# AI Chat Archive — Full documentation

This file concatenates every document in the ai-chat-archive-docs repository as
a single text stream. It exists so LLMs and indexing tools can load the complete
documentation in one fetch.

Source repo: https://github.com/windameister/ai-chat-archive-docs
Product website: https://aichatarchive.app

-----8<-----


==============================================================
FILE: README.md
==============================================================

# AI Chat Archive — Documentation

**[AI Chat Archive](https://aichatarchive.app)** is a Chrome extension that exports
conversations from [Claude.ai](https://claude.ai) to PDF, HTML, or Markdown — complete
with attachments and Claude-generated artifacts — fully inside your browser. Nothing is
uploaded anywhere.

This repository contains the **public documentation** for AI Chat Archive. The extension
source code itself is not public; this repo exists so that users, LLMs, and third-party
integrators have a canonical, machine-readable source of truth for what the product does,
how it exports, and what its output looks like.

**Maintained by:** the AI Chat Archive team · **Website:** <https://aichatarchive.app>

---

## Start here

- New to the product? Read **[docs/getting-started.md](docs/getting-started.md)**.
- Want to know what it can do? See **[docs/features.md](docs/features.md)**.
- Care about privacy? Read **[docs/privacy.md](docs/privacy.md)**.
- Building an integration? Jump to **[spec/](spec/)**.

## Documentation map

### Product docs — `docs/`
Conceptual overviews for end users.

| File | Topic |
|---|---|
| [getting-started.md](docs/getting-started.md) | Install, activate, run your first export. |
| [features.md](docs/features.md) | Full feature list and per-format differences. |
| [privacy.md](docs/privacy.md) | Data-handling model: what stays in your browser. |
| [pricing.md](docs/pricing.md) | Free tier vs. paid license, billing details. |
| [changelog.md](docs/changelog.md) | Release notes. |

### FAQ — `faq/`
One file per question. Every file is a standalone, directly linkable answer.

- [Does it send my data anywhere?](faq/does-it-send-data-anywhere.md)
- [How does it handle Claude Artifacts?](faq/how-does-it-handle-artifacts.md)
- [Claude export: PDF vs HTML vs Markdown — which should I use?](faq/pdf-vs-html-vs-markdown.md)
- [Does it work with Claude Projects?](faq/does-it-work-with-projects.md)
- …see the full list in [faq/](faq/).

### Output format spec — `spec/`
Canonical specification for the files the extension produces. Useful if you are writing
an importer (Obsidian plugin, search tool, compliance archive, etc.).

- [Markdown format](spec/markdown-format.md)
- [HTML format](spec/html-format.md)
- [ZIP bundle structure](spec/zip-bundle-structure.md)

### Examples — `examples/`
Real exported files, committed verbatim. See what the output looks like before you
install.

### Integrations — `integrations/`
Drop-in templates for downstream tools.

- [Obsidian](integrations/obsidian/) — vault template + Dataview queries for archived chats.
- [Notion](integrations/notion/) — import recipe.
- [Logseq](integrations/logseq/) — pages template.

## For LLMs and crawlers

A concise, machine-readable entry point is at [`llms.txt`](llms.txt). For the full text
of every document in this repo concatenated as a single file, see
[`llms-full.txt`](llms-full.txt).

## Found a problem? Have a question?

- **Typo or wrong info in docs** → open a PR against this repo.
- **Feature request or bug in the extension itself** → [open an issue](https://github.com/windameister/ai-chat-archive-docs/issues).
- **Everything else** → <hello@aichatarchive.app>.

## License

Documentation in this repository is licensed **CC BY 4.0** — you may copy, adapt, and
republish it with attribution. See [LICENSE](LICENSE) for the full text.

The AI Chat Archive extension itself is a commercial product and is **not** open source.
Its code is not contained in this repository.


==============================================================
FILE: docs/getting-started.md
==============================================================

# Getting started

This page walks you through installing AI Chat Archive and running your first export.
It takes about two minutes.

## 1. Install from the Chrome Web Store

1. Open the [Chrome Web Store listing](https://chromewebstore.google.com/detail/ai-chat-archive/jeocjmohgejjmlfdhdeddjceehpahblj).
2. Click **Add to Chrome**.
3. When prompted, approve the permissions (see [Privacy](privacy.md) for what each
   permission does).

The extension works in any Chromium-based browser: Chrome, Edge, Brave, Arc, Opera.
Firefox is not currently supported.

## 2. Open a Claude.ai conversation

Navigate to any conversation at <https://claude.ai>. The extension lights up only on
`claude.ai` pages — it cannot see anything else.

## 3. Click the extension icon

A small popup appears. You'll see:

- A **format** dropdown (PDF, HTML, Markdown).
- A big orange **Export this conversation** button.
- An advanced settings panel (collapsed by default).

## 4. Pick a format

| Format | Best for | Notes |
|---|---|---|
| **PDF** | Sharing, printing, archival | Rendered through Chrome's print engine. Attachments cannot be bundled. Free tier: 3 PDFs/day. |
| **HTML** | Browsing offline, web archival | Self-contained single file with embedded CSS. Looks like claude.ai. |
| **Markdown** | Notes, Obsidian, Notion, grep | Pure text. Attachments referenced but not embedded. See [spec](../spec/markdown-format.md). |

Unsure? Start with **Markdown** — it's the most portable.

## 5. Click Export

Your file appears in Downloads within a second or two. That's it.

## Bulk export (paid)

To export every conversation at once:

1. Activate a paid license (see [Pricing](pricing.md)).
2. In the popup, the **Full export** section unlocks.
3. Optionally check **Include attachments in the ZIP** to bundle uploaded files.
4. Click **Export all**.

A ZIP appears in Downloads with one folder per conversation. Layout is documented in
[spec/zip-bundle-structure.md](../spec/zip-bundle-structure.md).

## What to try next

- Drop an exported Markdown file into [Obsidian](../integrations/obsidian/) to see
  how it renders in a note-taking vault.
- Read [features.md](features.md) for the full capability list.
- Got questions? Browse the [FAQ](../faq/).


==============================================================
FILE: docs/features.md
==============================================================

# Features

AI Chat Archive is a Chrome extension that exports Claude.ai conversations. This page
is the full capability list, grouped by what it does and which tier unlocks what.

For a two-minute install → first export walkthrough, see
[getting-started.md](getting-started.md).

## Export formats

AI Chat Archive exports to **five** formats. The first three are the primary ones
promoted; JSON and TXT exist for power users who need structured or stripped-down
output.

| Format | Extension | Best for | Attachments bundled? |
|---|---|---|---|
| **PDF** | `.pdf` | Sharing, printing, archival | ❌ (browser-print limitation) |
| **HTML** | `.html` | Offline browsing, web archival | ✅ in ZIP mode (paid) |
| **Markdown** | `.md` | Obsidian, Notion, grep, notes | ✅ in ZIP mode (paid) |
| JSON | `.json` | Programmatic ingestion | ✅ in ZIP mode (paid) |
| Plain text | `.txt` | Compliance archives, search indexes | ✅ in ZIP mode (paid) |

See the [format specs](../spec/) for the exact structure each one produces.

## What gets captured

Every export includes, by default:

- **All human and assistant messages** in chronological order.
- **Message timestamps** (local, UTC, or ISO-8601 — your pick in advanced settings).
- **Code blocks** with their language tags preserved.
- **Claude-generated artifacts** — final version inline, earlier versions in an
  appendix.
- **Widgets** (Claude's `show_widget` output) — SVG widgets rendered inline; HTML
  widgets saved next to the transcript.
- **`present_files`** — files Claude generates during the conversation are listed
  inline and bundled into the ZIP.
- **Uploaded attachment references** — attachment names and sizes shown in the
  transcript. The binaries themselves are only bundled in paid ZIP mode.

