Weavestudio is the general-purpose platform for multi-agent simulation. Generate populations, design any world, and watch what unfolds when thousands of agents think at once.
AI agents are the fastest-scaling software category. $52.6B projected by 2030 (Grand View Research). But simulation infrastructure is fragmented. Everyone rebuilds from scratch.
| System | What it does | What it can't |
|---|---|---|
| Concordia DeepMind | Closest architecture. Entity-component agents. | Python-only. Tick-based. No emergence measurement. No composability across domains. |
| Generative Agents Stanford | Proved LLM agents can be believable. 25 agents. | Research demo. Not reusable infrastructure. |
| Project Sid Altera | 1,000 agents in Minecraft formed governments. | Locked to one game engine. No structured emergence tracking. |
| AgentTorch MIT | Scales to millions for population modeling. | Sacrifices individual agent richness. No event-driven time. |
Weavestudio is the general-purpose runtime. One engine, any domain, full reproducibility.
Turn a handful of parameters into thousands of demographically grounded agents, each with archetypes, beliefs, purchasing behavior, and a position in a realistic social graph.
Start with a grounded population. When a new scenario arrives, evolve it. Inject new knowledge. Shift sentiments. Drift demographics forward. Introduce new conditions. The platform mutates the population in place. You never regenerate from scratch.
{
"config": {
"population_name": "Urban US Consumers 2026",
"target_size": 2400,
"seed_intent": "tech-savvy urban professionals, 25-60",
"params": [
{
"name": "household_income",
"param_type": "continuous",
"range_min": 35000,
"range_max": 180000,
"distribution": "skewed_right"
},
{
"name": "tech_comfort",
"param_type": "continuous",
"range_min": 0,
"range_max": 100,
"distribution": "bell_curve"
}
],
"archetypes": [
{ "name": "Early Adopter", "proportion": 0.28, "param_overrides": { "tech_comfort": 88 }},
{ "name": "Skeptic", "proportion": 0.22, "param_overrides": { "tech_comfort": 35 }},
{ "name": "Follower", "proportion": 0.35 },
{ "name": "Loyalist", "proportion": 0.15 }
],
"relationships": [
{ "param_a": "age", "param_b": "tech_comfort", "strength_value": -0.35 }
]
},
"seed": 42,
"generate_names": true
}
// 202 Accepted
// { "task_id": "tsk_f4e2c9a1-7b3d-4c8e-9f2a", "status": "pending" }Declare the rules, the actions, the timing. Simulations run on continuous logical time. Event-driven, deterministic, reproducible to the last decision. Drop a population in and hit play.
{
"name": "jury-trial",
"version": "1.0.0",
"engine": {
"scheduling_policy": "phased",
"default_timing": { "cycle_duration": 5.0, "think_duration": 3.0 }
},
"action_registry": [
{
"id": "present_evidence",
"target": "scope",
"parameters": {
"evidence_type": { "type": "string", "enum": ["witness", "document"] }
},
"effects": [
{ "primitive": "add", "path": "evidence_presented", "value": 1, "apply_to": "scope" }
]
},
{ "id": "object", "target": "entity", "cooldown": 10.0 },
{ "id": "cast_vote", "target": "scope",
"parameters": { "verdict": { "type": "string", "enum": ["guilty", "not_guilty"] }}
}
],
"entities": [
{
"id": "judge",
"entity_type": "agent",
"prompt": "You preside over a criminal trial. Rule on objections.",
"actions": ["rule_objection", "instruct_jury"]
},
{
"id": "juror_1",
"entity_type": "agent",
"prompt": "You are a juror. Weigh credibility and reach a verdict.",
"actions": ["cast_vote"],
"state": { "role": "juror", "confidence_guilty": 0.5 }
}
],
"phases": [
{
"id": "evidence",
"allowed_actions": ["present_evidence", "object"],
"transition": { "condition": "world.evidence_presented > 4", "next_phase": "deliberation" }
},
{
"id": "deliberation",
"allowed_actions": ["cast_vote"],
"rules": [
{
"trigger": "on_condition",
"condition": "world.votes_cast == 12",
"effects": [
{ "primitive": "set", "path": "trial_complete", "value": true, "apply_to": "world" }
]
}
]
}
]
}Not every agent's decisions are worth learning from. Mark the protagonists: the key personas, the senior deciders, the social hubs. Extract only their trajectories. Fine-tune an open-weight model on that targeted signal. The result is a specialist for the task you actually care about.
Each specialist becomes the backend for the next run. The next run produces sharper protagonists. The next specialist is better trained. Compounding.
One runtime. Any domain where decisions matter and data is expensive.
NPCs that form factions, hold grudges, and rewrite the narrative without a single behavior tree.
Hundreds of NPCs interact autonomously. A merchant remembers the player stole from them 30 hours ago. Faction leaders drag a village into a civil war nobody scripted. Fork the world at any quest branch and test how different designs change emergent storylines.
Test your product launch on 500 AI consumers before spending a dollar on distribution.
Generate a population with realistic demographics, brand loyalties, and social influence. Consumers discover your product, share opinions, influence each other, and make purchase decisions. Fork the world: same population, different price point. Compare adoption curves.
Model cognitive architectures at population scale without a single IRB approval.
Agents with distinct cognitive profiles: varying working memory, risk tolerance, social conformity, belief revision rates. Study how beliefs propagate, how consensus emerges, how minority opinions survive. Freeze any agent and interrogate its full reasoning chain.
Any domain where agents make decisions, the runtime runs it.
Lokara is the first product built on the WeaveStudio runtime. It won't be the last.
AI-powered consumer research
Brands run synthetic consumer panels instead of focus groups. Generate a demographically grounded population, simulate product launches, measure sentiment and purchase intent, interview any consumer. All before the first dollar of spend.
The runtime is domain-agnostic. If your problem needs agents making decisions at scale, a WeaveStudio product can be built for it.
Explore partnershipThe window for a general-purpose agent runtime opened in 2024 and won't stay open long.
GPT-4-class intelligence now costs 100x less than two years ago. Agent behavior that was prohibitively expensive to simulate at scale is now economically viable. The cost floor crossed in late 2024.
The agent infrastructure space is a collection of point tools: orchestrators, memory modules, eval harnesses. But no general-purpose simulation runtime. The platform layer is unclaimed.
Every simulation run generates proprietary trajectories. Trajectories train domain-expert models. Better models run sharper simulations. The flywheel compounds, and it only works if you own the runtime.
Drives strategy, client relationships, and go-to-market.
Owns the simulation engine, AI infrastructure, and everything technical.
We're onboarding a small cohort of researchers, simulation engineers, and product teams. Tell us what you want to run.
We read every message. Expect a reply within 48 hours.