A Tiered Memory

Remembering isn't just piling things up. A true memory chooses what stays close at hand and what gets filed away. Here's how we taught XNeuronal to make that distinction on its own.

The Problem with a Flat Memory

Until now, every neuron (a reminder, a memo, a fact) lived on the same level as all the others. Convenient at first, but a memory that treats an important birthday and a passing remark the same way eventually gets lost in its own volume. The more you talk to XNeuronal, the more the background noise grows, and the harder it becomes to bring up what really matters without you having to ask for it again.

The answer isn't to remember less. It's to remember at different levels of availability, exactly like a human memory doesn't process a significant memory and an insignificant detail in the same way.

Three Tiers, a Dynamic Hierarchy

We've introduced three tiers: core (the essentials, always ready to be injected into the conversation), structural (solid context, mobilized when the topic comes up again), and archive (the rest, accessible but silent). A neuron doesn't stay fixed in its tier: information mentioned often, recently, or marked as important moves up to the core; what's no longer mentioned slowly moves down to the archive. A nightly pass rebalances everything with a bit of deliberate inertia. A memory doesn't switch tiers based on a single isolated signal; there needs to be a confirmed trend, otherwise the classification would flutter from one night to the next for no reason.

This hierarchy directly inspired how XNeuronal stores and retrieves what you entrust to it: it's the heart of its neural memory.

Waiting until the next day to reclassify a fresh memory left a shaky window open for a long time: information that had just arrived and was already very important temporarily remained drowned with the rest until the next recalculation. So, we made sure that every write (a creation, a reinforcement, an update) immediately triggers a reclassification of that specific neuron, without waiting for the nightly pass. The latter retains its role, but a more focused one: it shifts from a daily rhythm to a rhythm every six hours, to catch edge effects that a piecemeal reclassification alone cannot see, such as information that has suddenly become central because many other things are linked to it. In the constellation view, these crossroads-memories are also visually distinct: the more a neuron connects other memories, the larger its light point on the screen, a silent marker of what truly structures your memory.

The Inbox: What's Waiting to Be Seen

In parallel with the tiers, we've added an inbox. A reminder that comes in while you're sleeping, a daily event that deserves a summary upon waking: rather than letting this information blend into the flow, it lands in the inbox, with a discreet counter in the header. One tap, and XNeuronal reads them aloud to you: a way to pick up where you left off without having to search.

What "Organizing" Means Here

Organizing, for a neural memory, has never meant filing into folders. It means knowing, at all times, what deserves to be kept close at hand. That's also why a neuron deemed important can no longer be left by default in the middle of the pack: at the time of creation, XNeuronal must now estimate how much this memory matters, rather than putting everything on the same average level for convenience. It's an additional constraint in its reasoning with each turn of conversation. But it's the price for a growing memory to remain legible, and not a pile that gets heavier without ever sorting itself.

The compromise we accept: this hierarchy adds a layer of computation to every write. We prefer this cost, invisible to you, rather than letting memory flatten over time and become less useful as it grows.