TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Show HN: Upsonic: An AI agent framework with client-server architecture

9 点作者 gorkemcetin3 个月前
Three months ago, we started developing an open source agent framework. We previously tried existing frameworks in our enterprise product but faced challenges in certain areas.<p>Problems we experienced:<p>* We risked our stateless architecture when we wanted to add an agented feature to our existing system. Current frameworks lack server-client architecture, requiring significant effort to maintain statelessness when adding an agent framework to your application.<p>* Scaling problem - needed to write Docker configurations as existing frameworks lack official Docker support. Each agent in my application required a separate container (e.g., Twitter page analysis, website scraping, automatic documentation writing, etc.), necessitating individual agent deployment and health checks monitoring.<p>* Needed LLM calls for simple tasks - both fast and cost-effective solutions. With increased model capabilities, the framework should offer this option. I could handle LLM calls myself, but structured outputs required extra work within task structure.<p>Due to these problems, we decided to build a dockerized agent framework with server-client architecture. Though server-client architecture slowed development, we observe many benefits for users. We&#x27;re developing a task-centric approach as we expect agents to complete simple tasks and assist with work.<p>As tool support is crucial for completing tasks, we built a structure officially supporting MCP servers. Client-server architecture proved beneficial for MCP server stability.<p>Finally, we prioritized simplicity in the framework, developing a structure where outputs are more easily coded with object responses. We&#x27;d be very happy if you could check our repo. I&#x27;d love to hear any questions you may have.,

2 条评论

karakanb3 个月前
This seems really interesting, thanks for sharing.<p>I can see myself deploying something like this for some of our internal usecases, however the deployment options around such tooling seem to be rather complicated, especially when it comes to the need for RAG and stuff. Would I need to host something else on top of Upsonic for these sorts of usecases?
评论 #42975353 未加载
bartuu3 个月前
Generating documentation with different agents, each in their own image? Yeah, been there. A shared base image would&#x27;ve definitely simplified things.
评论 #42974807 未加载