Fact Sheet · 2026
VHS

AI for Behavioral Research — multi-agent simulation of realistic human behavior through internal psychological deliberation, dynamic specialist recruitment, and transparent reasoning.

Live at
vhs.skoonline.org
Part of
skoonline.org ecosystem
VHS creates virtual humans with inner lives. Instead of generating responses from a single prompt, it orchestrates 6+ psychological agents — drawn from a library of 3,400+ specialists across 15 parametric dimensions — through multi-round deliberation that produces emergent, human-like behavior. Compose a single VH from a sentence, or assemble entire teams for group consultation sessions. Every response reflects competing drives, emotional tensions, and cognitive trade-offs — the way real people actually think and respond.
Our philosophy

Most AI systems optimize for the best possible answer. VHS is built on a different premise: real humans don't always give the best answer — they give their answer, shaped by competing internal forces. Every design decision flows from six beliefs about what makes simulated behavior genuinely human.

Multi-Agent Deliberation
A person is not one voice — it is a committee that votes
Single-prompt chatbots produce monolithic responses. A Virtual Human has 6+ agents — each with its own personality, activation level, and influence weight — that perceive, debate, form coalitions, and reach consensus across 3 rounds. The response that emerges is not authored; it is negotiated.
Dynamic Composition
The team that answers you changes based on what you ask
A fixed agent roster cannot cover every domain. VHS's Central Executive recruits specialist agents from a gallery of 3,400+ specialists — turn by turn, based on detected gaps. The team that handles your first message may look nothing like the team on your tenth. The VH adapts its internal structure, not just its words.
Transparent Reasoning
You can see exactly what the VH was thinking — and why it chose what it chose
Black-box AI is useless for training and research. "Peek Into the Brain" exposes the full deliberation transcript: which agents spoke, who won, what coalitions formed, and what minority concerns were overridden. The process is as valuable as the output — especially for learners studying human psychology.
Psychological Realism
Flaws are not bugs — they are the entire point
An anxious student whose anxiety overrides their self-efficacy is not malfunctioning — that is the simulation. VHS preserves cognitive biases, emotional overreactions, and irrational defensiveness when they belong to the character. Training on perfect behavior teaches nothing about handling imperfect humans.
Self-Awareness
The VH knows what it is made of — and can tell you
An auto-injected Self-Awareness Monitor maintains real-time knowledge of the current agent roster: who is on the team, why each specialist was recruited, and what the deliberation dynamics look like. Ask "who are your agents?" and you get a truthful, live answer — not a canned description.
Emergent Behavior
The best responses are not designed — they are discovered
No prompt engineer wrote the VH's response. It emerged from agent interaction: empathy pushing for warmth while professional standards demanded boundaries, patience moderating the pace while anxiety injected hesitation. Emergence produces behavior that surprises even the designer.
Theory behind VHS
🧠
Society of Mind
Minsky, 1986
Intelligence emerges from the interaction of many simple agents, not from a single monolithic process.
🎭
Personality Psychology
Big Five · Trait Theory
Agents map to established personality dimensions: traits, states, and regulatory processes that shape human behavior.
⚖️
Deliberative Democracy
Consensus Theory
Multi-round debate with coalition formation produces outcomes that no single participant would have reached alone.
🏗️
Cognitive Architecture
Perception → Deliberation → Action
5-layer processing: appraisal, specialist recruitment, social-relational framing, personality deliberation, and meta-cognitive evaluation.
3,400+
Specialist Agents
in the gallery
15
Psych Dimensions
parametrically editable
900+
Pre-built VHs
across 18 categories
8
Avatar Providers
video & TTS
5
LLM Providers
with auto-fallback
Built for behavioral research — used across disciplines
🔬
Researchers & Behavioral Scientists
Design · Simulate · Analyze · Publish
🎓
Trainers & Educators
Practice · Rehearse · Debrief
  • Practice with simulated anxious students, defiant teens, or confused learners
  • Train clinical communication with virtual patients in denial, anxiety, or non-compliance
  • Rehearse giving feedback, de-escalation, and conflict resolution
  • Build customer service skills at scale — no risk to real people
  • "Peek Into the Brain" to debrief: see which psychological forces shaped each response
  • Choose from 900+ pre-built VHs or compose from scratch in seconds
  • Assemble multi-VH consultation teams for group case discussions
  • Tune 15 psychological dimensions with sliders — adjust difficulty and realism
⚙️
Developers
Integrate · Extend · Embed
  • RESTful API with SSE streaming for real-time responses
  • MCP integration — use VHS from Claude Desktop or any MCP client
  • Embeddable via iframe with full conversation UI
  • Bring your own LLM keys (OpenAI, Anthropic, Azure, DeepSeek, Z.ai)
