一位开发者在Hacker News上分享了其自主构建的AI Agent系统架构。[1]该系统由6个专业化Agent组成，分别负责执行协助（ea-agent）、运维（ops-agent）、开发（dev-agent）、市场推广（gtm-agent）、研究（research-agent）和基础设施管理（vps-agent），用于统筹管理多个产品和非营利组织的日常运作。[1]系统基于开源通讯平台Buzz v0.5.9和Nostr协议构建，运行在DigitalOcean的单台服务器上，配置为4个虚拟CPU、8GB内存和160GB存储空间。[1]
系统采用OpenAI模型驱动各Agent的智能决策，月度API订阅费用约为100美元，可根据需求扩展至200美元。[1]在记忆管理层面，系统为每个Agent profile配置了独立的Mnemosyne记忆库，同时维护一套基于Obsidian的共享知识库。[1]这一架构已在自动化Sentry错误响应、外包开发任务、生成日常工作汇总等场景中投入使用，自2026年8月11日部署后已运行约一个月。[1]
不过，作者坦诚指出，该系统当前的投入产出比尚不理想。[1]他直言"搭建这套系统花费的时间是自己手工完成上述工作所需时间的10倍"，意味着效率优化空间仍然较大。[1]
A developer has shared details of their custom AI Agent infrastructure, which consists of six specialized agents designed to manage multiple products and nonprofit organizations [1]. The system runs on a single DigitalOcean Basic Droplet configured with 4 vCPUs, 8 GB of RAM, and 160 GB of disk storage [1]. Built on the open-source Buzz communication platform and Nostr protocol, the setup leverages OpenAI models with a monthly subscription cost of approximately $100, scalable to $200 per month [1].
The six agents are organized by function: ea-agent for executive assistance, ops-agent for operations, dev-agent for development tasks, gtm-agent for go-to-market activities, research-agent for research operations, and vps-agent for infrastructure management [1]. The system employs a hybrid memory architecture combining per-profile Mnemosyne repositories with a shared Obsidian wiki [1]. The developer has demonstrated various use cases including automated Sentry response handling, outsourcing development tasks, and daily work summarization [1].
However, the developer acknowledged that the current return on investment remains suboptimal, noting that "I've spent 10x longer setting this up than it would have taken me to do any of the things above on my own" [1]. The system has been operational for approximately one month as of August 11, 2026 [1].