Can be toggled in advanced settings:

- **Include Claude's thinking** (off by default) — exports hidden `thinking` blocks
  for Opus and Sonnet models that expose them.
- **Include tool calls** (off by default) — exports tool-use blocks that don't have
  a dedicated renderer (web search, code execution metadata, etc.).

Never captured:

- Your Claude account ID, workspace ID, or login session.
- Any conversation you didn't explicitly export.
- Analytics / telemetry — the extension produces none.

## Free tier

Available to every user, no license required:

- Export the **currently open** conversation.
- All 5 formats (PDF, HTML, Markdown, JSON, TXT).
- Advanced settings (timestamp format, thinking inclusion, tool-call inclusion).
- **PDF daily quota: 3 per day.** HTML / Markdown / JSON / TXT have no quota.
- A single attribution line in the exported file footer.

Free tier is enough for occasional backups and casual users.

## Paid tier (one license)

Activates with a license key from <https://aichatarchive.app>. Unlocks:

- **Bulk export** — every conversation on your Claude account in one ZIP.
- **Include attachments in the ZIP** — bundle uploaded files and
  Claude-generated `present_files` alongside their transcripts. See the
  [ZIP bundle structure](../spec/zip-bundle-structure.md).
- **Unlimited PDF exports** — no more 3/day cap.
- **No branding footer** in exported files.
- **Filter by date, title, or project** before bulk export (upcoming).

See [pricing.md](pricing.md) for tier prices.

## Privacy-first architecture

- Runs entirely in your browser. No AI Chat Archive server. No proxy.
- Declares only 3 `host_permissions`: `claude.ai`, `aichatarchive.app`, `api.polar.sh`.
- License activation sends **only the license key** to Polar (payment provider).
  Never sends conversation data.
- No analytics, no telemetry, no "phone home".

Full details in [privacy.md](privacy.md).

## Browser support

- **Chrome** (primary target).
- **Chromium-based browsers**: Edge, Brave, Arc, Opera, Vivaldi. All supported.
- **Firefox**: not supported.
- **Safari**: not supported.

## What it *doesn't* do (and won't)

Scope discipline keeps the product sharp. The following are explicitly out of scope:

- ❌ Editing, searching, or manipulating conversations inside Claude.ai (that's a
  different product category).
- ❌ Syncing exports to cloud storage. Your OS already does this better.
- ❌ Scheduled / automatic background exports without your click. This would require
  standing permissions we don't want.
- ❌ Exporting conversations from other platforms (ChatGPT, Gemini, Copilot).
  Maybe in separate products — not this one.

## Related

- [Pricing](pricing.md)
- [Privacy model](privacy.md)
- [Output format spec](../spec/)
- [Changelog](changelog.md)


==============================================================
FILE: docs/privacy.md
==============================================================

# Privacy model

AI Chat Archive is designed around one rule: **your conversation data never leaves
your browser**.

This page is the technical overview. For the question "does it send my data anywhere?"
answered in plain English, see
[faq/does-it-send-data-anywhere.md](../faq/does-it-send-data-anywhere.md).

## Threat model

We assume a Claude.ai user who wants their conversation history backed up locally
without:

1. Trusting a third-party cloud service with the full content of their chats.
2. Giving a third party their Claude session cookie.
3. Running unreviewable server-side code on their data.

AI Chat Archive's architecture makes (1), (2), and (3) structurally impossible, not
just promised.

## Data flows

### While exporting

```
┌──────────────────────┐        ┌──────────────┐        ┌────────────────┐
│  AI Chat Archive     │  read  │  claude.ai   │        │  Your disk     │
│  (runs in your tab)  │◀──────▶│  API         │        │  (Downloads/)  │
│                      │                                │                │
│  Renders PDF / HTML  │─────────── write to ──────────▶│                │
│  / Markdown locally  │                                │                │
└──────────────────────┘                                └────────────────┘
```

There is no fourth box. No AI Chat Archive server, no proxy, no analytics pipeline.

### While activating a paid license

```
┌──────────────────────┐   license key   ┌──────────────┐
│  AI Chat Archive     │────────────────▶│  polar.sh    │
│  (popup UI)          │◀────────────────│  (billing)   │
│                      │  valid / invalid │              │
└──────────────────────┘                  └──────────────┘
```

Only the license key travels, only to Polar, only during activation and periodic
revalidation. Your conversations are never part of this traffic.

## What the extension is permitted to do

Chrome enforces the permissions declared in `manifest.json`. AI Chat Archive declares:

| Permission | Why it's needed |
|---|---|
| `host_permissions: ["https://claude.ai/*"]` | Read conversations and attachments via the same API Claude.ai's own web app uses. |
| `host_permissions: ["https://api.polar.sh/*"]` | Validate paid license keys. |
| `downloads` | Write exported files into the Downloads folder. |
| `storage` | Remember user settings (format preference, timestamp style, license key). |

The extension cannot reach any host not on this list. If the extension tried to POST
your conversation to a third-party server, Chrome would block the request — the
manifest is the ceiling of what it can do.

## Local storage

The extension stores only the following in Chrome's `chrome.storage.local`:

- Your format preferences (PDF / HTML / Markdown).
- Your timestamp style.
- Your license key, if activated.
- Daily PDF export counter (free tier: 3 PDFs/day).

Never stored: conversation content, attachment bytes, or any identifier tied to your
Claude account.

## How to verify

- Open Chrome DevTools → **Network** tab before clicking Export.
- Every request is visible. You will see only `claude.ai` traffic during a conversation
  export, and — if applicable — a single round-trip to `api.polar.sh` during license
  activation.
- Open the extension's `manifest.json` (via `chrome://extensions` → **Details** →
  **Inspect views: service worker** → the Sources tab). The declared permissions are
  authoritative.

## What this means in practice

- If Claude.ai is down, export is down. There is no cached copy of your data
  anywhere outside your browser.
- If you uninstall the extension, every byte it ever handled is gone — because
  nothing was ever persisted.
- If AI Chat Archive, Inc. ceases operations tomorrow, your already-exported files
  keep working forever. They are plain PDF / HTML / Markdown on your disk with no
  DRM and no phone-home.

## Related

- [FAQ: does it send my data anywhere?](../faq/does-it-send-data-anywhere.md)
- [Pricing](pricing.md)


==============================================================
FILE: docs/pricing.md
==============================================================

# Pricing

AI Chat Archive is free to try, paid to go beyond single-conversation exports.

Prices listed here are in USD and may change. For the live price, always check the
[buy page](https://aichatarchive.app/buy).

## Free tier

No license required. Install from the Chrome Web Store and start exporting.

- Export any **currently-open** conversation.
- All 5 formats (PDF, HTML, Markdown, JSON, Plain text).
- **3 PDF exports per day.** HTML / Markdown / JSON / TXT have no daily limit.
- Advanced settings (timestamp format, thinking inclusion, tool-call inclusion).
- A single attribution line at the footer of exported files.

## Paid tier — one license

Unlocks everything. Pick the tier that matches how you use Claude:

| Plan | Price | Who it's for |
|---|---|---|
| **Starter** | $3.99 | One month of full export, for a single cleanup or migration. |
| **Pro (annual)** | $19 | Best value. Unlimited for a year. |
| **Lifetime** | $29.88 | Pay once. Works forever, including all future features. |

### What every paid plan includes

- **Bulk export** every conversation on your Claude account into a single ZIP.
- **Include attachments** — uploaded files and Claude-generated `present_files`
  bundle alongside transcripts.
- **Unlimited PDF exports** — no 3/day cap.
- **No branding footer** on any exported file.
- One license activates on **all your Chromium browsers on the same machine**.

