Chapter 7-9: Knowledge-based Agents

Logical Agents

Knowledge-based agents typically store facts using logic. First-order logic can be expressed using the logic programming language Prolog. Here are two online examples using SWI-Prolog:

• A Simple Prolog Knowledge Base.
• The n-Queens Problem as a Logic Problem.

Python provides several modules for logic and symbolic mathematics. Here is a short primer on logic programming with Python.

Large Language Models

Large language models (LLMs) can be used as knowledge-based agents that work with natural language rather than logic. They can be used via an API or run locally. Important tasks include prompt engineering and context engineering.

Example:

• Running LLMs locally.
• Using Generative AI in Colab (without an API key)

Libraries and Prompting:

• OpenAI Python API Library.
• Hugging Face provides a large collection of downloadable pretrained LLMs.
• Prompt engineering guide from OpenAI.
• Textbook: Speech and Language Processing by Dan Jurafsky and James H. Martin. Part I contains an excellent introduction to language models, transformers, and large language models.

Agentic AI

An agentic AI solution uses a set of specially prompted LLM calls. It may involve any of the following:

• Multiple LLM calls
• LLMs can use tools (browse the web, access files, etc.) to interact with an environment.

• A planner coordinates the activities of the agents. It may use one of these approaches:

  • A developer-defined workflow using “prompt chaining” and feedback between LLMs.
  • An LLM that plans its own tasks (acting as an autonomous agent, hence the name agentic AI).

Video:

• Introduction to Agentic AI with Coding Examples (Code repository)

Articles:

• Components of A Coding Agent

Tools:

• OpenAI Agents SDK: native support for function calling, retrieval, and tool orchestration for the OpenAI ecosystem.
• CrewAI: orchestrate multiple specialized AI agents working collaboratively.
• LangGraph: a low-level LLM orchestration framework for building structured, reproducible agent pipelines.
• Model Context Protocol (MCP): An open protocol that enables seamless integration between LLM applications and external data sources and tools.

License

© 2025 Michael Hahsler. All code and documents in this repository are provided under Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) License.