The SML API is the standard method to get Soar agents to communicate with external environments, such as simulations or games. The API was written natively in C++, but has Java, Python, and Tcl bindings generated via SWIG. This page lays out the steps required to compile C++, Java, and Python SML client programs successfully on Linux/OSX/Windows. These instructions are for the 9.3.2 and later releases. Previous releases required a slightly more complex compilation process.

Building a program to call the SML API is conceptually straightforward in any language. Usually, only one or two dependencies must be specified. In practice, this can still be painful because the operating system must be told where to look for the dependencies. All dependencies are located in a single directory or one of its subdirectories. This will be SoarSuite/bin if you are using a pre-built release distribution of Soar, or (by default) SoarSuite/out if you built Soar yourself.

For all languages, compilers and operating systems, please make sure that you link 32-bit programs to the 32-bit version of Soar, and likewise for the 64-bit version. The error messages compilers produce for mismatched instruction sets tend to be cryptic, and you probably won't realize what the problem is from reading them. C++

To use the SML API in a C++ program, you must include the header file "sml_Client.h" and link to the Soar shared library. If you are using the pre-built release package, the Soar shared library will be in SoarSuite/bin, named (Linux), libSoar.dylib (OSX), or Soar.dll(Windows). The header will be in SoarSuite/bin/include. If you are building Soar from source, by default the build script will put the shared library in SoarSuite/out, and the headers in SoarSuite/out/include. Also note that when building from source, you need to build the "kernel" target (built by default) to get the shared library, and the "headers" target (not built by default) to produce the include directory. See the BuildSconsScript for more information. For the rest of this section, we will assume the reader is using the pre-built release package.

For convenience, here's a Hello World program that you can use to test your compilation process:

#include <iostream>
#include "sml_Client.h"

using namespace std;
using namespace sml;

int main(int argc, char *argv[]) {
        Kernel *k = Kernel::CreateKernelInNewThread();
        Agent *a = k->CreateAgent("soar");
        cout << a->ExecuteCommandLine("echo Hello World") << endl;
        string dummy;
        cin >> dummy;
        return 0;
Linux / Mac OS X

It's fairly straightforward to compile an SML client program from a shell using g++ or clang++. Both compilers take the same arguments. Supposing you unpacked the release into /home/user/SoarSuite, the compilation command should be:

g++ -L/home/user/SoarSuite/bin -I/home/user/SoarSuite/bin/include your_program.cpp -lSoar

-L specifies additional library search directories, -I specifies additional include search directories, and -l specifies libraries to link to. The linker automatically figures out that -lSoar refers to the file on Linux or libSoar.dylib on OSX.
Note that even though the linker can find the Soar shared library at link-time, the OS will still not know where it is at run-time (unless you use theRPATH feature). Therefore, you still need to set the environment variables LD_LIBRARY_PATH on Linux and DYLD_LIBRARY_PATH on OSX to/home/user/SoarSuite/bin when running your compiled program. More details can be found in BuildLibrarySearchPaths.

Windows with Visual Studio

Compiling an SML program using Visual Studio is a little more tedious because you have to navigate its cryptic configuration GUI. Here we present step-by-step instructions for setting up a new C++ project. These steps were tested using MSVC++ 2010 Express, but should be very similar or identical for other versions of Visual Studio. We assume you extracted the release into C:\SoarSuite.
  1. In the menu, select File/New/Project...
  2. Choose the "Empty Project" template. Enter a name for the project.
  3. In the Solution Explorer, under the newly created project, there should be a "Source Files" folder. Right click it, choose Add/New Item... Choose "C++ File" in the template selection window that pops up, and give the file a name. You need to do this now so that the Properties window will show options for compiling C++ programs. I also copied the contents of the above Hello World program into the new file, but you don't have to.
  4. Right click on the new project (NOT the solution) in the Solution Explorer, choose Properties.
  5. Click on "Configuration Manager" in the upper right corner of the Properties window. Make sure that your project's Platform matches the version of Soar you downloaded/compiled. That means if you have a 32-bit version of Soar.dll, you need to use the "Win32" platform, and if you have a 64-bit version, you need to use the "x64" platform. VC++ Express doesn't include a 64-bit compiler, so if you are using it, you must use the 32-bit Soar. When you are done, close the Configuration Manager.
  6. In the "Configuration" drop-down box in the top left corner of the properties window, select "All Configurations" so that the changes you make apply to both debug and release configurations.
  7. Under Configuration Properties / Debugging / Environment, enter PATH=C:\SoarSuite\out. This will set the PATH environment variable to find Soar.dll when you run your program from within the IDE.
  8. Under Configuration Properties / C/C++ / General / Additional Include Directories, add the entry C:\SoarSuite\bin\include.
  9. Under Configuration Properties / Linker / General / Additional Library Directories, add the entry C:\SoarSuite\bin.
  10. Under Configuration Properties / Linker / Input / Additional Dependencies, add Soar.lib.
  11. Click "Okay" in the Properties window to save the changes.
Now you should be able to build and run your project. Remember that if you want to run the program outside of the Visual Studio IDE, you need to add C:\SoarSuite\bin to your PATH environment variable, discussed in more detail in BuildLibrarySearchPaths.

Static Linking

You can compile Soar as a static library, as described in BuildSconsScript. To link your program to a static Soar library, the only change you have to make is to define the macro STATIC_LINKED before you include sml_Client.h. There are macros in the Soar headers that will expand to different values depending on whether STATIC_LINKED is defined. Specifically, the prefix declspec(dllimport) is prepended to all SML API functions when compiling with MSVC++. More information is available here, but you don't need to understand it to compile successfully.
  • With g++ and clang++, the easiest way to do this is to pass in the flag -DSTATIC_LINKED.
  • In Windows/VC++, go to the Properties window for the project. Under Configuration Properties / C/C++ / Preprocessor / Preprocessor Definitions, add the text STATIC_LINKED. Click "Apply", then look under Configuration Properties / C/C++ / Command Line. You should see/D "STATIC_LINKED" somewhere in the command.

