🔮 DeepSeek Chat Backup

Never lose a conversation again. 🚀

A lightweight Python toolkit to back up every DeepSeek chat to individual .txt files — complete with timestamps, chat-start times, and 12-hour AM/PM formatting.

📖 How It Works • ⚡ Quick Start • 🔑 Get Credentials • 📂 Output Format • 💬 Contact

🎯 Why This Exists

DeepSeek does not provide a chat export / backup feature. 😤 If you have hundreds of conversations with code, ideas, research, and late-night thoughts — they live and die inside the browser.

This project was born from a real frustration. It started as a fragile Selenium scraper and evolved — through trial, error, and DevTools spelunking 🕵️ — into a clean, direct-API backup tool that hits DeepSeek's own endpoints.

No browser automation. No DOM scraping. No selectors that break next week. 🎉

Just pure HTTP, verified against the live API.

✨ Features

Feature	Description
🚀 Direct API access	Uses DeepSeek's own `/api/v0/` endpoints — no Selenium at runtime
📄 One file per chat	Each conversation → its own `.txt`, named after the title
⏱️ Full timestamps	12-hour format `hh:mm:ss AM/PM - dd mm yy` per message
🕐 Chat start time	Header shows exactly when the conversation began
📌 Smart sorting	Pinned chats first, then newest-first
🔐 Auto credential grabber	Separate script grabs token + cookies from Chrome
📦 Zero-config deps	Script auto-installs `requests` if missing
🪟 Windows 10 friendly	Runs perfectly from Python IDLE
🎨 Emoji-safe filenames	Sanitizes illegal chars for Windows

📸 Preview

Input: your DeepSeek account with 50+ chats 🗂️ Output: a folder like this 👇

deepseek_backups/
├── JDSG 1st Year Students 📚.txt
├── WatchDog Telegram 👺.txt
├── Telegram forward bot ❤‍🔥🌚.txt
├── Termux AI Automation Tool.txt
├── Roll back GitHub commits.txt
└── ... (one file per chat)

Each file looks like this 👇

=== DeepSeek Chat Backup ===
Title: DeepSeek chat backup script
Chat started: 03:02:12 PM - 02 10 2026
Total messages: 14
=============================

User [03:04:47 PM - 02 10 2026]:
F12 → Network → Fetch/XHR...

Response [03:05:04 PM - 02 10 2026]:
# 🧭 No DevTools? No Problem...

User [03:06:11 PM - 02 10 2026]:
another question...

Response [03:06:29 PM - 02 10 2026]:
another answer...

🧭 The Journey

This project didn't appear fully-formed. It was 6 iterations from "hello world" to "100/100 reliability". Here's the honest story. 📖

🔴 v1 — The Naïve Selenium Scraper (confidence: 35/100)

The first request: "write a Python script that opens Chrome, logs into DeepSeek, and scrapes every chat."

What we built:

Selenium + undetected-chromedriver
Manual login in the opened Chrome window
DOM scraping of the conversation sidebar and message containers
Guessed CSS selectors (.fa81, .f9bf7997)

Why it failed:

❌ DeepSeek uses obfuscated, rotating class names
❌ Virtual-scrolling sidebar hid older chats
❌ Message extraction missed code blocks
❌ undetected-chromedriver version drift with Chrome 154

Lesson learned: 🌱 DOM scraping a modern React SPA is like building on sand.

🟡 v2 — The Hardened Scraper (confidence: 75/100)

We added:

✅ Auto-install of dependencies via subprocess.check_call([sys.executable, "-m", "pip", ...])
✅ Multi-strategy selectors (tries XPath, CSS, role=, JS fallbacks)
✅ JavaScript TreeWalker fallback when CSS fails
✅ CHROME_VERSION_MAIN config to fix SessionNotCreatedException
✅ Dedup + parent/child filtering

Still fragile — DOM-scraping is a moving target. 🎯

Lesson learned: 🌱 Heuristics help, but you're still guessing.

