◀ VibeCruiser

forgetting

A free generative art piece — play it in your browser, no install. an AI agent's memory, replayed inside a real human brain: vLLM's KV cache as points of light on a real cortex

by @dundysm · dundysm/forgetting on GitHubLANG JavaScriptHOSTED ON DUNDYSM.GITHUB.IOUPDATED 14 Oct 2026

Vote and save it — free account with email or sign in with GitHub

Made this, @dundysm? See how many people played it → or use email

Controls: Space

▶ Play full screen ▲ Vote: free account with email ▲ Or sign in with GitHub

GitHub may ask you to sign in, then to let VibeCruiser “Verify your GitHub identity” and “Act on your behalf”. We only read your profile and email, and never post anything.

Open the live demo Source on GitHub

From the README

An AI agent's memory, replayed inside a real human brain.

Every point of light is one token slot of vLLM's KV cache (the GPU memory an LLM server keeps of each conversation), replaying a real benchmark run event by event. The brain is a real person's cortex, from MRI.

Live: dundysm.github.io/forgetting. How it was made, with the encoding table and validation: about.

At 8 and 10 GiB, each miss is one 16-token block: the one holding the end of the previous prompt plus the model's reply. In those requests, the next scripted prompt happened to start with the same tokens the model had generated.

Read the full README on GitHub →

Built something like this? Submit your vibe — free account, takes a minute. How listing works.

More like this

Made this?

Claim it: create a free account, then press CLAIM THIS in the cruiser. Put the badge below in your README or on your demo page first; that link is how we know it's yours.

Your project is on the cruise, loaded live from your own site. Put the badge in your README or on the demo page: it shows your live play count, it links back to this page, and everyone who lands here is one click from running the real thing.

On VibeCruiser

How listing, the badge and claiming work →