  • No database required — zero-config in-memory architecture
  • Docker / Cloud Run ready
  • OAuth sign-in (Google, Microsoft, GitHub)
  • Token accounting & cost tracking per user and VH
🏛️
Institutions
Deploy · Manage · Scale
  • Role-based access: user, researcher, admin
  • Gallery management — curate which VHs your users can access
  • Admin dashboard with user management and system prompts
  • Google Cloud Storage for preset persistence and backup
  • Rate limiting and security middleware tiers
  • Multi-language support via hash-based i18n
  • Token usage analytics and cost monitoring
  • Self-host or use managed deployment
Core architecture
🧠
Multi-Agent Deliberation
6+ agents perceive, debate across 3 rounds, form coalitions, and reach consensus to generate each response
🔄
Dynamic Composition Default
Central Executive recruits from 3,400+ specialist agents turn-by-turn based on detected domain gaps
🪞
Self-Awareness New
Auto-injected monitor lets VH truthfully report its own roster, recruitment history, and deliberation dynamics
🏗️
5-Layer Cognition
Appraisal, specialist recruitment, social-relational framing, personality deliberation, meta-cognitive evaluation
📊
Behavior Synthesis
Three modes: dominant-only (psychological realism), coalition-informed (balanced), full-synthesis (best answer)
🎛️
15 Psych Dimensions
Big Five, confidence, stress, motivation, empathy, cognitive flexibility, impulse control, trust, attachment — all parametrically editable
📐
2-2-2 Agent Framework
Every VH has 2 Trait + 2 State + 2 Regulatory agents — ensuring balanced psychological coverage by design
📚
500+ Theoretical Foundations
Agents grounded in established psychological constructs with research citations — not arbitrary prompt engineering
Creation & composition
Compose from Blank
Describe a person in plain language — AI assembles a 6-agent team from the specialist gallery with rationale
🎨
AI Personality Generator
Generate full agent configurations from natural language descriptions — personality prompts, parameters, and theoretical grounding
🔧
Agent Workshop
Create custom specialist agents with configurable parameters, domain sensitivities, and personality prompts
🗂️
900+ Pre-built Gallery
Browse, filter, and launch from 900+ ready-to-use VHs across 18 professional categories — or fork and customize
Multi-VH interaction
🤝
VH-to-VH Simulation
Two virtual humans converse autonomously with intelligent role pairing — help-seeker meets help-provider
📹
Group Consultation New
Zoom-like multi-VH sessions — assemble a panel of virtual experts for group discussion, each with their own deliberation
👥
Team Creation Wizard
Dynamically compose VH teams for any purpose — advisory boards, case conferences, debate panels, consulting groups
🎭
Team Dynamics
Configurable discussion styles, role assignments, and interaction patterns — VHs are aware of what other VHs have said
Visualization & analysis
🔍
Peek Into the Brain
Inspect full deliberation: which agents spoke, who won, coalitions formed, minority concerns, activation levels
🌐
3D Deliberation Replay
Watch agent interactions unfold in 3D space — positions, influence flows, and coalition formation visualized spatially
🧩
Brain Visualization
2D interactive view of agent activation levels, influence weights, and domain sensitivities in real time
🔬
Psychological Analysis
Coherence checking across dimensions, agents, and foundations — coverage gap analysis and validity scoring
Research toolkit
🧪
Experiment Wizard
Multi-step guided creation of VH experiments — hypothesis, population, intervention, parameter config, and launch
🎚️
Parameter Exploration
Real-time parameter adjustment with live deliberation simulation — see how changes affect agent behavior instantly
📋
VH Evaluation
Configurable evaluation rubrics with stopping rules — systematic quality assessment of VH interactions
📊
Researcher Dashboard
Profile management, project tracking, experiment history, publication records, and activity analytics
Avatar & media
🎥
8 Avatar Providers
Azure Speech, HeyGen, D-ID, Simli, Tavus, Anam, Ready Player Me — unified registry with session-based credentials
🗣️
Text-to-Speech
Azure Speech Services integration — the VH speaks its responses aloud with configurable voice profiles
📽️
Presentations Hub
6 structured presentation decks (20–27 slides each) covering VHS architecture, research applications, and investment case
✍️
AI-Assisted Writing
Multi-mode writing (LLM Direct, Expert VH, Team VH) with article templates, non-linear editing, and comparative drafts
How it works
01
Choose or compose
Pick from 900+ pre-built VHs across 18 categories, or describe a person in plain language and let AI compose the agent team.
02
Agents perceive
Your message enters. Domain is detected. All 6+ agents perceive the input in parallel, each through their own psychological lens.
03
Deliberation & recruitment
Agents debate across 3 rounds. Coalitions form. The Central Executive recruits specialists if gaps are detected. Consensus emerges.
04
Response streams
The synthesized response streams token-by-token via SSE. Open "Peek Into the Brain" to see the full deliberation behind it.
Key differentiators
Deliberation
Multi-Agent
Consensus
Standard chatbot

Single prompt → single response → no internal conflict, no competing perspectives, no psychological depth.

