
4. Type python3 build_site.py (with the trailing space), then drag your export ZIP onto the window and press Enter.
5. When it finishes, it prints the path to index.html. Open that file.

Build a demo archive

The repository ships a generator for a fake export, so you can try the tool without touching your own data:

python3 tests/make_demo_export.py demo-export.zip
python3 build_site.py demo-export.zip

That gives you 450 conversations spread over three years, with images, math, code, tables, tool steps and a deliberately missing attachment. Pass --conversations N to the generator for a smaller or larger archive.

Limitations

• Official exports from mid-2026 no longer contain tool and system messages. Code interpreter output, web search results and Canvas documents are stripped on OpenAI's side before the ZIP is built. What remains, this tool shows; what is gone, no reader can recover.
• Attachments go missing. Exports regularly reference files they do not include: sometimes a random subset of generated images, sometimes all of them. That is why the missing-files report exists.
• Project names are not in the export, only opaque gizmo identifiers, so conversations that belonged to a project are shown without a project label.
• The format changes without notice. Field names appear and disappear between exports. The parser treats almost everything as optional and logs what it skips instead of failing.

How it works

build_site.py is a single file with no dependencies. It:

1. Finds conversation shards through export_manifest.json, falling back to a filename pattern, and opens the nested archive a Privacy Portal export wraps them in.
2. Streams each shard, so a 300 MB single-line JSON never lands in memory whole.
3. Walks each conversation from current_node up the parent chain, iteratively, because some threads run over 1200 nodes deep, past Python's recursion limit.
4. Resolves attachments through conversation_asset_file_names.json, restoring extensions from file signatures.
5. Writes HTML, a small search index, a lazy full-text index, and the missing-files report.

License

MIT