The high level architecture of the agent workflow framework

 

Both the CLI and GUI programs are very simple. 

The core library is just currently 600 lines long.  The main entry point of the core is exec_agent() which allows every agent to be treated identically.   All exec_agent() does is promote template and proc agents to a workflow so we can then call the identical exec_workflow() routine.

Then inside the exec_workflow() function we have three nodes marked in yellow, these are the only three kinds of step there is.  If a workflow is embedded as a step, it recursively calls workflow, otherwise it processes the template and proc agents at that level of workflow. 

The high level function graph of the core library/CLI system.

The GUI has been refactored into modules that isolate the functions you see in these vertical stacks.

The high level function graph of the GUI system.

This is the directory listing and line size of the files.

Directory listing of the editor GUI

If you don't tell the GUI where the core library is you get this error:

You did not give the path to the core library file if you see this.

The only thing this does is deactivate the run button.  It has no other effect on the GUI.

As you can see in the above core chart, to load the CLI program as a library you have to:

• Add the --lib-path to the dynamic_workflow_agents.py tiny core file. 
• Import a few needed entry points in the 
• Load the config.json file. 
• Set up logging
• Set up depth management

See the web services program for a concrete example.

All in all, this program punches well beyond its weight class of the size and complexity of its programs. By going with the grain of the data we get features that are just emergent and natural.