Eduardo H. Ramirez, PhD

Anatomy of an agent, step by step

What the model answers, what the harness executes, how observations come back and how the prompt grows in each iteration.

Demo indexEspañol
User Harness (agent program) Model (LLM) Tools Memory Instructions

User

States the request and receives the final answer.

Harness: the agent program

Runs the loop: builds the prompt, calls the model, reads its response, calls tools and stores observations.
Start

Instructions

Memory

Model (LLM)

Receives text and returns text. It does not run tools or remember previous iterations on its own.
LLM
Keyboard: arrow keys to move forward or back, space to play