Prediction tells one type of future from historical patterns. True simulation lets you test what hasn't happened yet — showing how and why customers, employees, or populations respond to new products, prices, campaigns, and policies.
Explore how behavioral simulation informs consequential decisions across your industry.
Built on the peer-reviewed breakthroughs from Stanford CRFM & HCI that defined foundation models, social simulacra, and 1,000-person interview-grounded Digital Twins.
On the Opportunities and Risks of Foundation Models, Stanford CRFM
Introduced the paradigm of foundation models and established rigorous distributional evaluation and uncertainty calibration as essential capabilities.
Graita tags every simulation and question with a predicted Total Variation Distance (TVD < 0.16 decision-quality threshold) and confidence tier (High, Moderately High, Medium, Low).
Social Simulacra: Creating Populated Prototypes for Social Computing Systems, UIST
Demonstrated how populated multi-agent simulations expose emergent social behaviors and second-order effects before a rule or product is shipped.
Our What-If Engine turns any product, pricing, creative, or policy question into comparable scenarios and shows both what people choose and why.
Generative Agents: Interactive Simulacra of Human Behavior ('Smallville'), UIST
Pioneered the cognitive architecture combining a long-term episodic Memory Stream, weighted Recency + Importance + Relevance retrieval, and recursive reflection.
Every Digital Twin preserves episodic memories across studies and propagates word-of-mouth opinions across a Small-World social graph.
Generative Agent Simulations of 1,000 People, Stanford CRFM & HCI
Interview-grounded agents with 4-domain Expert Reflections reproduced human responses at 85% of human self-retest accuracy and sharply reduced demographic stereotyping.
Every Twin is synthesized with Psychologist, Behavioral Economist, Sociologist, and Demographer reflections plus public AI Interviewer onboarding (/i/:token).
Finetuning LLMs for Human Behavior Prediction in Social Science Experiments, EMNLP
Training across 2.9M human responses from 210 experiments improved held-out behavioral distribution alignment by +26% and modeled the Say-Do Gap between stated survey intent and real action.
Graita reports both Stated Preference (%) and Habit-Adjusted Real-World Action Rate (%) to account for status-quo inertia and switching friction.
Knowing only what people did fails when conditions change; asking only what people say suffers from the Say-Do Gap. Graita constrains both together.
Every Digital Twin is anchored to a realistic monthly income, remaining discretionary wallet, and status-quo habit inertia — reporting both Stated Intent and Habit-Adjusted Real-World Action.
Retrieves episodic memories using Stanford's weighted Recency + Importance + Cosine Relevance formula, combined with reflections from a Psychologist, Behavioral Economist, Sociologist, and Demographer.
Round 0 captures independent reactions. Rounds 1..k propagate opinions across a Small-World social network (family, coworkers, neighbors, online mutuals) to measure word-of-mouth shifts.
Every simulation output and survey question is scored with a predicted Total Variation Distance (TVD) and Decision-Grade probability so you know when to trust a simulation and when to run fieldwork.
Select up to 6 skeptics or champions from any simulation run and open a live Focus Group room, or share a public `/i/:token` interview link with real human customers to compile them into Digital Twins.
Upload your existing CRM, loyalty, or survey spreadsheet to synthesize custom Digital Twins in seconds, and export every agent's verbatim internal monologue and answers as CSV or JSON.
Trigger population experiments programmatically with a workspace API key (`gr_live_...`), stream agent reasoning in real time over Server-Sent Events, and export structured JSON or CSV results.
curl -X POST https://api.graita.com/v1/simulations \
-H "Authorization: Bearer gr_live_9f8a7b6c5d4e3f2a1b0c" \
-H "Content-Type: application/json" \
-d '{
"cohort_id": "coh_global_id_urban",
"title": "Credit Card Reward Structure Conjoint",
"stimulus_type": "financial_offer",
"stimulus_context": "Memilih struktur benefit utama kartu kredit digital untuk profesional perkotaan.",
"enable_social_diffusion": true,
"diffusion_rounds": 1
}'1 credit = 1 Digital Twin response round. Every new workspace starts with a 14-day trial and immediate access to Verified Population Panels.
For product & growth teams testing concepts, copy, and pricing before launch.
For research, marketing, and policy teams running multi-round social diffusion.
Custom proprietary Digital Twin banks calibrated on your CRM & transaction data.