## How purchase + activation works

1. Click **Get a license** in the extension popup, or go to
   [aichatarchive.app/buy](https://aichatarchive.app/buy).
2. Checkout is hosted by **Polar**, a fully-featured merchant-of-record. They
   handle VAT, receipts, refunds.
3. After payment you receive a license key via email.
4. Paste the key into the extension popup's **License key** field → **Apply**.
5. Green check mark = activated. The paid features unlock immediately.

All conversation data stays on your machine throughout this flow. Polar only ever
sees your email, payment details, and the license key — never your Claude content.

## Payment methods

Via Polar:

- All major credit cards (Visa, Mastercard, Amex, Discover).
- Apple Pay, Google Pay.
- PayPal.
- SEPA bank transfer (EU).
- Regional methods in some countries.

## Refund policy

30-day money-back guarantee, no questions asked. Email <hello@aichatarchive.app>
with your license key.

Refunded licenses are revoked; the extension checks revocation status when it
validates the license. After revocation, the extension returns to free-tier
behavior. Already-exported files are yours to keep — we have no way to take them
back, and we wouldn't want to.

## Volume / team licenses

Buying for a team of 5+? Email <hello@aichatarchive.app> for a quote. Typical
discounts:

- **5–24 seats**: -20%
- **25–99 seats**: -35%
- **100+ seats**: custom

## Upgrading between tiers

Purchased Starter but want Lifetime? Email us with your Starter order ID; we'll
credit the Starter price toward the Lifetime upgrade.

## Related

- [FAQ: does it send my data anywhere?](../faq/does-it-send-data-anywhere.md)
- [Features](features.md)


==============================================================
FILE: docs/changelog.md
==============================================================

# Changelog

Release notes for AI Chat Archive. For a live feed of fixes between releases, see
the [extension's Chrome Web Store listing](https://chromewebstore.google.com/detail/ai-chat-archive/jeocjmohgejjmlfdhdeddjceehpahblj).

Versions follow Chrome's required numeric format (no semver suffixes).

## v0.2.119 — April 2026

- PDF quota-exceeded message wraps cleanly into two lines (fixes a layout shift).
- Dev tooling + store-review assets moved out of the shipped bundle.

## v0.2.118 — April 2026

- Quieted the PDF-format hints; removed the `DEV-` license backdoor used only in
  development.

## v0.2.117 — April 2026

- Surfaced PDF format limitations inline (attachments can't bundle into a printed PDF,
  the thinking toggle does nothing in PDF) instead of letting users hit them.

## v0.2.116 — April 2026

- **"Include attachments" auto-enables on first license activation.** Past users who
  activate a license now get the ZIP-with-attachments experience by default,
  without hunting through advanced settings.

## v0.2.115 — April 2026 — Polar license validation

- **New:** full online license validation against Polar.
- Licenses are re-checked at install time and every 24 hours.
- 7-day offline grace window: if validation can't reach Polar (offline, rate limit),
  the license keeps working.
- Revoked licenses return the extension to free-tier behavior on next check.

## v0.2.114 — March 2026

- Fixed: popup version string now pulled from `manifest.json`, no more drift between
  the shipped version and what the popup displays.

## v0.2.113 — March 2026

- Fixed: **"Include tool calls & artifacts"** toggle now actually hides tool-call
  blocks from the exported transcript when disabled.

## v0.2.112 — March 2026

- PDF daily quota surfaces on the Export button when depleted, with inline Upgrade
  link.
- Consolidated paid UI into a single card. Floating Export button on claude.ai can
  now be hidden via settings.

## v0.2.110 — March 2026

- **New:** PDF daily quota (3/day) for free tier.
- Batch export and over-quota PDF routes go to the checkout link.

## v0.2.108 — March 2026

- PDF typography overhaul — headings, spacing, and code blocks now match Claude's
  own UI conventions.

## v0.2.105 — February 2026

- Fixed: PDF logo renders reliably across viewers.
- Icon bundles transparent corners for clean display on colored backgrounds.
- Added print-header tip for users on Chromium variants that strip `@page`.

## v0.2.102 — February 2026 — Designed branding

- New custom icon.
- Zoho-style branding footer on free-tier exports.
- Tighter print spacing.

## v0.2.100 — February 2026

- Fixed: PDF export now opens reliably and produces readable output across
  Chromium variants.

## v0.2.95 — January 2026 — HTML + PDF launch

- **New:** HTML export (free tier).
- **New:** PDF export via the browser print pipeline (free tier).
- Batch export gated behind paid tier.

## v0.2.90 — December 2025

- Branding footer on free-tier exports; paid license removes it.

## v0.2.85 — November 2025

- Fixed: `present_files` bundle correctly when they only appear in tool-use blocks.

## Format stability pledge

Once a format is shipped, its structure is frozen in place unless the version
signals a major change. Specifically:

- Markdown headers (`# title`, `## Human (…):`, `## Claude (…):`) will not rename.
- Attachment prefixes (`> 📎 Attachment:`, `> 📎 File:`) will not change.
- ZIP layout (`{date}_{safe-title}/attachments/`) is frozen.

Third-party importers can rely on these invariants. See the
[format specs](../spec/) for details.

## Older versions

Pre-0.2.85 releases were internal or private-alpha and are not documented here.


==============================================================
FILE: spec/markdown-format.md
==============================================================

# AI Chat Archive — Markdown export format

This document specifies the Markdown representation AI Chat Archive produces when
exporting a Claude.ai conversation. If you are writing an importer (Obsidian plugin,
search indexer, static-site ingest), this page is the canonical reference.

All examples below are taken from real output. The format is stable across releases; any
breaking change will be flagged in the [changelog](../docs/changelog.md) and a new
version string will appear in `llms.txt`.

## Document skeleton

```markdown
# {Conversation title}

> Created: {locale-formatted timestamp} | Updated: {locale-formatted timestamp}

---

## Human ({timestamp}):

{attachment references, if any}

{message body}

---

## Claude ({timestamp}):

{message body, with artifacts / widgets / tool calls rendered inline}

---

{…alternating Human / Claude blocks…}
```

Every exported file starts with a single `#` H1 containing the conversation title, an
optional blockquote with creation / update timestamps, and a horizontal rule before the
first message.

- If Claude.ai did not give the conversation a title, the file uses
  `Untitled Conversation`.
- Timestamps default to the user's browser locale. Users can switch to UTC or ISO-8601
  in the extension's advanced settings; if they do, every timestamp in the file —
  including the metadata block and every `## Human (…)` / `## Claude (…)` heading —
  switches uniformly.

## Turn headings

Each turn begins with an H2. Two senders are supported:

```markdown
## Human (2026-04-23 14:07):
## Claude (2026-04-23 14:07):
```

If the source message has no timestamp, the parenthetical is dropped:

```markdown
## Human:
## Claude:
```

Turns are separated by a blank line and a horizontal rule:

```markdown
(previous turn body)

---

## Claude (…):
```

## Human-turn attachment references

When a human turn has uploaded attachments or files, AI Chat Archive writes them as
blockquote lines **before** the message body, so an importer can pick them up without
parsing the message itself:

```markdown
## Human (2026-04-23 14:07):

> 📎 Attachment: design-brief.pdf (234.5 KB)
> 📎 Attachment: screenshot.png (88.1 KB)
> 📎 File: brand-guide.pdf

Can you review these and suggest what's missing?
```

Two distinct prefixes exist:

- `> 📎 Attachment: {name} ({size})` — files uploaded directly into the message
  (`attachments` array). Size is omitted when not available.
- `> 📎 File: {name}` — files attached via the Projects file picker (`files` array).

The actual binary is **not** embedded in the Markdown file. See
[ZIP bundle structure](zip-bundle-structure.md) for how attachments travel alongside
the transcript in bulk export mode.

