How to read your Claude export

Claude.ai lets you download your full conversation history. What arrives is a ZIP of JSON files — complete, but not something you can comfortably read. Here's how to request the export, what's inside it, and how to read your conversations and artifacts as proper documents, offline.

Step 1 — Request your data export

1. Open claude.ai in a browser (or Claude Desktop — the export isn't offered in the iOS and Android apps).
2. Click your initials in the lower-left corner, then Settings.
3. Open the Privacy section.
4. Click Export data. There may be a short delay while the export is generated.
5. Anthropic emails a download link to your account address. Download it within 24 hours — the link expires after that, and you need to be signed in to use it.

Step 2 — What's inside the ZIP

• conversations.json — your complete chat history in machine-readable form: every conversation and message with timestamps. The valuable file — and the hard one to read raw.
• projects.json — Project instructions and attached documents, when present in your export.
• users.json — account metadata.

There's no browsable HTML in Claude's export — just the JSON. Opened in a text editor, a long history is effectively unreadable: one line per conversation, markdown escaped inside strings, no formatting.

Step 3 — Reading what matters, beautifully

Claude writes in Markdown — answers, reports, and especially artifacts. The work worth keeping can be saved as .md files and read properly:

1. Under any Claude response, click Copy — you get markdown, not flattened text. Markdown artifacts can be downloaded as files directly from the artifact panel.
2. Save it as a file ending in .md and get it to your device — Files, AirDrop, or any sync folder.
3. Open it in Read.md: tables, syntax-highlighted code, even Mermaid diagrams render the way they were meant to look — offline, on iPhone, iPad, Android, or Mac.

What you'll see

Here is a short saved conversation — a design question and Claude's answer with a numbered list, a code block and a comparison table — first as the markdown you copy out of Claude, then rendered. The rendered version is real HTML from the same engine as the markdown viewer online; paste your own answer there to try it.

# Rate limiter design — conversation with Claude

**You:** Design a sliding-window rate limiter for our API. Keep it simple.

**Claude:** Here's a minimal approach that avoids a cron job entirely:

1. Keep a list of request timestamps per API key.
2. On each request, drop timestamps older than 60 seconds.
3. Allow the request if the list is shorter than the key's limit.

```python
def allow(key, now):
    window = [t for t in store.get(key, []) if now - t < 60]
    if len(window) >= limit_for(key):
        return False
    store[key] = window + [now]
    return True
```

| Pros | Cons |
|---|---|
| No background job | Memory grows with traffic |
| Exact window | Needs a shared store across replicas |

**You:** What about replicas?

**Claude:** Use Redis with a sorted set per key — `ZADD`, `ZREMRANGEBYSCORE`, `ZCARD` — the same three steps, atomic in a small Lua script.

def allow(key, now):
    window = [t for t in store.get(key, []) if now - t < 60]
    if len(window) >= limit_for(key):
        return False
    store[key] = window + [now]
    return True

Frequently asked questions

What about the full archive?

Using ChatGPT too? See the guide to reading your ChatGPT export. Answers full of diagrams? The Mermaid viewer renders them, and the README viewer handles project docs you pull from GitHub. Everything starts from the Read.md homepage.