a nari project

Ruminari

The friction in using AI is rarely the model. It is the prompt, and everything a long conversation quietly loses. Ruminari is AI UX as a product: it works on both, and on nothing else.

Read on
  • the work AI UX itself
  • job one Sharper prompts
  • job two Context repair

the problem

The model is fine. The conversation is where it goes wrong.

Almost everyone using AI seriously runs into the same two walls, and neither of them is about how capable the model is. Both are experience problems, which is why a better model has never fixed either one.

WITH RUMINARI WITHOUT HELP ALL OF IT GONE CONTEXT RESTORED CONVERSATION LENGTH
The grey line is why a long session slowly stops behaving. The lit one drifts exactly the same way — it is the same conversation — but it is caught each time and carried back up, without stopping to re-explain anything. The band between them is the part you would otherwise be retyping.
asking well is hard

The prompt is the bottleneck

The same question, phrased two ways, gets two very different answers. Most people are left guessing at which phrasing the model wanted — and rewriting until something lands.

context decays

Long sessions forget

The further a conversation runs, the more of the early setup slips away. Constraints get dropped, decisions get re-litigated, and the work quietly drifts off the thing you agreed on an hour ago.

what it does

Fix the ask. Repair the thread.

Ruminari sits alongside the way you already work with AI and takes on the two jobs nobody should have to do by hand. Neither is something a model can do for itself — they are both work that happens around it.

prompt improvement

Say it the way it lands

Your intent goes in; a sharper, better-specified version comes out. Not a longer prompt for its own sake — a clearer one, so the first answer is closer to the one you were after.

context degradation, corrected

Hold the thread automatically

As a session grows long, Ruminari notices what has fallen out of context and restores it — without you noticing the repair, and without stopping to re-explain the setup for the fourth time.

The point is not to add a layer of tooling on top of the model. It is to remove the work that sits between you and the answer, so the time goes into the actual task. Do that well enough and the layer stops being something you could take or leave.

in the family

The same idea, turned on the conversation itself.

Studinari holds context for a student across weeks of study. Ruminari holds it across the length of a single conversation. Different surface, same spine, and the same reason neither is easy to walk away from — which is exactly what the Nari ending is supposed to mean.

where it is

In development, not yet open.

Ruminari is being built now. It gets announced when it holds its own thread well enough to deserve the ending — the same bar Studinari had to clear.

No public release date yet. Watch this page, or write to us.

Get in touch All projects