How Neurons Work

Everything you say becomes a node connected to others. Behind the simplicity of voice, there's a memory architecture designed to connect, retrieve, and evolve.

What is a neuron?

A "neuron" is a typed memory unit: a reminder, a task, a memo, a fact, or a contact. Each carries free content, metadata (a checklist, a number, an address, sources), and an embedding vector that captures its meaning. This vector is key: it allows two memories to be compared not by their exact words, but by what they mean.

Nodes connected by typed edges

Neurons don't live in isolation. Typed edges connect them: mentioned with, part of, met at, talks about... This graph structure is what distinguishes a true memory from a simple list of notes. It allows reasoning over time and connecting distant information, like the person you met somewhere, the project they told you about, and the reminder it triggered.

Find without organizing

Search combines two forces: semantic similarity (vectors, for meaning) and exact filters (type, people, dates). This combination allows you to answer a vague request, like "what was that restaurant Marc was talking about again?", without you ever having created a "restaurants" card. You didn't organize anything; memory knew where to look.

An evolving memory

A memory isn't set in stone once and for all. When information changes, the old neuron isn't overwritten: it's marked (historical, resolved, erroneous), a new neuron is created, and an edge connects the two. The timeline remains intact. It's a living memory that corrects itself without ever losing its past.