Skip to content
METATRON MIND

[ HOW IT WORKS ]

How it works

A question does not go straight to a language model. The engine looks in the store first, gathers the passages closest to what you asked, and hands on only those close enough to stand behind an answer.

Ask the engine → Create an account

The store

Everything the engine knows comes from documents read into it. Each subject keeps a dictionary of its own, and a document goes in whole rather than being scattered across many. Inside a dictionary, every passage stays with the section and document it came from — which is what lets an answer name its source instead of simply asserting.

A question is looked for across every dictionary at once, and the closest passages win on one measure only: how near each one is to the question. Popularity and earlier questions play no part.

The harness

The harness sits between the store and the writer, and decides what an answer may contain. It gathers the closest passages, checks whether they are close enough to stand behind the answer, and chooses what the writer is handed. When the store runs out, it can search the web; when you are signed in, it can also use the memory you have chosen to keep.

The writer

The writer puts the answer into words. What it is handed, and the note on where it came from, are the engine's. The writer does not invent the sources, and the answer carries its provenance.

What you can do

The honest limit

An engine that shows its sources can still be wrong about what a source means. The store makes the evidence inspectable; it does not make interpretation infallible. Treat an answer as a pointer to the passages, not as a verdict.

Try it

The explanation ends here. The engine does not.