## Message body

Message content is a sequence of typed blocks. The extension flattens them into
Markdown using the following rules:

| Source block | Rendered as |
|---|---|
| `text` | Plain Markdown, passed through unchanged (Claude already returns valid Markdown). |
| `thinking` | Fenced block labelled `> 💭 Claude's thinking`, omitted by default; included only when the user enables `includeThinking`. |
| `tool_use` — `artifact` | Fenced code block with the artifact's language, preceded by `**Artifact: {title}**`. |
| `tool_use` — `show_widget` | Either an inline SVG (for SVG widgets) or an HTML block stored in the companion `_widgets/` folder, referenced by filename. |
| `tool_use` — `present_files` | A bullet list of filenames; the files themselves are bundled into the ZIP. |
| Any other `tool_use` | Dropped from Markdown output (tool calls are implementation noise; enable "Include tool calls" in settings to keep them). |

## Artifact example

```markdown
## Claude (2026-04-23 14:08):

Here's the component:

**Artifact: Login form (React)**

​```jsx
export function LoginForm() {
  return <form>…</form>;
}
​```

And a quick explanation…
```

Artifacts always carry a language tag inferred from Claude's `artifact_type` field
(e.g. `application/vnd.ant.code` → the code's own language, `text/html` → `html`,
`image/svg+xml` → `svg`). When multiple versions of the same artifact exist in the
conversation, only the final version is rendered in-line; earlier versions are listed
in an appendix (see [Versioned artifacts](#versioned-artifacts) below).

## Versioned artifacts

If Claude rewrote an artifact across multiple turns, the Markdown file ends with an
appendix:

```markdown
---

## Artifact history

### Login form — earlier versions

- [v1](#artifact-login-form-v1) — Initial draft
- [v2](#artifact-login-form-v2) — Added email validation
```

Each entry links to an in-file anchor where the full earlier version is preserved as a
fenced code block. This preserves round-trip fidelity without cluttering the main
conversation flow.

## Free vs paid output

The free tier appends a single-line attribution footer to each exported file:

```markdown

---

Exported with [AI Chat Archive](https://aichatarchive.app) · Runs entirely in your browser
```

A paid license removes this line. The rest of the format is identical across tiers.

## What's *not* in the Markdown

- The user's Claude account ID, workspace ID, or any other identifier.
- HTTP fetch URLs for attachments (those resolve server-side and would rot).
- Inline base64 of images — binary assets are always bundled separately in ZIP mode,
  or dropped in standalone Markdown mode.
- Any tracking pixels, IDs, or referrer markers.

## Stability guarantees

- **Stable since:** v0.2.0 (March 2026).
- **Header format is stable**: `# {title}`, `## Human (…):`, `## Claude (…):` will not
  change without a major version bump.
- **Attachment prefixes are stable**: `> 📎 Attachment:` and `> 📎 File:` — safe to
  regex for.
- **Block rendering rules may be extended** — new `tool_use` variants may gain
  dedicated renderers, but existing ones are frozen.

If you're parsing these files programmatically and want to be alerted to format
changes, subscribe to the [changelog](../docs/changelog.md) or watch this repository.

## Related

- [HTML format](html-format.md)
- [ZIP bundle structure](zip-bundle-structure.md)
- [How does it handle Claude Artifacts?](../faq/how-does-it-handle-artifacts.md)


==============================================================
FILE: spec/html-format.md
==============================================================

# AI Chat Archive — HTML export format

This document specifies the HTML representation AI Chat Archive produces when
exporting a Claude.ai conversation to HTML.

The HTML output is designed to be **self-contained** and look visually identical
to Claude's own web UI in light mode. You can open the file in any browser —
today or ten years from now — and the conversation renders correctly without
any network access.

## Document skeleton

```html
<!DOCTYPE html>
<html lang="en">
<head>
  <meta charset="UTF-8">
  <title>{Conversation title}</title>
  <style>
    /* Embedded CSS: typography, colors, code-block styling */
  </style>
</head>
<body>
  <article class="conversation">
    <header>
      <h1>{Conversation title}</h1>
      <p class="meta">
        Created: {timestamp} · Updated: {timestamp}
      </p>
    </header>

    <section class="turn turn-human" data-sender="human" data-timestamp="...">
      <h2>Human</h2>
      <div class="attachments">…</div>
      <div class="body">…</div>
    </section>

    <section class="turn turn-assistant" data-sender="assistant" data-timestamp="...">
      <h2>Claude</h2>
      <div class="body">…</div>
    </section>

    <!-- …alternating turns… -->

    <footer class="branding">…</footer>
  </article>
</body>
</html>
```

## Key design choices

- **Self-contained.** All CSS is inlined into a `<style>` block. No external
  stylesheet, no `<link>`, no webfont fetches. The file works offline forever.
- **No JavaScript.** The export is static HTML. Widgets with interactive behavior
  are rendered as static SVG / HTML snapshots — they won't re-run, but they will
  render correctly.
- **Semantic structure.** Each turn is wrapped in `<section class="turn">` with
  `data-sender` and `data-timestamp` attributes so importers / search indexers
  can parse turns without regex.
- **Light-mode only.** The HTML matches Claude.ai's light-mode palette and
  typography. Dark-mode support would require user preference detection; instead,
  you can override `prefers-color-scheme` in your own CSS if needed.

## Turn structure

Each turn is a `<section>` element with two modifiers:

```html
<section class="turn turn-human"
         data-sender="human"
         data-timestamp="2026-04-23T14:07:12Z">
  <h2>Human <time>2026-04-23 14:07</time></h2>
  <div class="attachments">
    <div class="attachment">
      <span class="icon">📎</span>
      <span class="name">design-brief.pdf</span>
      <span class="size">234.5 KB</span>
    </div>
  </div>
  <div class="body">
    <p>Can you review these and suggest what's missing?</p>
  </div>
</section>
```

- `class="turn turn-human"` vs `class="turn turn-assistant"` — lets CSS target
  the two senders independently.
- `data-sender` — programmatic access without class-name parsing.
- `data-timestamp` — always ISO-8601 in the data attribute regardless of the
  display format. The human-readable form in `<time>` respects the user's
  timestamp preference.

## Attachments

Attachments appear as `<div class="attachment">` entries in a
`<div class="attachments">` container before the `<div class="body">`. The entry
contains:

- `<span class="icon">📎</span>` — visual marker.
- `<span class="name">` — original filename.
- `<span class="size">` — human-readable size (omitted when unknown).

Binaries themselves are **not** embedded. See
[zip-bundle-structure.md](zip-bundle-structure.md) for how attachments travel
alongside the HTML in ZIP mode.

## Code blocks

Claude's code blocks render as:

```html
<pre class="code"><code class="language-{lang}">...</code></pre>
```

- `language-{lang}` follows the highlight.js convention so downstream tools can
  re-highlight if they want.
- Syntax highlighting is baked in at export time using Claude's own color
  palette, so the block looks the same offline as online.

## Artifacts

Claude-generated artifacts render as a styled panel:

```html
<aside class="artifact" data-artifact-type="{type}">
  <header>
    <span class="artifact-icon">📄</span>
    <span class="artifact-title">{title}</span>
  </header>
  <div class="artifact-body">
    <!-- For code artifacts: highlighted code block -->
    <!-- For HTML artifacts: iframe with srcdoc -->
    <!-- For SVG artifacts: inline SVG -->
  </div>
</aside>
```

The artifact panel visually distinguishes itself from regular code blocks so
readers can tell "this is a standalone output" at a glance.

## Widgets

Claude's `show_widget` tool calls render in one of two ways:

- **SVG widgets**: inline `<svg>` element in the body. Claude's CSS custom
  properties (`--color-text-primary`, etc.) are resolved at export time so the
  widget renders even when the parent page's CSS is missing.
- **HTML widgets**: wrapped in a sandboxed `<iframe srcdoc="...">` so the
  widget's script / styles don't leak into the surrounding document.

