Prompta

What it is

A vault for AI prompts. Folders, encrypted attachments, and delivery of a prompt from one user to another by email. From any saved prompt you can launch straight into ChatGPT, Claude, Gemini, DeepSeek or Grok in one click.

Why it exists

Prompts that are worth keeping end up in notes apps, chat scrollback and text files, which means they are found by remembering where you put them. A prompt is a small reusable artefact, and it deserves the same treatment as any other one: somewhere to live, a way to organise it, and a way to hand it to someone else.

The hard problem

The one genuinely awkward part is that a prompt is not only text. It carries attachments, and those attachments may be private, so they are encrypted — while still being deliverable to another user who has to be able to open them.

The decisions

Shared-element morphs, not modals

Framer Motion’s layoutId moves the actual card into its fullscreen state, so the thing you clicked is the thing you are now reading. A modal that fades in over a grid loses the connection between the two.

Plain CSS, no utility framework

A deliberate constraint, taken to see what the tradeoff actually feels like from the inside rather than reading about it.

A left-aligned scrollbar via an RTL trick

Setting direction: rtl on the container moves the scrollbar to the left. It is the only way to do it in CSS today.

What I would do differently

The RTL trick works, and it costs more than it looks. Flipping direction inverts text direction for everything inside the container, so every child has to be flipped back one by one, and each new child is a chance to forget. For a purely cosmetic gain, that is a bad trade and I would not take it again.

Plain CSS was worth doing once. What it actually taught me is that the cost is not writing the first screen, it is that every subsequent surface starts from zero and the conventions live only in my head. On a project I intended other people to work in, I would not repeat it.