Agency用于构建LLM驱动的Agent系统的极简开源框架

4.6 (5)

审阅者 Daniel Nikulshyn·更新 2026年7月

Open Source Developer Tools LLM Framework AI Agents Code-First Lightweight Prototyping

概览

Agency 是一个轻量级框架，适用于希望在大型语言模型（LLM）驱动下构建集成代理（agent）的应用程序的开发者，而无需承受更重量级编排平台的额外负担。它提供了定义代理、工具和交互所需的基本组件，同时不会干扰您的系统架构。该框架强调简单性和可组合性，使其适合于原型设计实验性智能体行为或将智能体能力嵌入到现有系统中。其最小化的表面积帮助开发人员准确理解底层正在发生的事情，这对于调试和定制非常有价值。 Agency 非常适合那些喜欢直接使用 LLM API、并且更倾向于构建定制化解决方案的工程师，而不是采用那些固执己见的、集成度过高的智能代理平台。

主要功能

Agent的最小核心抽象
LLM集成原语
工具和函数调用支持
可组合的Agent交互
面向开发者的、以代码为中心的API

价格

模型: Freemium
分类: 王常服务
评分: 4.6 / 5 (5)

使用场景

原型实验Agent行为

快速建立最小Agent架构，在无需较大编排平台开销的情况下测试围绕工具使用和LLM交互的新想法。

将Agent嵌入现有应用程序

使用小型、可组合的API将LLM驱动的Agent能力集成到已建立的代码库中，而不会干扰现有架构。

构建自定义Agent架构

利用最小的核心原语设计定制Agent系统，您可以控制编排逻辑，非常适合需要完全透明和灵活性的工程师。

学习和调试Agent内部

使用框架的小表面积，准确理解Agent、工具和LLM调用如何交互，从而更容易调试和自定义行为。

优点 & 缺点

优点

轻量级且易于理解
灵活且无偏见的设计
适用于自定义Agent架构
低集成开销

缺点

比完整的平台需要更多的手动设置
生态系统和社区较小
对复杂工作流的内置工具较少

评测

4.6

5 个评分的平均值。

登录以留下评测。

George Papadakis

Feb 23, 2026

Solid for our team

We rolled this out across the team last quarter and low overhead for integration. Developer-focused, code-first API fits neatly into how we already work, and lLM integration primitives removed a step we used to do by hand. Less built-in tooling for complex workflows, which is the main caveat, but it has held up under daily use.

Linda Petersen

Nov 2, 2025

Solid for our team

We rolled this out across the team last quarter and flexible and unopinionated design. LLM integration primitives fits neatly into how we already work, and composable agent interactions removed a step we used to do by hand. but it has held up under daily use.

Wei Chen

Sep 24, 2025

Does the job

Pretty happy overall. LLM integration primitives just works and good for custom agent architectures. Smaller ecosystem and community can be annoying, but no dealbreakers — I'd recommend it to a friend without hesitating.

Robert Ainsworth

Aug 31, 2025

Compared a few options

Evaluated this against two competitors. Where it wins: minimal core abstractions for agents and low overhead for integration. On balance the feature set — especially minimal core abstractions for agents — justifies the 5 stars for our use case.

Aisha Khan

Aug 9, 2025

Skeptical, then convinced

I went in skeptical — most tools in this space overpromise. It actually delivers on composable agent interactions, and low overhead for integration caught me off guard. still, I'd recommend giving it a real trial.

问答

What kind of support or community resources are available for Agency?

Because Agency is a minimal, developer‑focused project, its community is smaller than that of larger platforms, meaning fewer pre‑built tools and less extensive documentation, so you may rely on the core repository, issue tracker, and contributions from the open‑source community for help.

Asked by Hannah Goldberg · Jul 27, 2025

Is Agency suitable for production‑grade agent systems?

Agency is lightweight and unopinionated, making it great for prototyping or embedding agents into existing apps, but it requires more manual setup than full orchestration platforms and has a smaller ecosystem, so additional engineering effort may be needed for large‑scale production use.

Asked by Tomáš Novák · Jul 6, 2025

What is the cost to use Agency?

Agency is an open‑source framework, so there are no licensing fees; you can download and use it for free.

Asked by Tobias Hartmann · Jun 4, 2025

Which large‑language‑model APIs does Agency work with?

Agency provides generic LLM integration primitives, so you can connect it to any model that offers an API (e.g., OpenAI, Anthropic, Cohere) as long as you implement the required request/response handling.

Asked by Katarzyna Zielinska · Jun 7, 2025