AchiralAchiral

Concepts · Humans and machines

SOAR working memory holds the current problem-solving situation: states, operators, objects, and working memory elements.

Published2026-09-090 reads
Editorial illustration of a small active working-memory tray surrounded by larger memory stores.

Working Memory in SOAR

What is the system looking at right now?

SOAR answers that with working memory.

Working memory is not the whole memory of the system. It is the current situation. It holds what SOAR is solving, which operator is selected, what objects are relevant, and what information is active enough to use.

The Soar manual calls working memory the current problem-solving situation. That is the easiest way to read it.

Working memory is active context

An AI agent may have old notes, saved preferences, documents, tool logs, and policies.

But it cannot act on all of them at once.

It needs a current state:

text
goal: resolve support issue
user: Susan
task: password reset
last result: failed
selected operator: ask for failure code

That sketch is not SOAR syntax. It only shows the idea.

Working memory is where the current problem becomes inspectable.

Working memory elements

SOAR represents details in working memory as working memory elements, often called WMEs.

A WME is a small piece of information. Several WMEs can describe the same internal object.

For example, one object may be described by facts like:

text
it is a task
its user is Susan
its status is failed

SOAR does not care whether the labels are elegant. The modeler chooses the labels and has to use them consistently.

That is a useful lesson for AI systems too. Memory is not helped by clever names. It is helped by stable structure.

Why this matters

Working memory gives decision-making a surface.

Without that surface, every action has to be inferred from raw text. The model has to re-read the situation, guess what matters, and choose a tool from the prompt.

With a working state, the system can ask clearer questions:

  • What is the goal?
  • What changed since the last step?
  • What operator is selected?
  • What memory is active now?

Those questions make agent behavior easier to inspect.

The takeaway

SOAR working memory is the system's current workspace.

It does not store everything. It holds the current problem in a form the rest of the architecture can use.

For AI builders, that is the lesson: before an agent acts, it needs a clean current state.

Sources

Go back to What Is SOAR Cognitive Architecture?, continue with Operators in SOAR, or return to the Concepts hub.