🟢 v3 — The Network-API Pivot (confidence: 80/100)

Everything changed when we opened DevTools → Network tab and saw the real API calls. 🔍

Discovered endpoints:

GET  /api/v0/chat_session/fetch_page?lte_cursor.pinned=false
GET  /api/v0/chat/history_messages?chat_session_id=<uuid>&cache_version=4

Key insight: DeepSeek is a thin SPA over a clean REST API. Why scrape HTML when you can call the API directly? 🤯

New architecture:

Pure requests — no Selenium at runtime ⚡
50x faster (30 chats in ~15 seconds)
No DOM selectors to break 🎉
Cursor-paginated conversation list

Lesson learned: 🌱 Always check the Network tab first.

🔵 v4 — Field Verification (confidence: 93/100)

We captured the real JSON responses and stopped guessing.

Confirmed from live data:

{
  "data": {
    "biz_data": {
      "chat_sessions": [
        {
          "id": "43d6c267-...",       // ✅ session UUID
          "title": "JDSG 1st Year...",  // ✅ chat title
          "pinned": true,               // ✅ sort key
          "updated_at": 1790945739.079  // ✅ unix timestamp
        }
      ]
    }
  }
}

Lesson learned: 🌱 Verify, don't guess. One DevTools copy beats an hour of fallback code.

🟣 v5 — The Final Breakthrough (confidence: 100/100)

We captured the history_messages response and discovered the critical bug:

🔴 Content is NOT in msg["content"] — it lives inside msg["fragments"][].content!

That single discovery fixed all empty files. 🎯

Verified message shape:

{
  "message_id": 24,
  "role": "ASSISTANT",
  "inserted_at": 1790954971.4259999,
  "fragments": [
    {
      "id": 2,
      "type": "RESPONSE",
      "content": "# The actual message text..."
    }
  ]
}

Final features added:

⏱️ Chat start time = first message's inserted_at
🕐 Per-message timestamps in 12-hour AM/PM format
📌 Pinned-first, newest-second sort order
🎨 Clean header block per file

Lesson learned: 🌱 The last 10% of confidence costs 90% of the effort — but it's worth it.

📊 The Confidence Graph

100 │                                     ● v5 ✅
 90 │                              ● v4
 80 │                       ● v3
 70 │                ● v2
 60 │
 50 │
 40 │
 35 │   ● v1
 30 └───────────────────────────────────────────
      1    2    3    4    5
              Iteration →

⚡ Quick Start

1️⃣ Clone the repo

git clone https://github.com/realximanta/deepseek-backup.git
cd deepseek-backup

2️⃣ Get your credentials 🔑

Run the helper script — it opens Chrome, you log in, and it auto-grabs the token + cookies.

python get_credentials.py

See How to Get Your Credentials below for the manual DevTools method too.

3️⃣ Paste credentials into the main script

Open deepseek_backup.py and edit the CONFIG block at the top:

AUTHORIZATION = "Bearer YOUR_TOKEN_HERE"
COOKIE        = "smidV2=...; ds_session_id=..."
X_DEVICE_ID   = "your-device-id"

4️⃣ Run the backup 🚀

python deepseek_backup.py

5️⃣ Check the output 📂

deepseek_backups/

All chats, timestamped, done. 🎉

🔑 How to Get Your Credentials

You have two options: automatic (recommended) or manual via DevTools.

🅰️ Option A — Automatic (via `get_credentials.py`)

python get_credentials.py

What happens:

🚀 Chrome opens automatically
🧑 You log in to DeepSeek manually
👀 The script sniffs Chrome's own network layer
📝 It captures the Authorization header + all cookies
💾 Writes them to credentials.txt and prints them to console

Zero DevTools, zero copy-paste. 🎯

🅱️ Option B — Manual (via DevTools)

👆 Click to expand the 11-step DevTools guide

Step 1 — Open DeepSeek

Go to https://chat.deepseek.com and make sure you're logged in. ✅

Step 2 — Open DevTools

Press F12 (or Ctrl + Shift + I).

Step 3 — Go to the Network tab

Click the Network tab at the top of DevTools. 🌐

Step 4 — Click Fetch/XHR filter

Filters the request list to only API calls. 🎯

Step 5 — Type `api` in the filter box

Narrows down to DeepSeek's API endpoints. 🔍

Step 6 — Clear the list

Click the 🚫 clear icon so you start fresh.

