How to Sync ChatGPT Conversations to Obsidian (2026)

The workable setup uses two tools. For a single conversation, the free XWX Exporter extension saves the open chat as Markdown in one click — drop the file into your vault and it’s a note. For the whole history, the XWX Backup desktop app batch-exports your ChatGPT archive into a local Markdown vault and mirrors it into an XWX Backup subfolder inside your Obsidian vault automatically — fetching only new and changed conversations, so re-syncing never duplicates anything. Both produce standard Markdown with YAML frontmatter; neither needs a converter script. The rest of this guide walks through each layer.

ChatGPT is where most people’s AI volume lives, and it fails as an archive in three ordinary ways:

• Search matches titles only. Ask about “PostgreSQL connection pooling” in five chats titled “New chat” and the built-in search becomes useless — it can’t see inside your messages.
• Deletion is permanent. Deleted chats are gone from OpenAI’s side; there’s no trash folder to dig through.
• The official export isn’t documents. Settings → Data Controls → Export data emails you a ZIP with conversations.json — complete, machine-shaped, and unreadable in Obsidian.

A note in your vault, by contrast, is a file you own: full-text searchable across everything you’ve ever written, linkable to related notes, and durable no matter what happens to your ChatGPT account. The frontmatter is what turns it from a text dump into a queryable note — platform, model, timestamps, message count, tags.

Methods 1–2 are worth knowing; methods 3–4 are the ones people actually live in. If you only ever save highlights, the extension alone is enough. If you want the whole library — searchable, synchronized, and growing on its own — the desktop app is the piece that’s missing.

Step 1 — Install the extension. Search for XWX AI Chat Exporter in the Chrome Web Store and click Add to Chrome. No account needed.

Step 2 — Open any ChatGPT conversation and click the floating export button. Step 3 — Choose Markdown. Code blocks, tables, LaTeX, and images survive the trip.

Step 4 — Move the file into your vault. Drop it in any folder — it’s a plain .md file with the YAML frontmatter already attached, so Obsidian indexes it immediately.

The full walkthrough, including how the frontmatter works inside Obsidian, lives in the general AI-chats-to-Obsidian guide. This method’s limit is obvious: it’s one chat at a time, and the vault only grows when you remember to feed it.

The XWX Backup desktop app (Windows 10/11) flips the model: instead of saving chats one by one, it reads your whole logged-in ChatGPT session and lands every conversation as a Markdown file in a local vault — in one run.

1. Install XWX Backup and connect the extension once — the pairing stays on your machine over localhost, and no password is ever involved.
2. Pick ChatGPT from the platform list. The app covers every conversation your logged-in session can reach, across Free, Plus, Team, and Enterprise workspaces.
3. Choose a count — e.g. the latest 200 — or open the conversation list, filter by date, and tick exactly the ones you want.
4. Start. Progress reports added / skipped / failed per item, with pause and resume. Conversations already in the vault are skipped by hash comparison, so re-running never creates duplicates.

The vault always saves Markdown (that’s what powers search), and can additionally generate PDF, Word, JSON, or TXT copies per conversation if you configure them under Export options.

This is the part that makes the archive feel native. In XWX Backup, open Settings → Vault → Export to Obsidian:

1. Select your Obsidian vault root — the app validates it (the folder must contain an .obsidian subfolder).
2. Conversations sync into an `XWX Backup` subfolder inside that vault.
3. Click Sync now for the initial copy. After that, auto-sync is on by default: new and changed conversations are mirrored within seconds.

Every note carries YAML frontmatter, which is what Dataview and other Obsidian tools read:

---
id: "a1b2c3d4"
platform: "chatgpt"
title: "Debugging the WebSocket reconnect loop"
model: "gpt-5.2"
created_at: 2026-08-12T09:14:00Z
updated_at: 2026-08-12T11:02:00Z
messages: 24
tags:
  - "xwx-backup/chatgpt"
---

Sync semantics are deliberately conservative: only new and changed files are written, notes you’ve edited inside Obsidian are left untouched unless the conversation itself changed, and deleting a conversation in XWX Backup does not delete its note from Obsidian. If you also use Claude, the same setup covers it — see the Claude-to-Obsidian guide for the platform-specific details.

A one-time import is stale the moment you finish it. On the Backup page, enable Scheduled export and pick an interval — every 30 minutes, hourly, every 3 or 12 hours, or daily — plus an optional limit per run.

Each run is incremental: it fetches only what’s new or changed since the last successful run, compares content hashes against the vault, and skips the rest. A typical run moves a handful of files and finishes in seconds. When a slot fires, the app starts the export through the extension automatically — the browser doesn’t need to stay open between runs; the app wakes it if necessary. If the machine was off, that slot is simply skipped and the next run catches up.

This is the layer that survives busy weeks — which is the only kind of backup that does.

Two retrieval layers come with the setup:

• Full-text search in the app. The vault is indexed by SQLite full-text search — message-level, across all 11 platforms at once, with snippets and filters for platform, tag, date, and starred status. When ChatGPT’s title-only search fails you, the vault’s doesn’t (see how to search your ChatGPT history for the standalone workflow).
• Dataview inside Obsidian. Because the frontmatter is on every note, queries just work:

TABLE platform, model, messages, updated_at
FROM "XWX Backup"
WHERE platform = "chatgpt"
SORT updated_at DESC
LIMIT 10

Tags and stars you assign in the app are written back into the frontmatter, so they survive outside it — your curation stays with the files, not inside a proprietary database.

• Windows only, for now. XWX Backup ships for Windows 10/11 (64-bit); macOS and Linux builds are planned. On a Mac, the extension route (Method 1) plus OpenAI’s official export is the current alternative.
• Batch export covers platforms with conversation lists. That’s 11 platforms including ChatGPT, Claude, Gemini, DeepSeek, Grok, Perplexity, Kimi, Microsoft Copilot, GitHub Copilot, Google AI Studio, and Tencent Yuanbao. NotebookLM and Google Search AI Mode don’t expose conversation lists, so they stay single-chat exports via the extension.
• The automation layer is Pro. The extension and single-chat exports are free; so are the local vault and its full-text search (generated files carry a small watermark). Batch export, scheduled backups, and Obsidian sync are the Pro features — they’re what turn a manual archive into a self-maintaining one.
• Big histories take time. The deliberate pacing means a multi-year archive can take a few hours on the first import. That’s the cost of keeping your account safe, and it’s a one-time cost — every run after that is incremental.

• Developers — keeping debugging threads and architecture decisions in grep-able Markdown, linkable from project notes; pairing the vault with an MCP server turns the same archive into an agent knowledge base (see how).
• Researchers and students — literature discussions and study threads archived per project, citable and searchable years later.
• Writers — draft iterations and brainstorming sessions as notes they can link into their writing vault instead of scrolling a sidebar.
• Anyone whose ChatGPT history has outgrown ChatGPT’s search — the moment “I know I asked this” becomes a weekly experience, title search has lost.