In Terra Engine, every person in the simulation is driven by an artificial intelligence. They make decisions, carry out actions, manage projects, and react to events across the world.
The system is built on a hybrid approach. Behaviors remain consistent at all times, while enabling generative AI adds a narrative layer where motivations and reasoning become visible.
Every person is an agent
Each person in the simulation is powered by a dedicated AI. It receives a structured overview of the situation, including key actors (the country being led, allies, rivals), ongoing relationships and dynamics, current priorities and opportunities, and the available actions.
It can then explore specific elements in more detail, such as the state of a country, the role of a key figure, or recent events, in order to refine its understanding before acting.
Based on this, the agent evaluates possible actions and selects the most relevant ones. The world evolves continuously, driven by contextual decisions unique to each profile.
Specialized agents for each type of decision
Multiple types of agents coexist, each handling a specific role within the simulation.
As new features are introduced, additional specialized agents will be added to expand the range of possible behaviors.
Two simulation modes
Both modes rely on the same foundation: the simulator identifies available actions based on the current situation.
Without language models, decisions are applied automatically based on their relevance.
With language models enabled, the agent selects from these same options and makes its reasoning and decision factors visible.
Assistant and journalist
In addition to the agents that represent each person, dedicated tools provide a broader perspective on the simulation.
These two approaches complement each other. One enables interactive exploration, while the other offers a structured way to follow the evolution of the world. They are part of a growing set of tools designed to provide multiple, complementary perspectives on the simulation.
Observing reasoning and decisions
When language models are enabled, decision-making phases become visible. A dedicated view shows what each person is analyzing, the elements they consider, and the actions they are evaluating.
Beyond this analytical view, a narrative layer also emerges: individuals express their reasoning, explain their choices, and can frame their decisions as statements or public communications.
An agent analyzing the situation: information reviewed, options considered, and decision in progress.
It is also possible to observe the strategy defined by the agent, and understand how it guides decisions over time.
A strategy used by an agent to guide decisions and maintain consistency over time.
Concepts: shared references for understanding the simulation
Concepts are contextual help elements displayed in the interface. They provide clear explanations of how mechanics work, the role of systems, and the implications of decisions.
These same concepts are also used by agents powered by language models. By relying on these references, they interpret situations more accurately and make decisions aligned with the rules of the simulation.
Contextual references used both by the interface and AI agents.
A fully local AI, integrated into the system
Language models run directly on the GPU, with all processing handled locally and no dependency on external services.
The model can be selected based on the available hardware, particularly GPU memory. This flexibility allows the AI to adapt to different configurations while remaining fully integrated into the simulation.
This approach enables continuous usage, with no network latency, no external costs, and no internet connection required.
See the system in action
The video below shows an accelerated one-month simulation. Leaders analyze the situation, define strategies, and make decisions, while their reasoning and positions become visible.
At the same time, an AI journalist reports on key events through weekly articles, offering a complementary perspective on how the world evolves.
Follow interactions between agents, showcasing the depth of decision-making and the dynamics of the simulation during a major crisis.
Summary
Terra Engine’s hybrid AI combines two complementary approaches. On one side, simulation-driven decision systems structure choices and ensure consistency. On the other, language models interpret these decisions, explain them, and bring a narrative dimension.
This combination preserves a strong decision-making foundation while making the simulation more readable and engaging.
As the simulator evolves, so will this approach. New mechanics will expand agent capabilities, and the way they reason, plan, and express decisions will continue to grow.