The Prompt Vault

Free. Version 0.2.2. Windows. No account, no key, no sign-up.

• Download for Windows (19 MB)

New in 0.2.1

Opening the app when it is already running now brings the copy you have back to the front, instead of starting a second one. Every TG Studios program works this way now.

New in 0.2.0

The first release was built to be used with the screen off, and it looked like it. This one is about the half I had not done.

• A details panel. Select a prompt and you see what it is for, which AI suits it, what you fill in, and the full wording. All of that used to be behind a dialog.
• A toolbar you can click. Copy, Fill in, Describe, Favorite, Edit, New. Every button says what it does and names its shortcut.
• Colour-coded categories, sortable columns, drag-to-resize panels, and a proper icon instead of the blank Windows default.
• It is no longer blurry. The window now scales properly on displays set above 100%, which is exactly the setting people with low vision use.
• Dark mode and high contrast are both followed. In high contrast every colour comes from your Windows theme and nothing is tinted.
• It updates itself now. New versions arrive the same way new prompts already did, and it always asks first.

Nothing about the keyboard changed. Same two lists, same shortcuts, plus Ctrl+1, Ctrl+2 and Ctrl+3 to jump straight to a pane.

The idea came from Matthew J (@bermudianbrit@mastodon.online), who suggested a prompt library that sits in the tray instead of scattered across a pile of Word documents. That is what this is.

Two lists, not a tree

Arrow down the categories. Tab across. Arrow down the prompts. Ctrl+C.

That is the whole app. Two flat lists instead of one nested tree, because a tree makes you keep track of how deep you are and two short lists do not.

Every row reads the title first, then the prompt itself, so you know what something is before you hear the whole thing.

The blanks are fields, not brackets

Most prompt collections hand you a wall of text with [YOUR TOPIC] buried somewhere in the middle and leave you to find it.

Press Enter here and you get one labelled edit field per blank, with a live preview of the finished prompt. Fill in four named fields, tab to Copy, done.

That single difference is why this beats a document. Hunting square brackets inside a paragraph with a screen reader is miserable. Filling in a field called "Target length" is not.

What is in it

502 prompts across 13 categories and 61 subcategories.

• Your AI Workspace, standing rules, memory, division-of-labour contracts, making the AI prove things rather than assert them
• Eyes on Demand, photos, packaging, screens, post, signs, appliances
• Code and Building, reading code, debugging, accessible apps, servers, agents in the terminal
• Video Production and Audio, Music and Podcasting, real pipelines, including the measuring step most guides leave out
• Images and Design, generating images you cannot see, and checking them
• Screen Readers and Accessibility, NVDA add-ons, audits, bug reports, and what to do when the AI and your screen reader fight each other
• Plus Foundations, Writing, Social, Everyday Life, Games, and Work

Every prompt has a plain-language description. Press Ctrl+D and it tells you what the prompt does, what you fill in, whether you need to attach a file first, and which AI it suits. That never calls an AI to explain itself, so it works offline and nothing you look at leaves your machine.

Keyboard

Make them yours

Edit any prompt to match how you actually work. Ctrl+R puts it back to the original whenever you want. Write your own and they live in My Prompts.

Nothing you write is ever uploaded. There is no account and no server that receives anything.

Your changes are stored separately from the shipped library, so an update can never overwrite a prompt you edited. When a prompt you changed gets a new official version, yours is kept and the new one waits behind Ctrl+R.

The library keeps growing

New prompts and whole new categories arrive on their own. The app checks once a day in the background and folds them in. If nothing arrived, it says nothing.

Those updates are signed, and the app checks the signature before it accepts anything. A prompt library is text that lands on your machine and then goes straight into an AI tool, so it is worth being careful about where it came from.

From 0.2.0 the app updates itself the same way. It tells you a new version is out and asks before it downloads anything. That one is signed with a different key from the prompts, because a key that can only publish text is a much smaller thing to lose than one that can run a program on your machine.

If you are on 0.1.0 you will need to download this one by hand, the once. There was no way to send a new version to a copy that had no way to receive one. After this it is automatic.

Where it came from

The prompts started with No Sight Required, my book on AI for blind users, and its bonus pack. The other 354 were written for this, and most of them came out of auditing what I actually do rather than guessing at what sounds useful.

That audit is also why the biggest original category is Your AI Workspace. The single most useful thing I found was not a clever prompt. It was that the things you tell an AI once, in a rules file, do more work than anything you retype every session.

Watch

Three short clips. Each one has burned-in captions, and the narration carries everything the picture does, so nothing is lost either way.

502 prompts in your tray

Arrow, tab, arrow, Ctrl+C, paste. The whole app in sixteen seconds.

Fields, not brackets

Why a labelled edit field per blank beats hunting square brackets in a wall of text.

Make them yours

Editing a prompt, reverting it, and what happens to your changes when the library updates.