◇ Prompt Tracer
Every passage in your library becomes a point on a map, placed by meaning, so related ideas settle into topic clusters. Here you're at the top, speaking a question to the model on the terminal below. Watch it reach out into the clusters, pull in the passages that fit, and answer you on screen — then read on for what happens at each step, and how it scales to your real library.
A guided, five-step walk-through — pick your own question or fine-tune it once it's running.
↓ The same five stages, one at a time — no jargon left unexplained
◇ Under the hood
The animation moves quickly. Here's what each step is actually doing between the moment you ask and the moment the answer lands — the whole pipeline the Prompt Tracer lights up, unpacked.
Every passage was turned into a string of numbers — an embedding — that captures what it's about, not just its wording. Your question gets the same treatment, and the engine looks for the passages sitting closest to it. That's how it surfaces the right note even when it shares not one word with what you typed.
At the same time, a classic keyword search runs the old-fashioned way. It matches the exact words you used and rewards passages where those words are rare and telling. It's the safety net: if you searched for a specific name, term, or number, the literal hit is never quietly missed.
Now there are two ranked lists — one by meaning, one by words. They're blended into a single order, so a passage both lists agree on rises to the top and a one-list fluke sinks. You can even lean the blend toward exact words or toward meaning — that's the knob you turn in the Prompt Lab →
The top handful gets a second, closer read. A slower model looks at your question and each passage together — not as two separate fingerprints — and re-sorts them by how well they actually answer you. It's expensive, so it only runs on the finalists, where it earns its keep.
Only the best few passages are handed to the language model, along with your question. It writes an answer grounded in those passages — and every claim carries a citation you can click straight back to the source. Nothing is invented from thin air, and nothing ever leaves your machine.
◇ At full size
What you just drove is a scale model. The map above shows a small sample so the whole thing fits on screen and runs in a browser tab. Point Alambique Index at your own files and the very same machinery runs at full size — on your hardware, with nothing uploaded.
The full Prompt Tracer is the app's Observatory — the same living map, rendered from your real index instead of a sample. It's the whole idea of retrieval made visible, at the scale of everything you've saved. Every dot stays on your machine; nothing is ever uploaded.
