The stage behind your agent draws what the conversation is about. This is its whole spoken vocabulary, generated straight from the stage's own code — 209 phrases across 5 groups — so what you read here is exactly what it can draw today.
The promise: Say any of these and the stage draws it as a full scene; say anything else and it appears as a stylized scene with the word as its label.
That second case is honest by design. An unlisted word first gets an ambient scene carrying the name card
<word> — stylized (say “lion” today and you get lion — stylized) — a label, not a likeness.
Then the stage tries to do better and says exactly what happened: in a live session your agent is asked to sketch the thing from its own
visual memory, and a stage note tells it honestly whether that landed or not; a short list of near-misses falls to a labeled stand-in preset
instead (“wolf” shows the dog, “tulip” the flower). Nothing is dressed up as the real thing. Everything listed below is a full preset scene.
How “saying” works today. On the Semantic Stage, press ● Go live · mic and talk. That needs a microphone and a
live voice line the site owner has switched on — if the line is not awake, the Stage says so and nothing is drawn. While live, your agent hears
every listed word and directs the stage, and direct commands fire on their own without waiting for it: show me a …,
turn into a …, take me to the …, compare X and Y, show me the process, go back.
Without a live line there is no text box yet — you click instead: the Stage's scene chips cover every character and placeable thing on this page,
a run of scenes and the four diagrams, and the scripted tour plays a fixed set.
Every word below is a listed phrase. Mix a scene with a thing, add a feeling or a gesture — the same shapes the stage's own director uses. Your agent hears these and directs the stage; the ones that open with a direct command (“show me”, “turn into”, “become”) also fire on their own.
Landscapes and settings drawn procedurally — say the word, or a phrase that ends in it ("a snowy mountain", "the night sky").
32 registry keys · 32 spoken phrases · from ARCHETYPE_INDEX.procedural
Nothing here matches.
Fleet-authored environments with their own light and weather. Aliases are keys too — "dunes" and "sahara" both open the desert.
24 registry keys · 24 spoken phrases · from ARCHETYPE_INDEX.worlds
Nothing here matches.
Curated figures whose open face zone carries your presenter's living face. Any listed nickname works as a whole word ("meet grandma").
12 registry keys · 44 spoken phrases · from ARCHETYPE_INDEX.figures (+ FIGURE_SYNONYMS)
Nothing here matches.
Ask for a picture of an idea — steps, a side-by-side, a sequence in time, or what leads to what.
4 registry keys · 4 spoken phrases · from ARCHETYPE_INDEX.diagrams
Nothing here matches.
Creatures and objects the stage can put into any scene — or stand alone center-stage. Say them on their own or inside a scene ("a camel walking the desert").
18 registry keys · 105 spoken phrases · from ENTITIES + ENTITY_ALIASES (entity registry; resolves via path 4b — not part of ARCHETYPE_INDEX)
Nothing here matches.
Generated by tools/gen-things-you-can-say.mjs from semantic/sem-archetypes.js
(sha256 30742841afbf93cd7c631ceccaa6bf1373504dc29605c4741588aaeaa0819fab). Dashed chips are nicknames that resolve to the solid key beside them.
Data twin: things-you-can-say.json.