Loka Labs builds the Economic World Model — a simulation infrastructure designed to mathematically verify the economic consequences of systemic decisions before they are deployed.

While traditional macroeconomic models assume identical, static populations, Loka provides the first simulation infrastructure purpose-built for economic systems to scale emergent behavior across millions of heterogeneous agents.

What we simulate: Economic systems — not physical reality, not individual consumer preferences. The "physics" of an economy is different from the physics of the real world: it is driven by incentives, information asymmetry, expectations, game theory, and emergent behavior.

Our infrastructure consists of three core computational layers:

• Layer 1 · Simulation (LLM Agents): Powered by open-source LLMs deployed locally, this layer replicates individual behavior, decision logic, and social interactions at scale, capturing human heterogeneity and irrationality.
• Layer 2 · Constraint (First-Principles Engine): An economic engine enforcing supply/demand, price elasticity, budget constraints, game theory, and market clearing. It introduces strict quantitative rules to prevent simulation drift. LLMs handle behavior; this engine handles the "physics" of the economy.
• Layer 3 · Verification (Formal Methods): A verification framework based on model-checking that mathematically proves boundary conditions and guarantees system properties before any simulated results are acted upon.

• Model-agnostic: Runs with any open-source LLM (Llama, Mistral, Qwen, etc.).
• Privacy-preserving: Fully local, sovereign deployment ensuring data residency.
• Distributed Scale: Architecture designed to scale to million-agent simulations efficiently.

┌─────────────────────────────────────────────────────┐
│               LAYER 3 · VERIFICATION                │
│            Formal Verification Framework            │
│  Mathematically prove economic properties hold      │
│  before results are acted upon                      │
├─────────────────────────────────────────────────────┤
│                 LAYER 2 · CONSTRAINT                │
│           First-Principles Economic Engine          │
│  Supply/demand · price elasticity · market          │
│  clearing · budget constraints · game theory        │
├─────────────────────────────────────────────────────┤
│                 LAYER 1 · SIMULATION                │
│             LLM-Powered Economic Agents             │
│  Individual behavior · decision logic ·             │
│  social interaction · heterogeneity                 │
└─────────────────────────────────────────────────────┘

Others stop at Layer 1. Loka Labs has all three.

01 · Technical depth no one else has Formal verification (Sifakis, 2007 Turing Award) + distributed AI/ML systems (Jialin Li, OSDI & NSDI Best Paper Award winner). This combination exists nowhere else in a single founding team.

02 · Full-stack simulation, not just LLM agents Three layers: behavior simulation (LLM) + economic constraints (first-principles engine) + formal verification. Others only have the first layer. The constraint and verification layers are what make results trustworthy enough for sovereign clients to act on.

03 · Sovereign-first architecture Designed from day one for local deployment. Data never crosses borders. Model-agnostic — clients use whatever open-source LLM fits their jurisdiction and compliance requirements. This is a hard architectural requirement, not a feature.

04 · Right market, right moment New economic systems are being built from zero. Charter cities, AI jurisdictions, sovereign economic zones have no historical data and no existing tools. We are not competing for a share of an existing market — we are defining a new one.

We start where policy risk is highest and experimentation is allowed.

• Wedge · Now — AI Regulatory Sandboxes: The smallest, fastest-moving sovereign clients. High uncertainty drives simulation demand. Self-governing — no legacy procurement cycles.
  • Frontier Tower, San Francisco — sovereign commercial entity
  • Utah AI Sandbox — first US AI regulatory sandbox
  • Oklahoma AI Sandbox — legislation in progress
• Expansion · 6–12 Months — Charter Cities & Allied Markets:
  • Próspera, Honduras — AI economic law verification pilot
  • Singapore MAS — via NUS academic channel
  • Southeast Asia economic zones
• Scale · 12 Months+ — Sovereign Wealth & Gulf Markets:
  • Saudi NEOM / Vision 2030
  • UAE regulatory sandbox
  • Global sovereign funds

Each pilot becomes the case study that wins the next client.

We simulate the economic impact of decisions before they are deployed.

• Input: Policy and regulatory design: tax changes, subsidy structures, AI jurisdiction rules, trade incentives, spending programs.
• Output:
  • Agent-level behavior simulation
  • Market-level outcomes (prices, employment, capital flows, GDP projections)
  • Risk distribution across scenarios
  • Failure mode identification
  • Policy comparison (Option A vs. Option B)
• Delivery:
  • 4–6 week deployment
  • Fully local compute — sovereign-compliant
  • Interactive dashboard and simulation reports
  • Model-agnostic (any open-source LLM)

Loka Labs serves sovereign clients with high contract value and structural lock-in.