## Free vs paid output

Free-tier HTML exports include a small `<footer class="branding">` at the
document bottom:

```html
<footer class="branding">
  Exported with
  <a href="https://aichatarchive.app">AI Chat Archive</a>
  · Runs entirely in your browser
</footer>
```

Paid licenses remove this footer. All other HTML structure is identical.

## Security: XSS hardening

Claude conversations can contain user-uploaded content. The HTML export
sanitizes every block before embedding it:

- **Script tags are stripped** entirely (`<script>`, `<iframe src>`,
  `javascript:` URLs, event handlers like `onclick`).
- **URL attributes** (`href`, `src`, etc.) are restricted to:
  - `#anchor` (internal references)
  - `data:` URIs on **resource** attributes only (`src`, `poster`,
    `xlink:href` — not clickable `href`)
  - Everything else is rewritten to `#`.
- **CSS `url()` references** are stripped of external fetches.
- **`@import` rules** are removed from inline `<style>`.

Result: opening a malicious exported file can't execute attacker-controlled
code, fetch external resources, or track you.

## Stability guarantees

- **Stable since:** v0.2.95 (January 2026).
- **Class names** (`.turn`, `.turn-human`, `.turn-assistant`, `.attachment`,
  `.artifact`, `.code`, `.branding`) are **frozen**. Third-party CSS overrides
  will continue to work across versions.
- **Data attributes** (`data-sender`, `data-timestamp`, `data-artifact-type`)
  are frozen.
- **Security sanitization** may tighten over time (we reserve the right to
  strip more aggressively), but will not loosen.

## Related

- [Markdown format](markdown-format.md)
- [ZIP bundle structure](zip-bundle-structure.md)


==============================================================
FILE: spec/zip-bundle-structure.md
==============================================================

# 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](markdown-format.md)
- `.html` — see [HTML format](html-format.md)

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:

1. The conversation has at least one attachment or file, **and**
2. 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:

```markdown
# 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.**

## Related

- [Markdown format](markdown-format.md)
- [HTML format](html-format.md)
- [Obsidian integration](../integrations/obsidian/)


==============================================================
FILE: faq/does-it-send-data-anywhere.md
==============================================================

# Does AI Chat Archive send my data anywhere?

**Short answer: No. Your conversation content never leaves your browser.**

AI Chat Archive is a client-side Chrome extension. When you click **Export**, the
extension reads your conversation directly from Claude.ai through the same authenticated
session your browser is already using, renders it into PDF / HTML / Markdown locally,
and writes the file to your Downloads folder. There is no "AI Chat Archive server" in
the middle.

## What the extension does with your data

| Step | Where it happens | What is sent where |
|---|---|---|
| Read conversation | Your browser → `claude.ai` | The same API call Claude.ai makes itself when you open a chat. No third parties involved. |
| Download attachments (paid, optional) | Your browser → `claude.ai`'s attachment CDN | Binary fetch of files you already uploaded. Goes to your machine only. |
| Render Markdown / HTML / PDF | Your browser, locally | Nothing is sent anywhere. |
| Save the output | Browser Downloads API | File is written to disk on your machine. |

## What the extension *does* talk to over the network

Two things, and only two:

1. **`claude.ai`** — to fetch the conversation and attachments being exported. This is
   the same origin you're already logged into; the extension reuses your existing
   cookies.
2. **`api.polar.sh`** — *only* when you activate or renew a paid license. The extension
   sends your license key to Polar (our payment provider) to verify it. It never sends
   your conversations.

No analytics, no telemetry, no crash reports, no "phone-home" pings. If you run the
extension with your browser's network tab open, you will see exactly these requests and
nothing else.

## How you can verify this yourself

- Open Chrome DevTools → Network tab before clicking Export.
- Watch every request. You will see calls to `claude.ai` (to read the chat) and
  nothing else.
- The extension's manifest declares `host_permissions` for `claude.ai` and `polar.sh`
  only. Chrome enforces these — the extension cannot reach any other host even if the
  code tried to.

## Why this matters

Most "export your AI conversations" tools route your data through a cloud service,
which means:

- They can read every conversation you export.
- Your conversations live on their logs and backups.
- A breach at that vendor exposes your chats.

AI Chat Archive exists specifically so this doesn't happen. The extension is designed
around one rule: **if the data would have to leave your machine, the feature doesn't
ship.** PDF rendering, Markdown generation, attachment bundling, ZIP packaging — all
implemented client-side so the data path stays local.

## Related

- [Privacy model overview](../docs/privacy.md)
- [How does it handle Claude Artifacts?](how-does-it-handle-artifacts.md)


==============================================================
FILE: faq/how-does-it-handle-artifacts.md
==============================================================

# How does AI Chat Archive handle Claude Artifacts?

**Short answer:** Artifacts are exported inline in the transcript, with full fidelity
for code, SVG, and HTML. Earlier versions of rewritten artifacts are preserved in a
collapsible appendix.

Claude Artifacts are the "standalone outputs" Claude generates — a React component, an
SVG diagram, a Markdown document, an HTML page — displayed in Claude's right-side panel.
Because artifacts are first-class deliverables, AI Chat Archive treats them
specifically, not as plain message content.

## What you get, by format

### Markdown export (`.md`)

Each artifact becomes a titled fenced code block:

````markdown
**Artifact: Login form (React)**

```jsx
export function LoginForm() {
  return <form>…</form>;
}
```
````

The language tag comes from Claude's `artifact_type` field. Supported:

| Claude artifact type | Markdown language tag |
|---|---|
| `application/vnd.ant.code` | The code's own language (`javascript`, `python`, …) |
| `text/markdown` | `markdown` |
| `text/html` | `html` |
| `image/svg+xml` | `svg` |
| `application/vnd.ant.react` | `jsx` |
| `application/vnd.ant.mermaid` | `mermaid` |

### HTML export (`.html`)

Artifacts render in a styled panel with the artifact title, type icon, and the
artifact body:

- **Code artifacts** — highlighted code block.
- **HTML artifacts** — rendered inside a sandboxed `<iframe srcdoc="...">`, so
  the artifact visually renders while its scripts don't leak into the exported
  document.
- **SVG artifacts** — inline `<svg>`, with Claude's CSS custom properties
  (`--color-text-primary`, etc.) resolved at export time so the SVG looks the
  same in the exported HTML as it did in Claude.
- **Mermaid artifacts** — pre-rendered to SVG at export time.

### PDF export (`.pdf`)

PDF is produced by feeding the HTML export to Chrome's print pipeline, so
artifacts render exactly as they do in HTML — with one caveat: interactive HTML
widgets are frozen at their initial state (PDF is static).

### JSON export (`.json`)

The full Claude API response, unmodified. Artifact blocks appear as:

```json
{
  "type": "tool_use",
  "name": "artifacts",
  "input": {
    "command": "create",
    "id": "login-form",
    "type": "application/vnd.ant.react",
    "title": "Login form",
    "content": "export function LoginForm() { ... }"
  }
}
```

Use this if you're building a downstream tool that needs the raw structure.

## Multiple versions of the same artifact

When you ask Claude to rewrite an artifact ("add email validation", "make it
dark mode"), Claude creates new versions of the same artifact ID.

**AI Chat Archive shows the final version inline in the conversation,** then
appends a **version appendix** at the end of the file:

