Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

DeepSeek DSH agent runtime targets constrained, multi-step execution

AI summary

掘金·人工智能 reports that DeepSeek has introduced DSH, also invoked as dsh, as an agent runtime environment for tool use, sandboxing, memory, approvals, and extensible plugins. The analysis describes a six-layer architecture and a Super-agent prototype featuring Router/Worker separation, chained recall, event-driven processing, watchdog checks, and device-cloud coordination, while emphasizing that the prototype is an architectural exploration rather than proof of current product performance.

In-depth insight

DSH is presented as infrastructure for moving memory, scheduling, approval, and recovery from users into a constrained execution environment, a notable approach to making agents handle multi-step tasks beyond a single chat response.

DeepSeek Harness 实战与架构解析 (1).png

DSH(命令行叫 dsh)是 DeepSeek 推出的 Agent 运行环境。它给模型提供工具、沙箱、记忆和可组装的插件系统,让模型在一个受约束的执行环境里把任务从开始做到交付。本文从一次“整理周报”的任务出发,拆解 DSH 的六层结构,讨论 Super-agent 原型中的 Router/Worker、链式召回、事件驱动和端云协同,最后给出上手路径与局限。

一、从一个周报任务说起

周三下午,同事在群里丢来一句:“把上周的测试报告整理成一页周报,下班前给我。”

用聊天窗口里的 AI,流程通常是这样:打开对话框,描述需求,把散在三个目录里的 log 逐个打开、复制、粘贴进去;它回一段排版不错的文字;你复制出来改两个数字,再贴到群里。

#DeepSeek#DeepSeek Harness (DSH)#dsh-superagent-prototype

Original · 15 min readRead the original
Group chat

No comments yet. Start the conversation.