Store a fact, note, or document into persistent long-term memory.

Search persistent long-term memory for relevant past facts and context. A result marked SUPERSEDED was true when it was written and is NOT true now -- do not answer with it, use it only to say what something USED to be. If the result is not the whole answer -- more memories matched than fit, or parts of a multi-part question got nothing -- the reply says so on its last line. Believe that line: it is counted, not estimated.

Every value a fact has held, oldest first, with the current one marked. The last entry is the current value. Use this when the user asks what something used to be, when it changed, or whether it changed at all. A fact that never changed returns a single entry, which is an answer, not an empty result. For a chain whose attribute was not declared with entity, nanomem_search can still rank a superseded value first; nanomem_changes is unfiltered and reports every write in a window.

Answer a question as the memory stood at a past moment. Records written after that moment are not considered, so a value that has since been superseded is returned wherever it was still current. Use this for 'what did I think in March', or to reconstruct past state.

What the memory learned in a time window, oldest first, with no query needed. Use this for 'what did I tell you since last week' or to catch up on what changed while away.

How often each remembered fact actually changes, measured from the memory's own revision log: how many times it has been restated, the typical interval between changes, and how long the current value has stood unconfirmed. Use this to decide which facts are probably stale and worth re-confirming with the user. These are measured statistics, not a prediction.

Return current memory vault size and document count.