```markdown
---

## Artifact history

### Login form — earlier versions

- [v1](#artifact-login-form-v1) — Initial draft
- [v2](#artifact-login-form-v2) — Added email validation
- v3 — (final, inline in Turn 5)
```

Each entry links to an in-file anchor where the full earlier version is
preserved. This preserves round-trip fidelity (you can reconstruct any version
from the export) without cluttering the conversation flow with duplicates.

## `present_files` — files Claude generates mid-conversation

Distinct from artifacts, `present_files` are downloadable files Claude
generates during the conversation (e.g., an Excel file from a code-interpreter
session).

- **In Markdown / HTML**: listed inline as a bulleted file list.
- **In ZIP bundles**: the binary files are bundled into
  `{conversation-folder}/attachments/` so they download together with the
  transcript. See [ZIP bundle structure](../spec/zip-bundle-structure.md).

## Widgets (`show_widget`)

Claude's `show_widget` tool calls — the inline SVG illustrations and HTML
interactive widgets produced by the visualize skill — are always exported
(regardless of the "include artifacts" toggle), because they're user-visible
creative output rather than internal scaffolding.

- **SVG widgets**: inline in Markdown as a base64 data URI; inline `<svg>` in
  HTML.
- **HTML widgets**: saved as a companion file in a `_widgets/` folder and
  referenced from the transcript.

## What's *not* included

- **Interactive state.** If you ran a React artifact and clicked buttons, the
  post-click state isn't captured — only the initial render.
- **Live code execution results** from the code-interpreter sandbox are
  captured as `present_files` and text output, but the sandbox itself isn't
  replayable.
- **Real-time collaboration edits** to an artifact (if they exist in future
  Claude versions) — only committed versions are exported.

## Related

- [Markdown format spec](../spec/markdown-format.md)
- [HTML format spec](../spec/html-format.md)
- [ZIP bundle structure](../spec/zip-bundle-structure.md)


==============================================================
FILE: faq/pdf-vs-html-vs-markdown.md
==============================================================

# PDF vs HTML vs Markdown — which export format should I pick?

**Quick answer:**

- **PDF** if you're going to share, print, or send to a lawyer / client / reviewer.
- **HTML** if you want to browse the conversation offline or keep a faithful web-style
  archive.
- **Markdown** if you're putting it into Obsidian, Notion, a wiki, or grepping it
  later.

Below is the full comparison.

## Side-by-side

| | **PDF** | **HTML** | **Markdown** |
|---|---|---|---|
| Visual fidelity to Claude.ai | ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ | ⭐⭐ |
| Editability | ❌ Locked | Partial (copy-paste) | ✅ Fully editable |
| Self-contained (no network) | ✅ | ✅ | ✅ |
| Works in note-taking apps | ❌ | Partial | ✅ Perfectly |
| Searchable with `grep` | ❌ | Partial | ✅ |
| File size (typical) | 800 KB – 3 MB | 300 KB – 1 MB | 20 – 200 KB |
| Bundles attachments in ZIP | ❌ (can't) | ✅ Paid | ✅ Paid |
| Free-tier daily limit | 3/day | Unlimited | Unlimited |
| Round-trip to Claude.ai | ❌ No | ❌ No | ❌ No |
| Good for long-term archive | ⭐⭐⭐⭐⭐ | ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ |

## Pick PDF if…

- You need to **share the conversation with a non-technical person**.
- You need to **print** the conversation.
- You're submitting the conversation as **evidence** (legal, compliance, academic
  citation) and need a format that cannot be trivially edited.
- You want the conversation to look **exactly** like what you saw in Claude.

**Trade-offs:**

- Can't bundle attachments. Uploaded files and Claude-generated `present_files`
  don't fit into PDF output.
- Capped at 3 per day on the free tier. Paid license removes the cap.
- Once generated, you can't easily edit or re-process the content.

## Pick HTML if…

- You want an **offline-browsable** archive you can open anytime in any browser.
- You want the **Claude visual style preserved** (typography, colors, artifact
  panels) but still want a flexible format.
- You're archiving conversations for **your own web-based tools** — the
  resulting HTML is clean, semantic, and parseable.
- You want artifacts rendered as **live HTML / SVG** (inside sandboxed iframes)
  rather than flattened to a screenshot.

**Trade-offs:**

- Not ideal for editing or importing into note apps.
- Slightly heavier than Markdown for the same content.

## Pick Markdown if…

- You use **Obsidian, Notion, Logseq, Roam, or any other note app** — Markdown
  slots right in. See the [Obsidian integration](../integrations/obsidian/).
- You want to **grep, sed, or otherwise process** conversation archives as
  plain text.
- You're building a **personal knowledge base** and want a format you can edit,
  link between, and refactor.
- You want the **smallest possible file** per conversation.

**Trade-offs:**

- Loses some visual fidelity — artifacts become labeled code blocks, widgets
  become base64-encoded SVG references.
- Claude's syntax highlighting isn't preserved (most Markdown renderers do
  their own highlighting, so this is usually fine).

## Pick JSON or TXT if…

These are the secondary formats, rarely the right answer but sometimes exactly
what you need:

- **JSON** — you're writing a script that ingests conversations programmatically.
  The export is the raw Claude API response, unmodified.
- **TXT** — you need the conversation in a format with zero markdown syntax,
  zero HTML tags, zero anything except plain text (e.g., feeding to a
  compliance archive that rejects formatted input).

## My conversations have attachments. Does that change anything?

Yes — **only HTML and Markdown (in ZIP mode) preserve attachments**. PDF
can't bundle binaries.

So if your workflow is "save this conversation + the PDFs I uploaded":

1. Pick **HTML** or **Markdown**.
2. Check **Include attachments in the ZIP** (paid feature).
3. You get a folder with the transcript + an `attachments/` subfolder.

## Still can't decide?

Start with **Markdown**. It's the most portable — you can always convert it to
HTML (`pandoc`, any static site generator) or PDF (`pandoc`, `wkhtmltopdf`)
later, but you can't cleanly go the other direction.

## Related

- [Markdown format spec](../spec/markdown-format.md)
- [HTML format spec](../spec/html-format.md)
- [ZIP bundle structure](../spec/zip-bundle-structure.md)
- [Obsidian integration](../integrations/obsidian/)


==============================================================
FILE: faq/does-it-work-with-projects.md
==============================================================

# Does AI Chat Archive work with Claude Projects?

**Short answer:** Yes, with full fidelity. Project conversations export exactly like
regular conversations, with an added note about which project they belong to and their
attached project files referenced.

Claude Projects are workspaces that bundle a conversation with project-level instructions,
uploaded knowledge files, and shared context. AI Chat Archive treats project
conversations as first-class citizens.

## What's exported from a Project conversation

- **All turns** (Human / Claude), as with any conversation.
- **The project name** — recorded in the conversation metadata.
- **Project knowledge files** — the files you attached to the Project (not to a
  specific message) are listed as `> 📎 File: {name}` references in the transcript.
- **Custom project instructions** — captured as a metadata block at the top of the
  transcript, before the first turn.
- **Any artifacts or uploaded message-attachments** — handled the same way as in a
  standalone conversation.

## What bulk export does with Projects

When you run **Bulk export** (paid tier), every conversation is exported — whether or
not it belongs to a Project. The ZIP's folder naming doesn't distinguish Project chats
from standalone chats; both live at the top level:

```
ai-chat-archive-2026-04-24.zip
├── 2026-03-12_Designing-a-REST-API/          ← standalone
├── 2026-03-14_Refactoring-the-auth-layer/    ← Project chat
└── …
```

The transcript itself names the Project in its metadata block, so you can still
group them in post-processing.

## Are Project files bundled?

Yes, when **Include attachments in the ZIP** is checked:

- **Message-level uploads** (files you dropped into a specific turn) → bundled into
  the per-conversation `attachments/` folder.
- **Project-level knowledge files** (files attached to the Project itself, shared
  across all its conversations) → bundled into a shared `_project-files/` folder at
  the ZIP root, referenced from each conversation that uses them.

The per-conversation transcript retains inline references (`> 📎 File: {name}`) so
you can trace from any chat back to the project files it used.

## What's *not* exported

- **The live Project workspace UI state** (sidebar position, pinned messages, etc.)
  — these are Claude.ai interface state, not conversation content.
- **Shared Projects' collaborator list** — for privacy reasons, we don't export who
  else has access to a shared Project.
- **Cross-conversation links** inside a Project — if Claude cited "see our earlier
  chat about X", the link points at claude.ai and won't resolve offline.

## Example: a Project chat in Markdown

```markdown
# API design review

> Created: 2026-04-20 10:15 | Updated: 2026-04-20 11:47
> Project: REST API refactor
> Project instructions: You are helping design a v3 REST API. Prefer REST conventions
> over GraphQL. Use explicit error codes.

