How letapis is worked with

Letapis reads everything you work with — code, documents, the record of decisions — and holds it as one corpus where every piece has an address and a revision.

Index once, and every question after that is asked of the same corpus, not of whichever directory happened to be open.

Ask by meaning — “where does the mail go out when an order is confirmed” — and it leads you to the place even when the code contains neither “mail” nor “confirmed”.

Use it on unfamiliar ground: you know what the system does, not what its authors called it.

Name a symbol and it shows who calls it, file and line, read from disk today rather than from an old snapshot.

Ask this before you change or remove anything: it tells you what breaks, on the branch you have checked out right now.

Ask for “all of them” and it says plainly that a ranked list is cut short by construction, then walks the corpus end to end to return not the best eight but all thirty-two.

When the answer is a count — an audit, a rename, a class of defect — the best few results are worth nothing.

When the parts pile up, it draws the map from above: what connects to what, where each thing comes from, and where you have not yet looked.

For when the facts are in and the picture is not. Before the first edit to a mechanism you have never met, or once the details stop fitting in your head.

And the one thing neither grep nor a model’s memory has: it remembers what you decided yesterday, and says so before you ask a colleague.

Worth a minute before reopening a settled question. It says whether it was settled, when, and on what grounds.