AI Agent Building

These terms explain how agents are composed, grounded, and protected against common attacks.

Agent Loop

An agent loop is an iterative cycle where a system observes context, selects an action, executes it, and updates its plan.

Tool Use

Tool use is when an AI agent invokes external functions or services (like search, code execution, or databases) to complete tasks beyond pure text generation.

Vector Database

A vector database stores embeddings so an application can retrieve semantically similar documents for tasks like search or RAG.

Prompt Injection

Prompt injection is an attack where untrusted content is crafted to steer a model into ignoring rules, leaking data, or taking unsafe actions.