The only requirement for compiling a Java SML client is that the file sml.jar be in your class path. This file should be located inSoarSuite/bin/java in the release, and SoarSuite/out/java if you built Soar yourself and included the target sml_java (built by default).

In the following, I assume you're compiling a source file with the following contents:

import sml.Kernel;
import sml.Agent;

public class HelloWorld {
        public static void main(String[] args) {
                Kernel k = Kernel.CreateKernelInNewThread();
                Agent a = k.CreateAgent("soar");
                System.out.println(a.ExecuteCommandLine("echo Hello World"));
To compile your program using javac directly from the command line, add sml.jar to the class path using the -cp flag. The command is essentially the same for all operating systems:

javac -cp /home/user/SoarSuite/bin/java/sml.jar (for Linux/OSX)
javac -cp C:\SoarSuite\bin\java\sml.jar (for Windows)

This should produce a file HelloWorld.class.

Oddly enough, running a Java SML client program is trickier than compiling it. The classes in sml.jar use JNI to call the C++ API under the hood. The JNI functions are compiled into a native shared library named (Linux),libJava_sml_ClientInterface.dylib (OSX), or Java_sml_ClientInterface.dll (Windows). This file should be in SoarSuite/bin orSoarSuite/out. When you run your program, you have to make sure the Java virtual machine can locate this library as well as the Soar shared library. This is explained in BuildLibrarySearchPaths.

After you set the library path correctly, you can run HelloWorld.class using the JVM. When you run the program, sml.jar should also be in the class path. Assuming HelloWorld.class is in the current directory, the command to run the program is:

java -cp /home/user/SoarSuite/bin/java/sml.jar:. HelloWorld (for Linux/OSX)
java -cp C:\SoarSuite\bin\java\sml.jar;. HelloWorld (for Windows)

You should see "Hello World" printed to the console. If you didn't set the library search path correctly, running your program will produce an error that looks something like this:

java.lang.UnsatisfiedLinkError: no Java_sml_ClientInterface in java.library.path
Exception in thread "main" java.lang.UnsatisfiedLinkError: no Java_sml_ClientInterface in java.library.path
        at java.lang.ClassLoader.loadLibrary(
        at java.lang.Runtime.loadLibrary0(
        at java.lang.System.loadLibrary(
        at sml.smlJNI.<clinit>(
        at sml.Kernel.CreateKernelInNewThread(
        at HelloWorld.main(

Many Java programmers use the Eclipse IDE. Here are the steps to create an SML project in Eclipse. Note that Eclipse is a general purpose IDE and comes in many different variations. These instructions were written for "Eclipse IDE for Java Developers". Remember that all we are doing here is getting Eclipse to find sml.jar at compile time and the JNI shared libraries at run time.
  1. In the main menu bar, choose File / New / Java Project. Give the project a name, then click Next.
  2. Choose the Libraries tab, then click "Add External JARs". Choose SoarSuite/bin/java/sml.jar in the file selection dialog. Click Finish.
  3. In the "Package Explorer", right click on the "src" folder under your project and choose New / Class, and create a class with a main function. For example, you can create a HelloWorld class and paste in the contents of the Hello World program above.
  4. Again in the "Package Explorer", right click on your project, choose Properties. Then choose "Run/Debug Settings", and click the "New..." button on the right. Choose "Java Application" in the pop-up window. In the configuration properties window that comes up, click "Search..." and choose the class you created with the main function.
  5. In the same window, select the "Environment" tab, click "New..." to add a new environment variable. For the Name field in the pop-up, enterLD_LIBRARY_PATH if you're in Linux, DYLD_LIBRARY_PATH if in OSX, or PATH if in Windows. For the Value, enter the full path to yourSoarSuite/bin directory. This will help the JVM find the JNI shared library when you run your project.
  6. Click OK, then OK again to close the configuration properties window.
Now you should be able to run your project by choosing Run / Run in the main menu bar. Python

To call SML via Python, you need to import the module Python_sml_ClientInterface, defined in the file This file should be in SoarSuite/bin in the release, and SoarSuite/out in your own build, if you built the target sml_python.
Like the Java SML bindings, the Python bindings also depend on a native library, called (Linux),_Python_sml_ClientInterface.dylib (OSX), or _Python_sml_ClientInterface.dll (Windows). Note the leading underscore in all versions. This file should also be in SoarSuite/bin in the release or SoarSuite/out in your own build. We will assume the reader is using the pre-built release package and both files are in SoarSuite/bin.
There are two ways to make the Python interpreter locate these files.
  1. Set the environment variable PYTHONPATH to include SoarSuite/bin. The Python interpreter will search SoarSuite/bin for modules for any program you run.
  2. Append SoarSuite/bin to the variable sys.path in the script itself before importing Python_sml_ClientInterface. This method is local to the script you applied it to, and is recommended if you have multiple versions of Soar on your computer.
Here is the Hello World program in Python, using the second method:

import sys
import Python_sml_ClientInterface as sml

k = sml.Kernel.CreateKernelInNewThread()
a = k.CreateAgent('soar')
print a.ExecuteCommandLine('echo hello world')
If you use the first method to modify the search path, the first two lines of the script are not needed. In any case, you should be able to run the script like a normal Python program:

$ python
hello world