Virtual Human System
Most AI gives you one voice.
A person is never just one voice.
The problem
Ask a single model to "act human,"
and you get a smooth surface with no inner life — no competing drives, no second thoughts.
The idea
So VHS gives a virtual human many minds.
6+ psychological agents — empathy, patience, anxiety, professional standards — each arguing its own case, in the first person.
The central abstraction
A virtual human is a triple:
Σ = ⟨ Composition · Parameters · Rules ⟩
who is in the room, how loud each voice is, and how they resolve.
How it works
Perceive → Deliberate → Act.
Agents read the moment in parallel, debate across 3 rounds, and the dominant coalition shapes one reply — about 18 model calls per message.
Why you can trust it
Nothing is hidden.
Every reply ships with the full deliberation transcript, and a self-awareness monitor can always tell you exactly who is in the room.
The numbers
3,436 specialists · 5 providers · 15 dimensions
A recruit-on-demand gallery, a 5-layer cognitive architecture, and one gateway for every model call.
What it can do
Describe a person in a sentence —
meet them a moment later. The team even re-composes itself turn by turn, recruiting specialists as the conversation moves.
Where it's used
Training. Research. Teaching.
Rehearse a hard conversation, run a psychology experiment, or stage a whole team of virtual humans in consultation.
Why it matters
Every drive is explicit, so you can tune it.
Every turn is logged, so you can audit it. The model stays frozen, so the behavior reproduces.
VHS
Virtual humans with an inner life.
The Virtual Human System.