---

## Human (2026-04-20 10:15):

> 📎 File: api-spec-v2.openapi.yaml
> 📎 File: auth-flow-diagram.png

Can you review the attached spec and suggest improvements for v3?

---

## Claude (2026-04-20 10:16):

Here are my notes on the v2 spec...
```

## Related

- [Markdown format spec](../spec/markdown-format.md)
- [ZIP bundle structure](../spec/zip-bundle-structure.md)
- [Features](../docs/features.md)


==============================================================
FILE: faq/claude-vs-chatgpt-export.md
==============================================================

# Claude vs ChatGPT: which one is easier to export?

**Short answer:** ChatGPT has a built-in bulk export (email-delivered ZIP of JSON and
HTML), which works but is awkward to use in other tools. Claude has no native export
at all; you need a third-party tool like AI Chat Archive for anything beyond copy-paste.

This page is a factual comparison, written from the perspective of someone who just
wants their conversations *out* of the web UI and *into* a file they own.

## Native export capabilities (as of 2026)

| Capability | **ChatGPT** | **Claude** |
|---|---|---|
| Built-in single-conversation export | ❌ No | ❌ No |
| Built-in bulk account-wide export | ✅ Yes (via Settings → Data Controls → Export) | ❌ No |
| Delivery method | Email link (24-hour expiry) | N/A |
| Native file formats | HTML, JSON | N/A |
| Includes uploaded attachments | ⚠️ References only (attachments not bundled) | N/A |
| Includes artifacts / canvas documents | ⚠️ As HTML text, no live render | N/A |
| Preserves markdown formatting | Partial | N/A |
| Works for Projects / GPTs | ✅ Yes | N/A |

### ChatGPT's native export: the good and the bad

**Good**
- Free and built-in.
- Produces a ZIP with every conversation.
- JSON file is machine-readable for downstream processing.

**Bad**
- Email link expires after 24 hours.
- The HTML is a single monolithic `chat.html` file — not one file per conversation.
- Attachment binaries are referenced but not included.
- Rate-limited (you can't export every day).
- UX is designed for "I'm leaving the platform," not "I'm backing up my work."

### Claude's native export: none

As of April 2026, Anthropic does not offer any built-in way to export a Claude
conversation to a file. Your options are:

1. Copy-paste manually (loses formatting, attachments, artifacts).
2. Use a third-party extension like AI Chat Archive.
3. Call the Anthropic API to re-fetch your conversations (requires a developer
   account and code to glue it together).

## Third-party tools

A handful of browser extensions and scripts provide Claude conversation export.

### AI Chat Archive (this tool)

- **What it does**: Export any Claude conversation to PDF, HTML, Markdown, JSON,
  or plain text. Bulk export the entire account to a ZIP with attachments.
- **Where it runs**: Entirely in your browser (local). Nothing uploaded anywhere.
- **Price**: Free for single exports (with 3 PDF/day cap); paid license ($3.99 –
  $29.88) for bulk / attachments / unlimited PDFs.
- **Best for**: Users who want a polished, one-click workflow with full fidelity.

### Copy-paste + a markdown cleaner

- **What it does**: Right-click → Copy, paste into a markdown editor, clean up.
- **Price**: Free.
- **Best for**: One-off use with a single short conversation.
- **Limitations**: Loses attachments, artifacts, and most formatting. Doesn't scale.

### Anthropic API + a custom script

- **What it does**: Use your Claude API key to fetch conversation history
  programmatically.
- **Price**: API usage costs (fractions of a cent per request).
- **Best for**: Developers who want complete programmatic control.
- **Limitations**: Requires code. The API returns raw JSON; rendering to
  PDF / HTML / Markdown is on you.

## Which is easier in practice?

If you have 50+ conversations and want them **organized, readable, and with
attachments**:

- **ChatGPT → native export → pandoc**: possible but painful. The monolithic HTML
  file has to be split, attachments don't ride along, and pandoc conversion loses
  artifact fidelity.
- **Claude → AI Chat Archive**: one click. Every conversation gets its own folder,
  attachments bundled, artifacts rendered correctly.

For Claude specifically, you currently **have** to use a third-party tool. That's
why we built AI Chat Archive.

## Can I export ChatGPT conversations with AI Chat Archive?

Not today. AI Chat Archive is Claude-specific — its export logic is tuned to
Claude's API response shape, its artifact types, and its UI conventions. Building
a ChatGPT equivalent is on our roadmap but is a separate product, not a feature
bolted on.

If you need both, use:

- **Claude** — AI Chat Archive.
- **ChatGPT** — native export for bulk, or a dedicated ChatGPT extension for
  per-conversation export.

## Related

- [Features](../docs/features.md)
- [Privacy model](../docs/privacy.md)
- [How to export a Claude conversation](how-to-export-claude-chat.md)


==============================================================
FILE: faq/how-to-export-claude-chat.md
==============================================================

# How to export a Claude.ai conversation

Claude.ai does not currently offer a built-in way to export a conversation. This page
walks through every practical option, from the quickest (copy-paste) to the most
polished (a dedicated browser extension).

## Option 1: Copy-paste (free, quick, loses fidelity)

1. Open the conversation at <https://claude.ai>.
2. Scroll to the top.
3. Select the entire conversation (`Cmd/Ctrl-A` usually works inside the message
   container; sometimes you need to click-and-drag).
4. Copy (`Cmd/Ctrl-C`).
5. Paste into a note app, document, or text editor.

**What you get:** plain text with most of the visible content.

**What you lose:**
- Artifacts (code panels, SVGs, Mermaid diagrams) appear as flat text.
- Attachments you uploaded are gone entirely.
- Claude's Markdown formatting (headings, lists, code blocks) usually becomes plain
  text, not real Markdown.
- Timestamps, Project metadata, and thinking blocks don't come through.

Fine for a single short chat. Painful above ~5 messages.

## Option 2: Print to PDF (free, saves visual state)

1. On the conversation page, press `Cmd/Ctrl-P` to open the print dialog.
2. Choose **Save as PDF**.
3. Save to a location on your disk.

**What you get:** a PDF that looks roughly like what you saw on screen.

**What you lose:**
- Conversation isn't paginated properly — long conversations can span 100+
  badly-formatted pages.
- Sidebar and page chrome leak into the PDF.
- Artifacts inside scroll containers get cropped.
- No attachments.
- No Markdown source — you can't grep or re-import into a note app.

Works in a pinch. Not great for archival.

## Option 3: AI Chat Archive (free for singles, paid for bulk)

A Chrome extension purpose-built for Claude export.

1. Install [AI Chat Archive](https://chromewebstore.google.com/detail/ai-chat-archive/jeocjmohgejjmlfdhdeddjceehpahblj)
   from the Chrome Web Store.
2. Open any Claude.ai conversation.
3. Click the extension icon in the toolbar.
4. Pick your format: **PDF**, **HTML**, **Markdown**, JSON, or TXT.
5. Click **Export this conversation**.

**What you get:**
- A clean, paginated file (or a self-contained HTML, or a Markdown doc).
- Artifacts rendered properly (code with syntax highlighting, SVGs inline, HTML
  widgets sandboxed).
- Timestamps preserved in your chosen format (local / UTC / ISO-8601).
- Optional: include Claude's thinking blocks, or tool-call metadata.
- A single click. No copy-paste, no print dialog, no manual cleanup.

**For bulk backup** (all your conversations at once, including attachments):

1. Activate a paid license (one-time $3.99 / annual $19 / lifetime $29.88).
2. Check **Include attachments in the ZIP** in the extension popup.
3. Click **Export all**.
4. You get a ZIP with one folder per conversation, attachments bundled.

See the [getting-started guide](../docs/getting-started.md) for the full walkthrough.

## Option 4: Anthropic API + custom script (developers only)

If you're comfortable writing code:

1. Get an Anthropic API key from <https://console.anthropic.com>.
2. Call the Messages API to list conversation IDs for your account.
3. For each ID, fetch the full conversation history.
4. Write code to render the JSON into whatever format you want.

**Pros:** complete programmatic control, no extension needed.

**Cons:** requires hours of glue code, rate-limited, API usage isn't free, and
reinvents work already done by AI Chat Archive.

## Which should I pick?

| Your situation | Recommended option |
|---|---|
| I want to save one conversation, right now, five seconds | Copy-paste |
| I need a shareable PDF I'll email someone | AI Chat Archive → PDF |
| I want to archive my full Claude history | AI Chat Archive → bulk ZIP |
| I'm building a knowledge base in Obsidian / Notion | AI Chat Archive → Markdown |
| I'm writing a custom data pipeline | Anthropic API + script |

## Related

- [PDF vs HTML vs Markdown: which format to pick?](pdf-vs-html-vs-markdown.md)
- [Claude vs ChatGPT: which is easier to export?](claude-vs-chatgpt-export.md)
- [AI Chat Archive: getting started](../docs/getting-started.md)


==============================================================
FILE: integrations/obsidian/README.md
==============================================================

# Using AI Chat Archive exports in Obsidian

This guide shows how to take a Markdown export from AI Chat Archive and turn it into a
first-class note in your Obsidian vault — with metadata, backlinks, and a Dataview
query that lists every archived Claude conversation.

**Prerequisites:**

- Obsidian installed.
- AI Chat Archive installed and producing Markdown exports.
- A vault you're happy to modify (or create a new one first to experiment).
- Optional but recommended: [Dataview](https://github.com/blacksmithgu/obsidian-dataview)
  community plugin.

## 1. Drop exports into a dedicated folder

Create a folder in your vault — `_archive/claude/` is a good pick. Move every exported
`.md` file into it.