VHS

6+ agents debate across 3 rounds → coalitions form → dominant position emerges with minority concerns preserved. The response is negotiated, not authored.

Dynamic
Specialist
Recruitment
Fixed-persona AI

Personality is defined once in a system prompt → cannot adapt internal structure → same capabilities every turn.

VHS Dynamic Mode

Central Executive detects coverage gaps each turn → recruits from 3,400+ specialist gallery → the team that handles turn 1 may differ from turn 10. Internal structure adapts, not just words.

Transparency
Peek Into
the Brain
Black-box AI

Response appears → no way to see what influenced it → useless for training debrief or research analysis.

VHS

Full deliberation transcript exposed: every agent's position, coalition dynamics, activation levels, minority concerns. The process is as valuable as the output.

Realism
Flaws as
Features
Standardized patient / actor

Expensive → limited availability → inconsistent across sessions → cannot scale to many learners simultaneously.

VHS

Anxiety overriding self-efficacy, denial suppressing rational assessment, defensiveness blocking empathy — all reproducible, configurable, and scalable. Flaws are the simulation, not bugs in it.

Teams
Group
Consultation
One-on-one chatbot

Single AI persona → single perspective → no cross-pollination of viewpoints → no group dynamics.

VHS Team Mode

Assemble a panel of VH experts — each with full internal deliberation — that discuss, debate, and build on each other's perspectives in a Zoom-like group session. Multi-VH, multi-mind.

Composition
3,400+
Agents
Manual prompt engineering

Write a system prompt by hand → hope it captures the personality → no theoretical grounding → no validation.

VHS Composition

Describe a person in plain language → AI selects from 3,400+ specialist agents grounded in 500+ psychological constructs → assembles a balanced 6-agent team with rationale → coherence-validated.

Potential applications
AI Safeguard Layer
A deliberative middle layer between humans and raw AI models
Instead of letting users interact directly with a bare LLM, place a multi-agent VH layer in between. Specialist agents act as guardrails — checking for safety, appropriateness, domain accuracy, and emotional sensitivity — before any response reaches the user. The committee deliberates so the raw model doesn't have to be trusted alone.
Professional Training at Scale
Replace expensive standardized patients and actors with reproducible VHs
Teacher training, clinical education, customer service, crisis intervention, soft skills — learners practice with psychologically realistic counterparts that are consistent, configurable, and available 24/7. Debrief with "Peek Into the Brain" to understand why the VH reacted the way it did.
Behavioral Research Platform
Controlled experiments on simulated human behavior — at a fraction of the cost
Manipulate personality variables, run VH-to-VH simulations, analyze deliberation transcripts, and publish — all from one platform. VHS gives researchers the experimental control of a lab with the ecological validity of naturalistic observation. 500+ psychological constructs, 15 parametric dimensions, full reproducibility.
Humanoid Robotics
Give physical robots a psychologically grounded personality layer
VHS provides the cognitive and emotional architecture that humanoid robots need to interact naturally. Instead of scripted responses, the robot's behavior emerges from internal deliberation — making it feel genuinely human rather than programmatically polite.
Multi-Expert Consultation
Assemble virtual advisory panels for complex decisions
Compose teams of VH experts — each with domain specialization and distinct psychological profiles — for group consultation on medical cases, legal questions, business strategy, or policy analysis. Multiple minds, multiple perspectives, one session.
AI-Assisted Research Writing
Write papers with a panel of VH writing experts, not a generic chatbot
Step-by-step article composition with three modes: LLM Direct for speed, Expert VH for domain-specific voice, or Team VH for multi-perspective drafting. Article templates, comparative drafts, section-level goals, and a full journey trail for reproducibility. The AI doesn't write for you — a committee of specialists writes with you.
900+ Virtual Humans across 18 categories
Intelligent Tutoring Rehabilitation Patients Educational Clinical & Therapeutic Healthcare Professionals Workplace & Professional Crisis & Emergency Parenting & Family Marriage & Couples Hospitality & Service Elderly Care Recovery & Addiction Social Services Customer Service Peers & Colleagues Teen Issues Students with Issues Coaching & Personal Dev
Technology
React 18 + Vite
Node.js + Express
Anthropic Claude
OpenAI GPT-4
Azure OpenAI
DeepSeek
Z.ai
SSE Streaming
MCP Integration
xAPI / LRS
Google Cloud Storage
Hash-based i18n
Docker / Cloud Run
Three.js (3D Viz)
Azure Speech TTS
HeyGen · D-ID · Simli
OAuth (Google · MS · GitHub)
JWT Authentication