aifollow.news
← TrendingLatest development Oct 31 21:36
STORY IN FOCUS

DeepSeek releases dsh-superagent coding agent research prototype

The DeepSeek Harness team released the MIT-licensed dsh-superagent research prototype, which dispatches authorized GitHub requests to isolated sessions for parallel work. Proactive work defaults to observe mode, which only records what it would accept. Its task completion results come from synthetic messages and simulated tools; real coding quality was not measured.

1 developments3 reports3 sources

Event developments

Coverage of the same occurrence shares a node; subsequent developments have their own nodes.

DeepSeek proposes Super-agent architecture for parallel task execution

DeepSeek proposed a Super-agent architecture in which a Router assigns tasks and Workers execute them concurrently in isolated sessions. In a synthetic workload test, the team measured a p95 task completion time of 15.7 minutes, but the experiment did not assess actual performance on individual tasks; the published prototype plugin is not an official product.

View source reports · 3 sources · 3 reports
All reports · 3

By original publication time, with each report's bookmarks and feedback preserved.

MarkTechPost AI 研究报道 Machine translation

DeepSeek releases dsh-superagent coding agent research prototype

The DeepSeek Harness team released the MIT-licensed dsh-superagent research prototype, which dispatches authorized GitHub requests to isolated sessions for parallel work. Proactive work defaults to observe mode, which only records what it would accept. Its task completion results come from synthetic messages and simulated tools; real coding quality was not measured.

量子位 AI Machine translation

DeepSeek shares Super-agent architecture and prototype for ongoing cross-platform collaboration

DeepSeek has shared an engineering-focused Super-agent architecture and a GitHub prototype link. Its Router–Worker design aims to carry tasks across group chats and PR comments while handling work in parallel. The team ran preliminary tests in a simulated setting with about 300 messages and 20 requests; its asynchronous ReAct mechanism is not enabled by default.

IT之家 科技新闻 Machine translation

DeepSeek proposes Super-agent architecture for parallel task execution

DeepSeek proposed a Super-agent architecture in which a Router assigns tasks and Workers execute them concurrently in isolated sessions. In a synthetic workload test, the team measured a p95 task completion time of 15.7 minutes, but the experiment did not assess actual performance on individual tasks; the published prototype plugin is not an official product.