hirameki

What it does

A tool for scientists: it generated new research ideas and helped manage the literature behind them, using semantic search over a corpus of papers to find connections a keyword search would miss. No client, no route table, no README survives from the app itself, backed by a Postgres database with the vector extension enabled, which means it stored and searched embeddings.

How it is built

A single Go service on :8080, fronted the same way every app on the box is, with pgvector doing the retrieval work underneath. It is the platform's oldest resident: the one app that predates the per-project convention and still sits in AWS's shared default network as a fossil of that.

Where it stands

Its source directory is gone. What's left is one built image, pinned by digest so it can never drift or be overwritten, kept running and archived off-site, and never touched again. The name is 閃き, hirameki: the flash of insight, the moment before an idea becomes anything, the whole point of a tool built to help find one.

This entry was rewritten from Jonathan's own description of what the app actually did; the infrastructure details (source gone, pinned by digest) are unchanged from the earlier draft. Still to come: confirming Status and Stack are still accurate, and however much more he wants to say about what "research ideas" and "literature management" meant in practice.