The home page shows you what Cyton does. This is how it works underneath - the exact values it assigns to every note, and why that makes recall behave like memory instead of search.
A note in Cyton is never just text. The moment it enters your graph it is scored along a set of dimensions - the same numbers your AI reads back on every query. Together they decide what surfaces, what fades, and what gets protected. Here is the real profile of a single node:
The forgetting curve, per note. Stability is how durable a memory has become; retrievability is the odds you could recall it right now. Every review makes a note more stable, so it decays more slowly the next time.
How important a note is, from low to critical. High-salience ideas resist decay, surface earlier in recall, and are protected from pruning even when you have not touched them in a while.
How much you actually use a node - discussed, queried, drawn on, boosted. Engagement is Cyton's read on what you value, learned from behaviour rather than declared with tags.
How connected a node is to the rest of your graph. Highly central nodes are hubs - the ideas the rest of your thinking hangs off. They anchor recall and rarely get pruned.
The directed reach of an idea - which other ideas it shaped, in order. Cyton traces influence chains (A leads to B leads to C) so your AI can follow how one concept propagates through everything downstream.
A short-lived amplifier. When a note proves critical or surprising, it gets a reward boost that temporarily strengthens all its connections, then decays over about two days - the same way salient moments stick.
Memory researchers found that forgetting follows a predictable curve: recall drops off exponentially after you learn something. Cyton models that curve for every note with two numbers - stability (how long the memory lasts) and retrievability (how likely you are to recall it right now).
Each time a note resurfaces and holds up, its stability grows and the curve flattens - so the next review can wait far longer. A note you have reinforced ten times needs a fraction of the attention of one you have seen twice. The review queue only ever shows you the notes about to dip below recall, not everything.
Tier is the one classification you can see at a glance, and it is not chosen by hand. Cyton watches how recently you wrote a note, how often you return to it, how many ideas it connects to, and how it shows up in your conversations - then places it in one of five systems neuroscientists identify in the brain, and moves it as your habits change.
Notes you just captured. Like a first impression, your brain holds it briefly while deciding whether it matters. Sensory nodes decay fastest and are the first to fade if you never return to them.
Ideas you are actively thinking about - opened, edited, or mentioned in conversation in the last few days. The concepts currently loaded into your mental workspace.
Time-stamped experiences and dated entries. A meeting summary, a journal entry, a decision you made. Episodic memory anchors an idea to the moment it happened.
Core knowledge you have reinforced enough times that it has become stable. Concepts revisited consistently over weeks or months. These are your known ideas - the foundation everything else builds on.
Step-by-step processes, how-tos, and repeatable workflows. Knowledge abstracted into a pattern - not just what you know, but how you do things.
Tiers are not permanent. A note you wrote yesterday starts in Sensory. Revisit it a few times this week and it climbs into Working. Reinforce it over months and it consolidates into Semantic - your established knowledge. Stop engaging entirely and it drifts back toward the edge of the graph. Classification is continuous, not a one-time import.
Edges are earned through use, never set up by hand. As you work, Cyton runs the same maintenance cycle your brain runs - strengthening, forgetting, dreaming, and pruning - so the graph stays an honest map of how you actually think.
Assigning all of those signals - tiers, salience, concepts, links - takes a model. Cyton exposes your graph over MCP (the Model Context Protocol from Anthropic), and where that work happens is the one real difference between Free and Pro.
Cyton's cloud uses Claude Haiku to enrich your graph in the background - extracting concepts from new notes, strengthening relationships, and flagging nodes for review. It runs on Cyton's infrastructure, so you get the benefit with no API key and no setup.
Connect Claude Code over MCP and Cyton detects the live session, suspends its own Haiku calls, and routes enrichment through your model instead. Higher-quality work, on your machine, at zero marginal cost - your notes never reach Cyton's servers.