AI Chat Archive — ZIP bundle structure

When a user runs bulk export (a paid-tier feature) or exports a single conversation with Include attachments in the ZIP checked, AI Chat Archive produces a single .zip archive with a deterministic layout.

This document specifies that layout so third-party importers can reliably traverse it.

Top-level contents

ai-chat-archive-{YYYY-MM-DD}-{HHMMSS}.zip
├── 2026-03-12_Designing-a-REST-API/
│   ├── Designing-a-REST-API.md
│   └── attachments/
│       ├── design-brief.pdf
│       └── screenshot.png
├── 2026-03-14_Refactoring-the-auth-layer/
│   ├── Refactoring-the-auth-layer.md
│   └── attachments/
│       └── error-log.txt
├── 2026-04-02_Late-night-debug-session/
│   ├── Late-night-debug-session.md
│   └── _SKIPPED.md
└── …

Exactly one top-level folder per conversation. No shared flat directory.
Folder name is {created_at_date}_{safe_title}.
Each folder contains a single transcript file — .md or .html depending on the format the user picked — named after the conversation's safe title.
If the conversation had attachments and the user opted to bundle them, an attachments/ subfolder sits alongside the transcript.
If some attachments failed to download (private files, expired URLs, etc.), the folder contains a _SKIPPED.md listing each one and the reason.

Folder-name convention

{created_at_date}_{safe_title}

created_at_date — the conversation's creation date, formatted YYYY-MM-DD. Time component is omitted so folders sort naturally.
safe_title — the conversation's title, sanitized:
- Spaces replaced with -.
- Characters illegal on Windows/macOS/Linux filesystems (<>:"/\|?* and control characters) replaced with -.
- Collapsed runs of - trimmed.
- Truncated to 80 characters to stay under Windows MAX_PATH.
- If the title was empty, the fallback is Untitled-Conversation.

Example:

Original title: "My project: v2 🎉 (final)"
Sanitized folder: 2026-03-12_My-project-v2-final

Transcript file

The transcript's filename matches the folder's safe_title, with an extension matching the format:

.md — see Markdown format
.html — see HTML format

PDF is not supported in ZIP mode because PDF generation uses the browser's print pipeline, which cannot be batched.

`attachments/` subfolder

Present only when:

The conversation has at least one attachment or file, and
The user checked Include attachments in the ZIP.

Files inside preserve their original filenames, with the same sanitization rules as folder names. If two attachments share a name, the second gets a -2 suffix and so on.

The transcript's > 📎 Attachment: {name} references use the sanitized name, so an importer can join {folder}/attachments/{sanitized-name} reliably.

`_SKIPPED.md`

If one or more attachments could not be downloaded during export, AI Chat Archive writes a _SKIPPED.md file into the conversation folder:

# Files that could not be downloaded

- design-brief.pdf — network error
- old-prototype.fig — attachment expired

The transcript is still produced in full; only the binary attachments are missing.

What's not in the ZIP

Metadata files like manifest.json, conversations.json, or any machine index. Each conversation is self-describing via its transcript.
Hidden system files (.DS_Store, Thumbs.db). The ZIP is written by the extension directly, never by the OS.
Compression metadata beyond what DEFLATE naturally produces. No encryption, no passwords, no AES.
Attachments for conversations the user did not select. Bulk export respects the current filter — filtered-out chats are never touched.

Stability guarantees

Stable since: v0.2.0 (March 2026).
Top-level layout (one folder per conversation) is frozen. Future versions will not flatten or introduce a metadata index at the root.
Folder-name pattern is frozen — importers can regex ^\d{4}-\d{2}-\d{2}_(.+)/ to extract titles.
attachments/ subfolder placement is frozen.