Step 7 — Click a conversation in the sidebar

Pick any chat — this fires a history_messages request. 👆

Step 8 — Find the `history_messages` request

Look for /api/v0/chat/history_messages in the list. 🎯

Step 9 — Click the request

A detail panel opens on the right.

Step 10 — Go to the Headers tab

Scroll to Request Headers and copy:

authorization: Bearer eyJhbGciOiJIUzI1NiIs...
cookie: smidV2=...; ds_session_id=...

Step 11 — Also grab `x-device-id`

Scroll further down. Copy the x-device-id value.

Done! ✅

Paste them into deepseek_backup.py.

📂 Output Format

🗂️ Folder Structure

deepseek_backups/
├── Chat Title 1.txt
├── Chat Title 2.txt
├── Chat Title 3_1.txt   ← duplicate title auto-suffixed
└── ...

📄 File Format

=== DeepSeek Chat Backup ===
Title: <chat title>
Chat started: hh:mm:ss AM/PM - dd mm yy
Total messages: <count>
=============================

User [hh:mm:ss AM/PM - dd mm yy]:
<user message>

Response [hh:mm:ss AM/PM - dd mm yy]:
<assistant response>

User [hh:mm:ss AM/PM - dd mm yy]:
<next user message>

...

Timestamps use your local timezone. 🌍

🛠️ Technical Details

🔬 Click to see the API contract

Endpoint 1 — List Conversations

GET /api/v0/chat_session/fetch_page?lte_cursor.pinned=false
Headers:
  authorization: Bearer <token>
  cookie: <cookies>
  x-client-version: 2.5.0
  x-device-id: <device-uuid>

Response:

{
  "code": 0,
  "data": {
    "biz_data": {
      "chat_sessions": [
        {
          "id": "uuid",
          "title": "string",
          "title_type": "USER | SYSTEM",
          "pinned": true,
          "model_type": "default",
          "updated_at": 1790945739.079
        }
      ],
      "has_more": false,
      "next_cursor": null
    }
  }
}

Endpoint 2 — Fetch Messages

GET /api/v0/chat/history_messages?chat_session_id=<uuid>&cache_version=4
Headers:
  authorization: Bearer <token>
  cookie: <cookies>
  referer: https://chat.deepseek.com/a/chat/s/<uuid>

Response:

{
  "code": 0,
  "data": {
    "biz_data": {
      "chat_messages": [
        {
          "message_id": 24,
          "role": "USER | ASSISTANT",
          "inserted_at": 1790954971.4259999,
          "fragments": [
            {
              "type": "REQUEST | RESPONSE",
              "content": "message text here"
            }
          ]
        }
      ]
    }
  }
}

⚠️ Security Notes

🔴 Your token is a session key. Anyone who has it can read every chat in your account.

🔒 Never commit credentials to GitHub.
🚫 Never paste tokens in public chats/issues.
🔄 Rotate your token if you accidentally expose it — just log out and log back in on DeepSeek.
💾 credentials.txt is in .gitignore — keep it that way.

This repo ships with placeholder credentials only. You must supply your own.

🤝 Contributing

Pull requests are welcome! If DeepSeek changes their API and something breaks:

🍴 Fork the repo
🌿 Create a branch (git checkout -b fix/api-change)
🔧 Fix it
✅ Test it
📤 Open a PR

📜 License

MIT License — see LICENSE for details.

You are free to use, modify, and distribute this tool. Attribution appreciated but not required. 💚

💬 Contact

Ximanta 👨‍💻

📧 realximanta@gmail.com 🐙 github.com/realximanta

Built with frustration, coffee ☕, and DevTools.

⭐ If this saved your chats, give it a star! ⭐

Made with ❤️ by Ximanta

Because your conversations deserve to outlive the browser tab. 🌱

Name		Name	Last commit message	Last commit date
.gitignore		.gitignore
.nojekyll		.nojekyll
CNAME		CNAME
LICENSE		LICENSE
README.md		README.md
deepseek_backup.py		deepseek_backup.py
get_credentials.py		get_credentials.py
index.html		index.html

Latest commit

History

Folders and files

Repository files navigation