A circular logographic writing system inspired by the heptapod language in Arrival. Words become ink on a ring — no beginning, no end, no linear time.
Live at torus.steven-geller.com
"time is a circle"
Type English text. Torus classifies each word by linguistic category (entity, action, property, relation, particle, negation, question), assigns semantic roles, decomposes words into universal semantic primes, and renders a deterministic circular symbol where visual features encode meaning.
Key properties:
- Order-independent — "time is a circle" and "a circle is time" produce the same symbol. Input order doesn't matter; meaning does.
- No tense — Torus uses aspect (timeless, bounded, unbounded) instead of past/present/future.
- Deterministic — Same words always produce the same symbol.
- Reversible — Symbols can be decoded back to their source text.
- Calligraphic — Brush dynamics modeled on Chinese calligraphy: pressure zones (顿笔), bone structure (骨法), ink texture (墨韵), dry brush (飞白), and taper (收笔).
Each word category produces a distinct mark on the ring:
| Category | Mark |
|---|---|
| Entity (nouns) | Outward tendrils with branching sub-tendrils and satellite splotches |
| Action (verbs) | Stroke-width rhythmic waves |
| Property (adj/adv) | Inner ripples along the edge |
| Relation (prepositions) | Silk-thin bridges with bow |
| Particle (the, and) | Decisive notches |
| Negation (not, never) | Deep inverted voids |
| Question (?, why) | Gap in the ring |
See LANGUAGE.md for the full linguistic system.
┌──────────────────────────────────────────────────┐
│ Axum server (:3031) │
│ │
│ POST /api/generate → symbol geometry JSON │
│ POST /api/generate-svg → SVG with embedded text│
│ POST /api/decode → text from SVG upload │
│ GET /og/{text} → 1200×630 OG image PNG │
│ GET /png/{text} → 600×600 download PNG │
│ GET / → frontend │
│ GET /health → "ok" (monitoring) │
└──────────────────────────────────────────────────┘
| File | Lines | Purpose |
|---|---|---|
src/main.rs |
716 | Web server, API endpoints, OG image rendering (resvg), structured logging |
src/symbol.rs |
768 | Symbol generation engine: polar geometry, brush dynamics, mark rendering |
src/language.rs |
944 | Linguistic engine: word classification, aspect detection, semantic roles, NSM primes |
src/decode.rs |
618 | Symbol recognition: metadata extraction, fingerprint matching, geometric reverse-engineering |
src/word_primes.rs |
968 | Semantic prime decomposition database (500+ words) |
- Classify — Each word gets a category (entity/action/property/...) and semantic role (agent/patient/modifier)
- Sort — Words sorted by role priority + content hash (guarantees order independence)
- Brush pressure — 2–3 Gaussian pressure zones placed on the ring
- Bone structure — Sinusoidal radius modulation creates the ring's overall shape
- Word marks — Category-specific geometry placed at each word's angular position
- Ink texture — Fractional Brownian motion noise for organic variation
- Dry brush — Stochastic stroke gaps at 12–30% intensity
- Taper — Automatic width reduction at stroke extremes
- Embellishments — Accent dots, wisps, inner arcs, branching structures
- SVG render — Polar → Cartesian conversion, dual-path ring (fill-rule evenodd)
Symbols can be decoded back to text through four methods (tried in order):
- Embedded metadata — SVGs from the generate endpoint carry base64-encoded source text
- Fingerprint lookup — Visual fingerprint matched against known symbols database
- Dictionary match — Pre-computed fingerprints for all dictionary words
- Geometric analysis — Detect marks from path geometry, classify by shape, search candidate word combinations
- Rust 1.75+
cargo build --release
./target/release/torusServer starts on 127.0.0.1:3031. For production, put it behind a reverse proxy with TLS.
Set RUST_LOG=info (default) or RUST_LOG=debug for verbose output. The server handles SIGTERM/SIGINT gracefully, draining in-flight requests before shutdown.
[Unit]
Description=Torus logographic language
After=network.target
[Service]
Type=simple
WorkingDirectory=/home/user/torus
ExecStart=/home/user/torus/target/release/torus
Restart=on-failure
[Install]
WantedBy=multi-user.targetcurl -X POST https://torus.steven-geller.com/api/generate \
-H 'Content-Type: application/json' \
-d '{"text": "time is a circle"}'Returns JSON with word analysis, semantic primes, and ring geometry (outer/inner point arrays).
curl -X POST https://torus.steven-geller.com/api/generate-svg \
-H 'Content-Type: application/json' \
-d '{"text": "time is a circle"}'Returns an SVG string with the source text base64-encoded in a data-torus-text attribute.
curl -X POST https://torus.steven-geller.com/api/decode \
-F 'file=@symbol.svg'Returns JSON with decoded text, confidence level, and method used.
curl -O https://torus.steven-geller.com/png/time%20is%20a%20circleReturns a 600×600 PNG on dark background.
- Real-time symbol generation with morphing animation between states
- Word mark labels positioned near their marks with linguistic metadata
- Semantic prime decomposition displayed for each word
- SVG and PNG download
- SVG upload for decoding
- Shuffle button to demonstrate order independence
- Arrival-inspired ambient sound (spacebar or on generate)
- Dark/light mode