Today's systems accumulate. engram remembers, holding what matters, composing it on demand, and returning it whole.
One memory across text, image, sound and signal. What is learned in one form is available in every other, with no translation step in between. A phrase can surface an image. A sound can recall a document. A diagram can answer a question first asked in words.
Assembles the context a moment needs, not a heap to sift through. Rather than returning everything that matches, it returns what belongs together, shaped to the question asked. What arrives is already relevant, and already in order.
Returns not only what was true, but what was believed, and when. Every fact carries its own history: how it entered, how it changed, and what it displaced. You can ask the memory to answer as it stood at any earlier moment, not only as it stands today.
Accounts for what it holds, how sure it is, and where it is blank. It can tell what it knows from what it merely assumes. When a question falls outside its reach, it says so, rather than inventing an answer. A memory that can report on itself is one you can build upon.
Reorganizes, abstracts and lets go over time. Left running, it draws out the patterns that recur and retires the detail that no longer earns its place. What remains grows more useful, not merely larger. The memory you return to is better than the one you left.
Two clocks, native to the substrate: when a thing was true, and when it was known. Most memories collapse the two, and lose the difference the instant they write. Keeping them apart lets engram answer a harder class of question: not just what happened, but what was understood at the time. It can revisit an earlier belief without pretending the present never arrived. Time is not metadata here. It is part of what the memory is.
Provenance, confidence and doubt are first-class here, not bolted on after the fact. The memory knows where each thing came from, how far it can be trusted, and what it has no basis to claim. Most systems store answers. engram also holds its own standing toward them: the difference between a memory that recites and one that can be reasoned with.
One substrate for every modality, not a stack of separate indexes stitched together. The usual practice keeps text in one place and images in another, then hopes the seams hold. Here there are no seams: a single space where every form of knowing lives together and relates directly. Meaning is not confined to the medium it arrived in.
Built to live across many minds as one shared memory, not per-agent silos that sync after the fact. Distribution is the starting assumption, not a feature added once a single store runs out of room. Many systems can draw on the same memory at once, and what one learns, the rest can hold.
Recall is composition, not the store-and-fetch that has defined memory to date. An answer is assembled at the moment of asking, from everything the memory holds, rather than read back from where it was filed. This is closer to how a mind remembers than to how a database reads. It means a memory that can meet a question it never filed an answer for.