Load a 3D scan of any place. Converse with an AI on a Solari flip-board. Group words by color, draw edges between ideas, and send them into space as annotated nodes — building a shared memory palace one session at a time.
The hippocampus doesn't just store memories — it builds spatial maps. Every memory palace tradition from Simonides to Giordano Bruno exploited this: anchor ideas to places, and the brain retrieves them like a route. Semantic Field makes this literal. Load any real 3D scan, converse with an AI, group the words that matter, connect them with edges, and place them in space — anchored to the geometry that already lives in your body.
Start with a question. End with a memory palace. Here's the path.
The entry point. Type any question, concept, or provocation into the input field below the board. Hit Enter or SEND. The AI responds on the flip-board — letters scrambling into place like an airport departures sign. Response words and recognized phrases drop automatically onto your canvas.
Drop a .ply file
onto the app or use the file button. Exported from Lixel Color Studio, Polycam,
RealityCapture, or any photogrammetry tool. The parser handles both ASCII and binary PLY.
Switch to ⊕ Place Node mode and click anywhere in the scene to drop an annotation node at that 3D position. A modal opens immediately for you to name it and begin annotating.
Click any node to open the Meaning Editor. Each object holds up to 6 named layers — Literal, Cultural, Contested, Speculative, Poetic, Technical. Layer through them like tuning a radio. Let the AI co-evolve any layer.
Words from the board land on the Canvas panel. Group them by color — no meaning required. Add meaning later via double-click. In the scene, use ⌁ Connect mode to draw edges between nodes. Save a world.json and commit it to a shared GitHub repo to sync with collaborators.
git add world.jsongit commit -m "..."git pushgit pullRaw scans always have noise — floating points, background clutter, scan artifacts. Switch to Erase mode and paint over any points to remove them. Each stroke is one undo step.
| Action | Input | Notes |
|---|---|---|
| Orbit scene | Left drag | Horizontal + limited vertical — can't go below ground |
| Pan scene | Right drag or Alt + drag | Moves the orbit pivot point |
| Zoom | Scroll wheel | Exponential — works at any scale |
| Reset view | Double-click canvas | Returns to full-scene overview |
| Move a node | Drag the ring | Slides on camera-aligned plane in 3D |
| Open annotation editor | Click the ring | Short click = open, drag = move |
| Delete a node | Right-click node | Removes node and all its edges |
| Draw an edge | ⌁ Connect → click A → click B | Optional relation label on confirm |
| Delete an edge | Right-click edge line | Immediate, no confirm |
| Add word to canvas | Click word on board | Or automatically from AI responses |
| Group canvas token | Click token → click color swatch | 6 color groups, no type required |
| Label a token | Double-click token | Optional meaning — shown on hover in scene |
| Send canvas to scene | ◈ Send to Scene | Commits canvas as a new graph node |
| Save world | ⟳ Sync → Save world.json | API keys never written to file |
| Load / merge world | ⟳ Sync → Load world.json | Non-destructive merge — newer revision wins |
| Erase points | ◌ Erase mode → paint | Screen-space brush — hides, doesn't delete |
| Undo erase | ↩ Undo button | One undo per stroke, stackable |
The canvas uses color groups as the primary organizing tool — assign any word to a color without needing to name what the color means. Groups can be labeled later, or left as pure visual clusters. When you send a canvas to the scene, the dominant group maps loosely to one of five underlying node types — a scaffold you can override at any time.
Types are a lightweight scaffold — useful for export and querying, but never a gate. Start with color. Name it when it matters.
Semantic Field is designed for iterative return. Each session leaves a world.json that the next session — or next collaborator — picks up from.
.ply.Deep in the medial temporal lobe sits the hippocampus — a seahorse-shaped structure that is the brain's cartographer. It does not merely store memories; it builds cognitive maps. O'Keefe & Nadel's 1978 discovery of "place cells" — neurons that fire precisely when an animal occupies a specific location — established that the hippocampus is a genuine spatial computing engine, earning a Nobel Prize in 2014.1
The maps it builds are not limited to geography. The hippocampus encodes relational structure of any kind: the layout of ideas, the sequence of events, the connections between concepts. Memory palace techniques work precisely because they hijack this spatial machinery — anchoring abstract information to imagined places so the hippocampus can retrieve it like a route.
This plasticity cuts both ways. A 2020 study found that habitual GPS use correlated with a steeper decline in hippocampal-dependent spatial memory over three years — and importantly, the decline was observed even in people who started with strong navigation ability, suggesting GPS use caused the drop rather than selecting for weaker navigators.3
Chronic stress compounds the problem. The hippocampus is densely packed with cortisol receptors. Sustained elevation of cortisol — a hallmark of urban crowding, noise, social precarity, and nature deprivation — suppresses adult neurogenesis in the dentate gyrus and causes dendritic atrophy.4 A study of 156,075 UK Biobank participants found that urban environments characterized by social deprivation, air pollution, and high density were associated with measurable brain volume differences along stress-response pathways.5 A separate analysis of ~1,000 adults found that those living in the most disadvantaged neighborhoods had hippocampal volumes equivalent to seven additional years of age-related atrophy.6
Note: these studies show correlation, not simple causation — genetic selection, socioeconomic confounds, and reverse causality are active areas of debate. The mechanism (chronic stress → cortisol → hippocampal atrophy) is well-established; the urban attribution requires care.
Alfred Korzybski's maxim — coined 1931 — is the permanent disclaimer on any knowledge system. No map is complete. Every annotation carries its author's blind spots, the period's assumptions, the language's limits. But sharing and refining maps between people with different vantage points is the closest collective process we have to becoming less wrong together. The semantic field is not truth — it is a convergence process.
A cipher is a systematic substitution: replace each element of a message with a symbol from a chosen alphabet. Memory palaces work identically — replace each idea with a vivid image placed in a location. The code is spatial. Semantic Field extends this: each node is a cypher unit, its meaning layers the key, its attestations the checksum. A shared knowledge graph is a collectively-legible cipher for a shared world.
The entire app — 3D point cloud engine, PLY parser, Solari board, color canvas, edge graph, meaning layers, and world sync — ships as one self-contained HTML file. No build step. No server. Drop it anywhere. Fork it freely.