liveBook · Manning

Memory module

As explained in Build a Multi-Agent System (from Scratch)

A component that stores information from agent task execution and retrieves selected information into the agent's processing context.

A memory module stores potentially useful information from an agent’s task execution and makes some of it available in the context for future tasks. Reusing past results can avoid repeating tool calls. section 1.4

A typical memory flow follows the agent’s processing loop:

  • At task start, check memory and load information that may help with the current task. section 1.4
  • During execution, save results from sub-steps and tool calls, along with the final task result. section 1.4

When and what to load or save are design decisions. A framework’s memory module can provide infrastructure and recipes for episodic memory. section 1.4 section 1.7

Figure 1.8 A mental model of the LLM agent processing loop that has memory modules for saving and loading important information obtained during task execution.

The processing-loop diagram illustrates where memory loads and saves fit into task execution. section 1.4

How Build a Multi-Agent System (from Scratch) covers it

How deep it goes
Defines memory and traces its load-and-save flow in the agent loop, while deferring implementation details.
Examples
Uses stored sub-steps, tool-call results, and final task results as reusable information.
What it assumes
Assumes an LLM agent processing loop and defines context as the input text used to generate new text.
What you can do after
Understand how to check memory at task start and save sub-step, tool-call, and final results.
How it connects
Connects agent processing loops with context, task trajectories, tool calls, and episodic memory.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build a Multi-Agent System (from Scratch). The section links point to the parts of the book it draws on.