Multi-agents technologies developed by CERV are used by CERVVAL to build simulators. Four different kind of agents are used :

• Entity-agents are used since 1997. These agents are most classically used in agent-based simulator. Each entity that populates the environment to be simulated is modelized by a single autonomous agent. Such an entity can be a water particle, an atom, a living being, ... The main advantage of entity-agents is their implementation is conceptually very simple.
• Reaction-agents are used since 2000. They are based on a decomposition of the world which is no longer entity-based but phenomenon-based. Ie, the atom of the simulator is an autonomous agent representing a single phenomenon. This modelisation has two main advantages: on one hand, it involves far less agents. On the other, it permits the addition or the removal of a new phenomenon, by simply adding or removing an agent, without any need to update existing agents.
• Interaction-agents are used and patented by CERVVAL since 2003. Interaction-agents take into account the mutual influence of phenomenon. Particularily Interaction-agents are used for implementation of diffusion phenomenon. Their role is then to transfer resources (particles, heat, energy, ...) from the influence areo of one agent to another. Since agent interact on each other only by way of their action on the structure of the environment, they are said to be structurally coupled.
• Enaction-agents are used and patented by CERVVAL since 2004. They turn statically defined structural coupling into an adaptative structural coupling. In an enaction-based modelisation, the environment is not explicitly structured (eg by entity agents or by finite elements) by the engineer. Instead the structuration is build by the phenomenons (ie the enaction-agents) of the simulated world. Each enaction agent builds a piece of environment structure, according to the knowledge it needs to work.