Select
Highlight the part of the reply that's actually worth keeping — not the whole conversation, just the good part.
a local, one-click memory keeper
Some AI answers are worth keeping.
You know the ones — the career advice that actually landed, the workout plan you meant to follow, the fix that saved your afternoon. AI Vault saves them where you'll actually stumble back onto them: a plain Markdown file on your own computer.
Free. Open source. Nothing leaves your computer.
the whole workflow
Three steps, in this order, every time.
Save
Click the small Save button that appears, or press Ctrl+Shift+S. No popup, no typing a filename, no folder to pick.
Rediscover
Weeks later, you open your Downloads and stumble onto it again — the same way you'd find a good note in a drawer.
what actually gets created
Just a file. Nothing clever.
No account, no sync, no AI-written summary pretending to know what your conversation meant. Just the text you chose, with a few honest details about where it came from.
--- source: claude.ai date: 2026-07-09, 6:41 PM title: "Systems Engineering Internship Roadmap" preview: "Start with the AWS fundamentals track, then move into..." url: https://claude.ai/chat/... --- Start with the AWS fundamentals track, then move into infrastructure-as-code once you're comfortable reading CloudFormation templates end to end.
deliberately not doing more than this
What this won't do, on purpose.
No AI-written titles
Filenames and previews are your own literal words, not a model's summary of them. You should recognize the file as something you saved, not something that was filed for you.
No cloud, no account
Nothing is uploaded anywhere. The extension writes a file to your Downloads folder and does nothing else.
No sorting by app
Claude and ChatGPT saves land in the same folder, on purpose. A good chat is a good chat, regardless of which app it came from.
where it works today
Two sites, done properly.
claude.ai
chatgpt.com
more, if people ask
Want another site supported? Open an issue — that's the actual roadmap.
open source
Small enough to read in one sitting.
MIT licensed. No build step, no framework — just the extension's own source, readable end to end.