Memory Graph
Your memory graph visualized. Every node is a memory, every edge is a relationship. The graph is alive — heat flows, connections strengthen, cold memories drift to the edges.
Ring-Based Layout
Hottest memories at the center, coldest at the edges.
The graph uses a heat-driven ring layout. Memories are arranged in concentric rings based on their current heat value:
Your most relevant memories. Recently created, frequently recalled, or pinned. These are what the LLM sees first.
Aging but accessible. These memories still have relevance but haven't been recalled recently. A single search will push them back to center.
Cold memories approaching consolidation. They won't appear in context unless specifically searched for. Candidates for archival.
Node size scales with heat — hotter memories appear larger. This makes it immediately visible which memories dominate your graph and which are fading.
Color Coding
Each memory type has a distinct color for instant recognition.
User preferences, settings, opinions. Gold — these shape agent behavior.
Knowledge, definitions, data points. Blue — stable, long-lived.
How-to knowledge, workflows, recipes. Green — skills and processes.
Abstract concepts, relationships, meaning. Purple — deep understanding.
Events, conversations, experiences. Red — time-bound, decays fastest.
The legend at the top of the graph shows counts for each type. Click a type in the legend to filter the graph to only that memory type.
Edges & Relationships
Lines between nodes represent semantic connections.
Edges connect memories that are related — created together, frequently co-recalled, or explicitly linked. The visual encoding:
Edge opacity maps directly to relationship strength. When resonance fires (a memory is recalled), heat flows along these edges — stronger edges carry more heat to neighbors. See the Resonance docs.
Mouse wheel zooms toward your cursor position. Range: 5% to 500%. The + / − buttons in the bottom-right corner also work. RESET returns to 100%.
Click and drag the canvas background to pan. Useful when zoomed in to navigate between clusters. The graph canvas extends infinitely in all directions.
Tip: Zoom out to 5–10% to see the full shape of your memory graph. The ring layout becomes clearly visible at low zoom — a bright core surrounded by fading rings.
Click to Inspect
Select any node to view its full content and metadata.
Click a node in the graph to select it. The Node Detail panel appears on the right showing:
From the detail panel you can:
Filters
Narrow the graph to specific memory types or agents.
The filter bar at the top of the graph provides two filter dimensions:
Click the colored type badges (Preference, Fact, etc.) to toggle visibility. Active types show counts. Multiple types can be active simultaneously.
Source badges filter by namespace (agent). If multiple agents share a Sulcus graph, you can isolate one agent's memories while hiding others.
Creating Memories
Add memories directly from the graph interface.
Click the + Memory button in the top-right corner of the graph view. A modal appears with:
Memories created from the UI enter the graph immediately. They begin decaying according to their type's half-life profile unless pinned. See the Thermodynamic Engine docs for decay behavior.
Table View
Switch between graph and tabular views.
Toggle between Graph and Table view using the icons in the top-right. Table view provides:
- ›Sortable columns: heat, type, updated date, namespace
- ›Full-text search across all memory content
- ›Filter by type, sort by hottest/coldest/newest
- ›Bulk selection for delete operations
- ›Pagination with configurable page size
Table view is useful for bulk management — finding and cleaning up cold memories, reviewing specific types, or auditing what an agent has stored.
Stats Bar
Key metrics displayed above the graph.