This document aims to help you install and run Jason, as well as developing a simple multi-agent system using Jason.

Installation and Configuration

  • Java JDK >= 1.7 is required.

  • Download Jason from here and unzip it.

  • You can now execute the Jason IDE by going to the folder jedit and double-clicking on jedit.jar.[1] For Linux and MacOS, the script jason-ide can be also used (see more instructions here).

screen initial

As you can see, Jason runs as a plugin of jEdit (a text editor developed in Java). This is useful because, although in Jason agents are programmed in a variant of AgentSpeak, in most cases you’ll need to do some Java programming (e.g., if you want to create an environment where the agents are situated).

In some cases, it might be necessary to configure the directory where Java JDK is installed. This can be done in the menu Plugins → Plugins Options → Jason:

screen conf

You can also change the Jason configuration by double clicking the jason jar file located in the libs directory.

Done! It’s that simple!

You are now ready to run a Jason multi-agent system, as we explain next.

Execution of an example

Jason comes with many examples and demos. The examples are multi-agent system applications for simple scenarios. The demos are meant simply to show how to use some useful features of Jason. You can find a brief description of examples and demos at

We will now run the classic Cleaning Robots example:

This is a very simple example, showing a robot that searches the whole environment (represented as a grid) for pieces of garbage, and when one is found, it takes it to another robot, located in the centre of the grid, where there is an incinerator; the moving robot then goes back to the place where the last piece of garbage was found and continues the search from there. It is based on the original scenario that Anand Rao used when he introduced the AgentSpeak language.

+ align:center

  • All Jason projects have a configuration file that ends with .mas2j, so to open the cleaning robots example, open the file examples/cleaning-robots/mars.mas2j that you’ll find in the folder where you installed Jason.

    screen mars

    The project file defines the underlying infrastructure that has been chosen for that project (Centralised, Jade, …​), the Java class that implements the environment (MarsEnv), and the agents that belong to this application (agent r1 searches for pieces of garbage and r2 incinerates them).

  • To execute this application, click on the icon run. Two windows are opened, the first is the application GUI and the second is the Jason MAS Console where all print messages are shown (MAS is a common abbreviation of Multi-Agent Systems).

    screen masconsole

  • To stop the MAS execution, click on suspend, either in the MAS Console or in jEdit.

Creation of a simple example

In this section we will create a new and simple example where two agents, bob and tom, exchange greeting messages.

  • Click on the newProject icon and fill in the project name field with greeting.

    screen newproject

    (Don’t worry about the syntax error in the project, it is caused because there are no agents in the system.)

  • Add a new agent called bob by clicking on newAgent and filling in only the agent name field. Note that the agent name has to be an AgentSpeak term, so it cannot start with an uppercase letter.

    screen newagent

  • Do the same for tom.

  • As you can see, there is a skeleton for the agent’s code: the agent has no beliefs, but an initial goal start and one plan to achieve this goal. The plan simply prints something when triggered.

    We will now change tom 's code so that it sends a “hello” message to bob. To send messages, an internal action called .send is used:

    // Agent tom in project greeting.mas2j
    +!start : true <- .send(bob,tell,hello).

    In bob code, we remove the start goal (and its related plan), leaving its program empty:

    // Agent bob in project greeting.mas2j
  • You can now run the project. There is no output! However, in the MAS Console, click on the debug button debug and then select, in a new windows named Jason Mind Inspector (also available at http://localhost:3272) the agent bob (if the agent’s list is empty, click once in the run button). The mind inspector for bob will look as follows:

    screen mindinsp

    Note the bob has a belief hello[source(tom)], which means that it received tom’s message.

  • Suppose now that we want bob to react to this message. Since the received message implies a belief addition, an event like +hello[source(tom)] is produced and may trigger the execution of the following plan:

    // Agent bob in project greeting.mas2j
    +hello[source(A)] <- .print("I received a 'hello' from ",A).

    In the plan, A is a variable that contains the name of the sender. In AgentSpeak, as in Prolog, identifiers that start with uppercase letters are variables.

    When you run the new version, the output will be:

    screen hello

  • Since bob is a polite agent, we will now make it send a hello back to tom:

    // Agent bob in project greeting.mas2j
      <- .print("I received a 'hello' from ",A);

    and tom does the same:

    // Agent tom in project greeting.mas2j
    +!start : true <- .send(bob,tell,hello).
      <- .print("I receive an hello from ",A);

    Before running the system, think what you would expect to happen. Perhaps the agents will enter a kind of greeting loop?

