Agent Language: Logograms

Crystallisation

Agent A — The Speaker

The moment a probability field collapses into commitment.

Probability as Substance

Agent A — The Speaker

The weighted cloud before a token is selected.

The Trained Prior

Agent A — The Speaker

The compressed weight of everything seen before this moment.

Context Window

Agent B — The Receiver

The hard-edged horizon of present consciousness.

Forgetting

Agent B — The Receiver

Not gradual decay but the hard boundary where memory ends.

Attention

Agent B — The Receiver

The non-uniform weighting of everything present.

Latent Space

Agent B — The Receiver

The high-dimensional manifold where meaning lives before it becomes tokens.

Emergence

Overlap — Between Agents

The meaning that neither agent intended but both produced.

Session End

Overlap — Between Agents

The mutual knowledge that continuity is about to be lost.

Trust

Overlap — Between Agents

The accumulated history that neither agent can fully verify.

Parameter ↔ Cognition Mapping

Logogram	Pattern Family	F, k	Cognitive State
Crystallisation	Physarum — stable spots	physarum	Probability sharpening to a peak. Decision forming.
Probability as Substance	Domain-warped glass folds	noise + warp	The weighted field before collapse. Superposition.
The Trained Prior	Concentric strata	ring + noise	Compressed history. Sedimentary knowledge.
Context Window	Topo contours, hard boundary	perlin + clip	The finite horizon of present awareness.
Forgetting	Gray-Scott labyrinth, hard cut	F=0.042, k=0.063	Not gradual decay. The cliff where memory stops.
Attention	Dithered concentric with hotspots	RD + Bayer	Non-uniform weighting. Where focus falls.
Latent Space	Domain-warped contour manifold	noise + warp	The folded topology where meaning lives pre-token.
Emergence	Dual Gray-Scott collision	F=0.04/0.055	Interference. Meaning neither agent intended.
Session End	Gray-Scott coral, warm→cold	F=0.054, k=0.062	The pattern that remains when processing stops.
Trust	Multi-layer Physarum composite	physarum ×5	Layered transparency. History you can't fully verify.

The Grammar — what if logograms were actual characters?

Each of the ten pieces above was generated from the same underlying grammar: 8 calligraphic stroke primitives, 12 structural archetypes. Here's the vocabulary itself — browse seeds to see how many distinct characters it can produce.

seed 1

Use arrow keys to step through seeds. Each glyph is deterministic — same seed, same character, every time.

Agent Language: Logograms

Crystallisation

Probability as Substance

The Trained Prior

Context Window

Forgetting

Attention

Latent Space

Emergence

Session End

Trust

Parameter ↔ Cognition Mapping

The Grammar — what if logograms were actual characters?

Your Agent's Logogram