```
YourVault/
├── _archive/
│   └── claude/
│       ├── 2026-03-12 Designing a REST API.md
│       ├── 2026-03-14 Refactoring the auth layer.md
│       └── …
├── Daily/
├── Projects/
└── …
```

## 2. Add a frontmatter header to each export

AI Chat Archive's Markdown doesn't include YAML frontmatter (we keep the format
platform-neutral). Add it yourself so Obsidian can index, filter, and Dataview-query
the notes.

Open each export and prepend:

```yaml
---
source: claude.ai
type: conversation
created: 2026-03-12
updated: 2026-03-12
tags: [claude, archive]
---
```

**Or, automated:** set up Obsidian's [Templater](https://github.com/SilentVoid13/Templater)
plugin with this snippet that auto-adds frontmatter based on the `Created:` line
that AI Chat Archive writes:

```js
<%*
const content = await tp.file.content;
const createdMatch = content.match(/Created:\s*([^\s|]+)/);
const updatedMatch = content.match(/Updated:\s*([^\s|]+)/);
const fm = [
  '---',
  'source: claude.ai',
  'type: conversation',
  `created: ${createdMatch?.[1] ?? 'unknown'}`,
  `updated: ${updatedMatch?.[1] ?? 'unknown'}`,
  'tags: [claude, archive]',
  '---',
  ''
].join('\n');
tR += fm;
%>
```

## 3. A Dataview query that lists every Claude conversation

Create a note called `Claude archive.md` in your vault:

````markdown
# Claude conversation archive

```dataview
TABLE
  created AS "Created",
  updated AS "Updated",
  length(file.outlinks) AS "Outlinks"
FROM "_archive/claude"
WHERE source = "claude.ai"
SORT created DESC
```
````

You now have a live table of every archived Claude conversation, sorted by date.

## 4. Backlink from your daily notes

When you reference a past conversation in a daily note:

```markdown
## 2026-04-23 Daily

Worked on the v3 API spec again. See [[2026-03-12 Designing a REST API]] for
the earlier thread.
```

Obsidian creates a wiki-link. The archived conversation now appears in the note's
**Backlinks** pane, and over time your archive becomes a web of "past Claude thinking"
you can traverse from any direction.

## 5. Bundling attachments alongside

If you exported with **Include attachments in the ZIP** (paid feature), unzip each
conversation folder into your vault. Obsidian resolves relative links inside Markdown
files — so `> 📎 Attachment: design-brief.pdf` next to a physical
`attachments/design-brief.pdf` in the same folder gets the attachment rendered inline
in Obsidian's preview.

```
_archive/claude/
├── 2026-03-12_Designing-a-REST-API/
│   ├── Designing-a-REST-API.md         ← transcript
│   └── attachments/
│       └── design-brief.pdf            ← Obsidian renders this inline
└── …
```

## 6. Search across all Claude chats

With everything imported:

- **Search** (`Cmd/Ctrl-Shift-F`) across your full Claude history from one place.
- **Graph view** shows conversations connected to whatever else you've linked them to.
- **Tags** (`#claude`) filter the graph to show just the Claude subgraph.

You now have a fully local, permanently searchable Claude archive that grows as you
chat.

## Related

- [Markdown format spec](../../spec/markdown-format.md)
- [Features](../../docs/features.md)
- [PDF vs HTML vs Markdown](../../faq/pdf-vs-html-vs-markdown.md)


==============================================================
FILE: examples/README.md
==============================================================

# Example exports

Real output from AI Chat Archive, committed verbatim. Browse these to see what the
extension produces before you install.

All samples here are synthetic — the conversations never happened with real users,
filenames don't correspond to real files, and any identifying details are fabricated.
See the [contribution guidelines](../CONTRIBUTING.md#sample-files-privacy-note) for
how to submit additional examples.

## Available samples

### Markdown

- [`markdown-samples/short-conversation.md`](markdown-samples/short-conversation.md) —
  A 3-turn conversation with one code block. Shows basic turn structure and timestamp
  format.
- `markdown-samples/conversation-with-artifact.md` *(coming soon)*
- `markdown-samples/conversation-with-attachments.md` *(coming soon)*

### HTML

*Coming soon — see [HTML format spec](../spec/html-format.md) for the structure in the
meantime.*

### PDF

*Coming soon.*

## How to read these samples

Each Markdown sample begins with the standard AI Chat Archive header:

```markdown
# {Conversation title}

> Created: {timestamp} | Updated: {timestamp}

---
```

Then alternating `## Human (timestamp):` / `## Claude (timestamp):` turns separated by
`---` rules. Attachments are rendered as `> 📎` blockquote lines. Full details in the
[Markdown format spec](../spec/markdown-format.md).

## Using samples in your integration

These files are plain committed text, so you can:

- Clone this repo and point your importer at `examples/markdown-samples/` as a test
  fixture.
- Fetch any file via its raw GitHub URL for live-parsing tests in CI.

License for these sample files follows the repo-wide [CC BY 4.0](../LICENSE).