Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

Glossary

ReAct agent

The simplest agent loop, often written by hand: the model writes a thought, picks an action (a tool call), reads what comes back, and repeats until it can answer. It comes from the 2022 paper ReAct: Synergizing Reasoning and Acting in Language Models.

A loop you write yourself has no framework in the way, and none of a framework's safeguards either: approvals, a sandbox, a limit on steps and logging are yours to add. The paper's authors warned that connecting a model to actions can lead it to look up private information or take harmful actions, and limited their own agent to specific websites.

Open the glossary in the Learning Center