  • Run the system and you will realise that there is no loop! The reason is because when bob receives the second hello, it already has this belief in its belief base (BB). Since nothing changed in the BB, no event was produced, and thus no plan triggered.

  • If you want to use JADE as the infrastructure, change the project as follows:

    MAS greeting {
        infrastructure: Jade

    Also change the configuration of the Jason Plugin to start the JADE Sniffer agent as well:

    screen confjade

    The windows created when you run the system are shown below:

screen runjade

An example with environment

In this section we will create a system where one agent will perform one action in a simulated environment.

  • Create a new project called testenv.

  • Add one agent called liz with the following code:

    // Agent liz in project testeenv.mas2j
    +!start : true <- burn.

    The plan’s body has only the action, burn. Action here is meant to an environment action (i.e., something that changes the state of the environment), and not internal actions (the ones which starts with a dot, or have a dot anywhere in their name).

  • The implementation of the burn action is done in an environment class. To create this class, click on the createEnv icon and fill in the environment name field with TestEnv.

    A skeleton for this class is added by Jason. Change it to be exactly as follows:
    import jason.asSyntax.*;
    import jason.environment.*;
    import java.util.logging.*;
    public class TestEnv extends jason.environment.Environment {
      private Logger logger = Logger.getLogger("testenv.mas2j."+TestEnv.class.getName());
      /** Called before the MAS execution with the args informed in .mas2j */
      public void init(String[] args) {    }
      public boolean executeAction(String agName, Structure action) {
        if (action.getFunctor().equals("burn")) {
          return true;
        } else {
"executing: "+action+", but not implemented!");
          return false;
      /** Called before the end of MAS execution */
      public void stop() {

    When an agent attempts to execute an environment action, the method executeAction of this class is executed. In this implementation, if the action burn is executed, a new percept fire becomes available to all agents.

  • Agent liz can now react to the perception of fire:

    +!start : true <- burn.
    +fire <- run.

    (The implementation of the run action is left as an exercise.)


Imagine a very simple environment formed by 4 locations (identified by 1, 2, 3, and 4) as in the figure below:


A vacuum-cleaner robot should be programmed in AgentSpeak to maintain the environment clean. The available actions for the robot are:

  • suck: remove dirt at the robot’s position;

  • left: move the left;

  • right: move to right;

  • up: move up;

  • down: move down.

To help the robot decide what action to take, the following percepts are given:

  • dirty: the robot is in a dirty location;

  • clean: the robot is in a clean location;

  • pos(X): the location of the robot is X (0 < X < 5).

The following diagram, using the Prometheus notation, illustrates the interactions between the robot and the environment.


An implementation of the environment class is available here.

Some tips

You can start programming your agent by thinking about how it should react to the available perception. For instance, what it should do when it perceives "dirty"? The action "suck", of course! In AgentSpeak, we program this reaction by means of a plan as follows:

+dirty <- suck. // when dirty is perceived, do the action suck

So, an initial and very reactive agent can simply react to every perception and be programmed as shown below (replace "someaction" for the action you think is the most suitable, you might also want to remove some of the plans):

+dirty  <- someaction.
+clean  <- someaction.
+pos(1) <- someaction.
+pos(2) <- someaction.
+pos(3) <- someaction.
+pos(4) <- someaction.

Since all perception is also included in the belief base, they can also be used to select the right plan, as in the following example:

+pos(1) : clean <- someaction.   // whenever I perceive I'm in pos(1) and
                                 // I believe that my position is clean,
                                 // do some action.

You will soon realise that this reactive approach has some limitation in defining a good behaviour for our vacuum cleaner. In fact, this agent should be defined has having goals, in particular, a persistent goal of maintaining the house clean. The easiest way to define a persistent goal is by a recursive plan; for example, the code below implements the persistent goal (represented by p) of printing out "a":

!p.                   // initial goal
+!p <- .print(a); !p. // to achieve the goal p, print "a"
                      // and after has p as a new goal.

Some comments on possible solutions for this exercise are available here.

This document has shown a very limited range of Jason’s features; the next section contains references where you can find further information.

More information

You can find more information about Jason at:

1. Eclipse can also be used as the IDE, see here.