# Agents, Tools, and Agent Frameworks

Modern AI increasingly acts as **agents** that plan, use tools, browse, and remember. This sub-article expands [Section 2.5 of Hardware & Software](../hardware-software.md) and [Recent Topics](../recent.md).

## What Is an Agent?
An agent perceives, decides, and acts toward goals. In current practice, an LLM is wrapped in a loop that:
- Observes the environment (text, APIs, screens, sensors).
- Reasons about next steps.
- Calls tools (search, code execution, calculators).
- Updates memory.

## Frameworks and Libraries
- **LangChain**, **LlamaIndex**, and **AutoGPT**-style stacks orchestrate agent loops.
- **OpenAI Functions/tool-calling** and Anthropic's **computer use** let models invoke tools natively.

## Reasoning Patterns
- **Chain-of-thought** and **ReAct (Reason+Act)** interleave thinking and acting.
- **Tree-of-thought** explores multiple reasoning branches.

## Memory
- **Vector databases** (Pinecone, Weaviate, FAISS) store embeddings for long-term recall.
- Conversation history, scratchpads, and external knowledge bases compose agent memory.

## Why It Matters
Agents connect models to the world, enabling automation, research, and real-world action—and raising new safety questions.

## Sources
- Yao, S. et al. (2022). *ReAct: Synergizing Reasoning and Acting in Language Models.*
- Lewis, P. et al. (2020). *Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks.*

*Back to [Hardware & Software](../hardware-software.md) · See [Agents & Autonomy](agents-autonomy.md).*