Company

Make worlds runnable.

Enreef builds the simulation layer for living worlds: persistent, branchable simulations populated by synthetic people, businesses, institutions, infrastructure, and AI agents. Compile a place that exists — or create a world that never has — and run both on one engine.

The problem

Consequential decisions are made with fragmented tools: GIS describes where things are, spreadsheets describe past measurements, traffic models simulate transportation, surveys estimate opinions, and LLMs reason conversationally without a grounded, persistent society. But real outcomes are coupled — housing changes affect commutes, budgets, retail, schools, transit, and migration. Enreef's thesis is that a shared world state, where these systems interact, is now buildable.

Why now

AI can supply behavior without owning the simulation

Language models help with planning, communication, and unusual decisions. They should not replace deterministic constraints, statistics, economics, or routing where those are better tools. The architectural idea is hybrid intelligence.

Large populations no longer need a model call per tick

Event-driven simulation, precomputed behavior, cached decisions, offline policy compilation, and selective frontier reasoning make persistent city-scale populations tractable.

Real places are becoming compilable

Open geospatial standards, Overture and OSM, census data, transit feeds, and local-government GIS make a reproducible World Compiler practical for the first time.

Created worlds can be structured, not scripted

A creator can describe population, geography, institutions, and rules, then inspect and edit the compiled specification — instead of hand-authoring every character and event.

Product principles

One engine, many worlds — grounded and synthetic worlds share runtime primitives.
San Francisco first, Honolulu second — the company is not a city-specific simulator.
Simulation before photorealism — an excellent, inspectable 2D/2.5D world comes first.
Hybrid agents, not an LLM call every tick — frontier reasoning is reserved for salient decisions.
Persistent worlds — time, memory, relationships, and consequences accumulate.
Fork anything — any checkpoint can produce an alternate reality.
Version everything — inputs, rules, models, scenarios, seeds, and runs are reproducible.
Validate grounded worlds — exploration is separated from calibrated forecasting.
Synthetic people are not inferred real people.
World state is renderer-independent — map, dashboard, character view, or game.
Creation is a platform primitive — worlds buildable through data, code, templates, or language.
Usage compounds the moat — worlds, rules, modules, and benchmarks improve the platform.

The team

Enreef is a founding team of three, building across simulation engineering, geospatial data systems, applied AI, and product. The vision: any real place or invented society can become a persistent, versioned world — where humans and AI explore alternate futures, test interventions, and understand consequences before they reach reality.

Want to be part of it? See Careers.