Your best thinking happens in AI chats. Keep it.

Right now those conversations are scattered across Claude, ChatGPT, and Gemini — locked in someone else's database. chatdump sits quietly in your menu bar and files each one as plain Markdown in a folder you choose: easy to search, easy to back up, readable by your agents. Nothing leaves your Mac.

↓ Download for macOS View source on GitHub

Works with Claude ChatGPT Gemini

Open source No lock-in No telemetry

Your whole sidebar, archived

Every conversation, filed in one folder.

Each thread across Claude, ChatGPT, and Gemini becomes its own Markdown file in your vault — the archive grows as you chat.

chatgpt.com — john@example.com

What should I see in Kyoto in 3 days?

Day 1: Fushimi Inari early, then Tofuku-ji and an evening walk through Gion.

Day 2: Arashiyama bamboo grove, Tenryu-ji, and the Sagano riverside.

Day 3: Kinkaku-ji, Daitoku-ji, then Nishiki Market before you leave.

~/Vault — zsh — 76×20

$ tree Vault
Vault
├── chatgpt/john@example.com
│   ├── 2026-07-06_Kyoto_3-day_itinerary_8f3a2c1d.md
│   ├── 2026-07-04_Regex_lookbehind_help_c41b9e02.md
│   ├── 2026-07-01_Budget_planner_formula_1f77ab00.md
│   └── …and 142 more
├── claude/john@example.com
│   ├── 2026-07-05_Sourdough_starter_rescue_a7d02e55.md
│   └── …and 38 more
├── gemini/john@example.com
│   ├── 2026-07-01_Trip_photo_captions_3b8cf104.md
│   └── …and 11 more
└── .chatdump/cache

3 directories, 205 files
$ ▋

Every thread in your sidebar lands as Markdown you own — one file per conversation, one folder per account.

From chat to plain text

What a synced chat looks like.

One conversation in, one Markdown file out — YAML frontmatter for the metadata, a heading for each turn.

chatgpt.com

What should I see in Kyoto in 3 days?

Day 1: Fushimi Inari early, then Tofuku-ji and an evening walk through Gion.

Day 2: Arashiyama bamboo grove, Tenryu-ji, and the Sagano riverside.

Day 3: Kinkaku-ji, Daitoku-ji, then Nishiki Market before you leave.

2026-07-06_Kyoto_3-day_itinerary_8f3a2c1d.md

---
title: "Kyoto 3-day itinerary"
created: 2026-07-06T09:12:34.000Z
updated: 2026-07-06T09:14:02.000Z
model: auto
source: chatgpt
id: "8f3a2c1d-90ab-4cde-8123-456789abcdef"
parser_version: 2
---

## User

What should I see in Kyoto in 3 days?

## Assistant

Day 1: Fushimi Inari early, then Tofuku-ji and an evening walk through Gion.
Day 2: Arashiyama bamboo grove, Tenryu-ji, and the Sagano riverside.
Day 3: Kinkaku-ji, Daitoku-ji, then Nishiki Market before you leave.

The frontmatter keeps the title, timestamps, model, source, and conversation id — the rest is just the conversation.

Tue 09:41

Menu bar preview chatdump's menu bar dropdown, listing connected accounts and sync status

The whole app lives in one dropdown — accounts, sync status, and a manual sync button.

What it does

A small app with one job, done well.

No workspace to configure, no browser tabs to keep open — connect your accounts once and let it run.

Three providers

Claude, ChatGPT, and Gemini — each with its own accounts and independent sync state.

Plain Markdown

Every conversation lands as a readable .md file, so it works with any note app — Obsidian, Logseq, or just a folder of files.

Multiple accounts

Connect several accounts per provider. Each one syncs to its own vault, on its own schedule.

Set your rhythm

An auto-sync interval keeps things current, plus a per-account window controlling how far back each sync reaches.

Lives in the tray

No dock icon, no stray windows. Sign in once in an embedded browser and forget it's there.

Memory for your agents

A stdio MCP server and CLI let Claude Code, Codex, or your own tooling sync and read your archived conversations.

Own your data

Nothing leaves your Mac.

chatdump has no backend. Signing in to a provider opens an embedded browser window right inside the app; the resulting session cookies are written to local storage on your Mac and never sent anywhere else. Every sync reads conversations from the provider and writes Markdown straight to disk — there's no server in the middle to trust.

It's the same archive either way: a folder of Markdown files you already own, that you can read, grep, back up, or sync with anything you like.

Sign-in Embedded browser window

Sessions Stored locally, on your Mac

Servers None

Telemetry None

Output Markdown files you control

For agents & the terminal

Give your agents a memory.

Your archive is more than a backup — it's context. chatdump doubles as a stdio MCP server, so any MCP-aware agent — Claude Code, Codex, or your own harness — can sync accounts and read individual conversations in full. It reuses the same accounts and sessions as the menu bar app.

ask Ask a question through a signed-in session

conversation Fetch a conversation as Markdown — by id or share link

accounts List configured accounts & sync status

sync Sync selected accounts to their vaults

Full CLI & MCP reference →

Run it

$ /Applications/chatdump.app/Contents/MacOS/chatdump mcp

Project-scoped MCP client

{
  "mcpServers": {
    "chatdump": {
      "command": "/Applications/chatdump.app/Contents/MacOS/chatdump",
      "args": ["mcp"]
    }
  }
}

Call a tool

{
  "tool": "ask",
  "arguments": {
    "accountId": "openai:user@example.com",
    "prompt": "Summarize this thread."
  }
}