A project or knowledge area you work in (e.g. "frontend", "devops"). Domains are the large hub nodes — everything else orbits around them. Size reflects how many sessions you've had in that domain.

The typical way you start working in a domain — recurring first-message patterns clustered from your session history. For example, "fix bug in auth" or "add API endpoint" are entry points if you frequently begin sessions that way.

A recurring behavioral pattern detected across your sessions — keyword sequences that appear repeatedly. These reveal your habitual workflows, e.g. "read → grep → edit → test" or specific technical terms you consistently use together.

A Claude Code tool you frequently use in this domain (Read, Edit, Grep, Bash, etc.). Size reflects usage ratio — larger means you use it more often. Shows your preferred workflow style per domain.

A behavioral feature learned by sparse dictionary analysis of your session patterns. These are abstract dimensions like "exploration depth" or "refactoring tendency" — computed from 27 behavioral signals, not named by you.

A piece of information stored by JARVIS — decisions, facts, errors, or context you asked it to remember. Memories have heat (how recently/frequently accessed) and importance. They cool over time and compress from full text to summaries.

A named thing extracted from memories — functions, files, dependencies, technologies, errors, or decisions. Entities form the nodes of the knowledge graph, connected by typed relationships like "imports", "calls", or "resolved_by".

A cross-domain connection — shared patterns, vocabulary, or structural links between two different domains. Bridges show how your knowledge transfers across projects.

Two entities that frequently appear together in the same memories. The more often they co-occur, the stronger (brighter) the connection.

A cause-effect relationship between entities — e.g. an error caused by a dependency, or a bug resolved by a fix. Red edges indicate causal chains.

Code-level relationships — one module importing another, or one function calling another. Extracted from memory content automatically.

Memories with emotional charge get colored by their dominant emotion: red = urgency, dark red = frustration, green = satisfaction, amber = discovery, purple = confusion. Emotional memories have pulsing rings — faster pulse = higher arousal. Based on amygdala-hippocampal coupling (Wang & Bhatt 2024).

Four chemical channels modulate how strongly memories are encoded: Dopamine (reward from fixing errors/passing tests), Norepinephrine (arousal from error density), Acetylcholine (novelty enhancement), Serotonin (exploration vs exploitation balance). These affect node size and glow intensity.

LTP: entity edges strengthen when both entities co-appear in memories (Hebbian learning). LTD: inactive edges weaken over time. STDP: when entity A consistently appears before B, the A→B direction strengthens — learning causal structure from temporal ordering.

During consolidation, weak edges (low weight + stale + cold endpoints) get eliminated, and orphaned entities get archived. Like brain microglia that prune unused synapses to keep the network efficient.

When a strong (important) memory is stored, older weak memories that share entities with it get retroactively promoted — their importance and heat are boosted. Like protein synthesis tags in biology (Frey & Morris 1997).

Node size — reflects importance: session count for domains, frequency for patterns, heat+importance×emotional boost for memories, usage ratio for tools.

Node glow — intensity reflects heat (recency). Hot nodes glow brighter, cold nodes are dim.

Edge brightness — reflects relationship weight/strength. Stronger connections are brighter.

Flow particles — animated dots traveling along edges, representing active information flow between connected nodes.

Cluster shells — translucent spheres grouping nodes by domain. Visible when zoomed out (L1/L2 zoom levels).

Click node — select it, see details + connections, camera flies to it
Click background — deselect
Scroll — zoom in/out (zoom level changes: Neural → Constellation → Galaxy)
Drag — orbit camera
M — toggle monitoring log
R — reset camera
Esc — close panels