# Agent SDK

> A code kit that runs the agent loop for you. It calls the model and the tools.

**Field:** Software Development  
**Category:** AI Engineering  
**Also known as:** Agents SDK, AI Agent SDK, Agent Library

## Think of it like

It is like a robot vacuum. You press start, and it keeps going until the floor is clean.

## Definition

An agent SDK is a code library that runs a whole agent loop for you. You give it a model, instructions and tools. It sends the task to the model and runs each tool the model asks for. Then it sends the result back and repeats until the work is done. Most also handle limits, memory and a log of each step.

## Example

Lucy wants a bot that reads a bug report, searches the code and drafts a fix. With a client SDK she would write the loop herself. With the Claude Agent SDK or the OpenAI Agents SDK, one call runs every step.

## Why it matters

The loop looks simple but hides hard parts: when to stop, what to do on errors, and how to save the work. An agent SDK has already solved them.

## Common mistake

Reaching for one when you do not need it. A task with one question and one answer needs only a client SDK.

## Three principles

- **The loop is built in.** Model, tool, result, repeat.
- **You bring the tools.** And the instructions.
- **Limits and logs.** Turns, cost and a trace.

## Related terms

- [Client SDK](https://learn.benibix.com/term/development/client-sdk)
- [LLM Framework](https://learn.benibix.com/term/development/llm-framework)
- [OpenAI Agents SDK](https://learn.benibix.com/term/development/openai-agents-sdk)
- [Google Agent Development Kit (ADK)](https://learn.benibix.com/term/development/google-adk)
- [Claude Agent SDK](https://learn.benibix.com/term/ai/claude-agent-sdk)
- [Agent Loop](https://learn.benibix.com/term/ai/agent-loop)
- [Agent Harness](https://learn.benibix.com/term/ai/agent-harness)

## Told apart from

- [Client SDK vs Agent SDK](https://learn.benibix.com/compare/client-sdk-vs-agent-sdk): People think an agent SDK is just a client SDK with more features. It is a different job. A client SDK sends one request and hands back one reply. Your code reads it, runs any tool the model asked for, and calls again. An agent SDK runs that whole loop for you, tools and all, until the task is done.

![Your code makes one call to the agent SDK with a task, a model and a list of tools. Inside the SDK a loop runs: the model asks for a tool, the SDK runs the tool, the result goes back to the model, and this repeats. When the model says it is done, one final answer comes back to your code.](https://learn.benibix.com/diagrams/dev-agent-sdk.svg)

---

Source: https://learn.benibix.com/term/development/agent-sdk
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-20
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