• Phase 1 · Pilot Projects: Bespoke simulation for a specific policy question. Delivered in 4–6 weeks on local infrastructure. First revenue on contract signature. Purpose: build proof points, establish trust, create case studies.
• Phase 2 · Local Deployment License: Government acquires perpetual local deployment rights plus annual maintenance. Data sovereignty guaranteed. Once deployed, the switching cost is extreme — the simulation becomes embedded in institutional decision-making.
• Phase 3 · Ecosystem Platform: Developer tools, API access for research institutions, modular licensing for specific simulation components. Long-term compounding revenue as the platform becomes standard infrastructure.

The structural logic: One client signed = multi-year lock-in. One pilot = the next client's proof point. Local deployment = natural moat.

• Próspera · Honduras
  • Use case: AI economic regulation simulation
  • Stage: In scoping discussions for verification pilot
  • Contact: Direct engagement with legislation team
• Frontier Tower · San Francisco
  • Use case: Sovereign commercial entity economic simulation
  • Stage: Relationship established, scoping in progress
• Honduras AI Legislation
  • Use case: First live policy simulation case
  • Stage: Direct contact with team drafting AI economic law

• Taylor Swift Eras Tour (in development): Economic impact simulation vs. verified city-level data. Demonstrates model accuracy against public, verifiable outcomes.
• 2026 FIFA World Cup (prediction live before tournament): Pre-tournament economic impact prediction for host cities across the US, Mexico, and Canada. Results verifiable post-tournament — a public, timestamped accuracy test.

Joseph Sifakis and Jialin Li are co-authoring a formal verification paper for economic simulation systems — the academic anchor for Loka Labs' technical approach.

• Lora Li — Co-founder & CEO (Parsons School of Design · Harvard University): Led strategic finance for a leading Ethereum L2, backed by Sequoia China, Tencent, Lightspeed, and Vitalik Buterin. Rare founder combining design thinking, capital markets judgment, and deep tech conviction.
• Akasha Zhu — Co-founder & CRO (Tsinghua University · MIT): Founded Vinci (AI headphone). Deep academic network spanning AI and economics. Bridges research and real-world application.
• Jialin Li — Co-founder & CTO (NUS School of Computing): OSDI and NSDI Best Paper Award winner. Published research: PipeOffload (ICML '25), TAPAS (ICPP '25), Cortex and Libra (NSDI '26). Research focus: distributed AI/ML systems and decentralized infrastructure — the exact technical stack Loka Labs is built on.
• Joseph Sifakis — Chief Scientist (2007 Turing Award): Pioneer of model checking — the mathematical verification method now used in Airbus, NASA, and safety-critical systems worldwide. Actively publishing on World Model safety (IEEE ISSREW 2024, co-authored with Huawei RAMS Lab and VERIMAG). Research credibility is our moat: it is why sovereign clients trust us with their most sensitive economic data.

Joseph Sifakis · Chief Scientist

• World Models: The Safety Perspective (IEEE ISSREW 2024 · Zeng, Zhang, Liu, Sifakis et al. · Huawei RAMS Lab & VERIMAG): Identifies four structural safety failures in World Models: out-of-distribution generalization failure, hallucination propagation, world instability, and inability to formally verify safety properties. Proposes the research agenda Loka Labs is executing — ODD formalization, uncertainty quantification, neuro-symbolic hybrid architectures, and runtime monitoring — mapped directly to economic simulation. This is not a retrofitted application of Sifakis's work. The research agenda was defined for exactly this class of problem.

Jialin Li · Co-founder & CTO · Selected Publications

• PipeOffload (ICML 2025): Memory optimization for pipeline parallelism in large neural networks. Directly enables large-scale economic agent simulation without prohibitive memory overhead.
• TAPAS (ICPP 2025): Automatic derivation of tensor parallel strategies for large neural networks — 160× speedup. Core to running heterogeneous agent populations at scale.
• Cortex (NSDI 2026): Low-latency LLM access via semantic-aware knowledge caching. Core to efficient local deployment of the simulation engine.
• Libra (NSDI 2026): Flexible request partitioning and scheduling for unbalanced and dynamic LLM workloads. Handles the heterogeneous, time-varying computation patterns of economic agent simulation.

This repository contains the core components of the Loka Network. We welcome contributions from the community!

1. Fork the repository
2. Create a feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

Detailed development documentation can be found in our Documentation.