[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-deepseek-super-agent-router-worker-architecture":3,"topics-all":38,"news-related-05fe5d2f-b83d-4aa9-9a68-5d0274b0d558":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"05fe5d2f-b83d-4aa9-9a68-5d0274b0d558","DeepSeek 开源 Super-agent:双层架构调度","DeepSeek 公开 Super-agent 架构并开源 MIT 原型仓库。常驻 Router 分派、隔离 Worker 执行,配两级事件过滤与 Thread 直通。官方消融实验自报 p50\u002Fp95 完成 12.4\u002F15.7 分钟,单 Session 为 20.4\u002F29.5 分钟,代价是读缓存 token 更多。","10 月 31 日,DeepSeek Harness 团队在微信公众号放出《Super-agent 设计探索:迈向更自主的 Agent 系统》,同一天开源了参考实现 [deepseek-ai\u002Fdsh-superagent-prototype](https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002Fdsh-superagent-prototype)。仓库 README 第一句就写明:这是研究原型,不是 DeepSeek 官方产品。\n\n文章开头抛出一个观察:一个人通常只能同时跟进 3 到 5 个活跃 Session。在单会话为单位的 Agent 设计里,单任务的天花板取决于模型能力,多任务的天花板取决于人的注意力。作者的解法是把 Session 从用户面前撤走,退化成系统内部的实现细节。\n\n## 双层架构:Router 与 Worker 分工\n\n第一个设计是拆成两层。常驻的 Router 层承接所有入口的人类交互,汇总跨任务信息与决策,是全局的序点、协调点和记忆。它不直接执行具体任务,甚至可以不配备 Bash 工具。作者的解释是,Router 的任务没想象中复杂,轻量模型就够。\n\nWorker 层负责真正干活。每个 Worker 上下文隔离、挂载完整的执行工具,专注完成单一任务;因为直接决定任务结果与代码质量,通常需要配置能力最强的模型。Worker 内部还能再创建 Subagent,对上层 Router 保持透明。用户看到的一直是同一个助手,背后是多个会话并行推进。\n\n## 事件驱动:两级过滤与 Thread 直通\n\n第二个设计是让 Agent 由环境事件驱动,而非等用户下指令,触发源包括群聊消息、邮件与日历提醒、代码评审评论、CI 结果。真实群聊噪音极大,官方用计算机系统的 Fast Path \u002F Slow Path 思路做两级过滤:程序层按来源、群组、白名单等确定性条件先做准入,通过的候选再交给 LLM 按提示词策略判断是否行动。\n\n同一 GitHub PR、同一 IM 话题下的连续事件构成一条 Thread。Router 用首条事件创建 Worker 后,该 Worker 订阅它,后续事件由程序直通,不再经过 Router。上下文则按缓存思路管理:Inline 内联近期信息,Spill 把超预算内容留在窗外只留引用,再由各类对象的唯一 ID 串成引用图,支撑按需多跳的链式召回。\n\n## 消融实验:12.4 分钟对 29.5 分钟\n\n官方文章的自评实验同时驱动群聊与 GitHub CI 消息流:单次实验约 300 条消息,其中约 20 条是工作请求,编码工作用一个只消耗时间的工具代替,衡量调度与协作能力;消息间隔与工作时长按 1\u002F4 缩放,每种方案跑 5 次。\n\n按[原文](https:\u002F\u002Fmp.weixin.qq.com\u002Fs\u002FzoS7GlReWPhfCjENGxXCdg)给出的数字,Super-agent 的任务完成时间 p50 与 p95 分别是 12.4 分钟和 15.7 分钟。同一份消融表里,带 Subagent 工具的单 Session 方案在中性提示下 p50\u002Fp95 为 20.4\u002F29.5 分钟,提示优先委派时为 13.8\u002F29.0 分钟。原因写在原文里:中性提示下模型 5 次运行一次都没调用 Subagent,工作全在主 Session 里被新消息反复打断;改成优先委派后 5 次里有 4 次委派,剩下 1 次模型判断工作\"很小\"自己执行,形成 p95 长尾。\n\n## 代价写在同一张表上\n\n同一张表还列了另一半账。Super-agent 单次实验的读缓存 token 高于单 Session,官方正文只给了方向;[第三方对同一张消融表的整理](https:\u002F\u002Fagentspulse.github.io\u002Ftutorials\u002Fdeepseek-super-agent-architecture)显示,Super-agent 为 23.23M,单 Session 两种提示配置分别是 6.05M 与 10.82M。去掉 Thread 直通机制后,完成时间与完整版基本一致,但 Router 需要做出的决定增加约 30%,总读缓存 token 从 23.23M 涨到 31.77M。\n\nRouter 自己还有一层批处理。官方数据是:Trigger 到达速率从每分钟约 1 条升到约 40 条时,平均 batch 大小从 1.0 升到 2.8,单次请求耗时仅从 2.5 秒增到 3.9 秒,单个决定分摊的推理时间因此从 6.4 秒降到 1.6 秒。据此官方判断,当前团队规模下一个常驻 Router 够用。\n\n## 开源原型给了什么\n\n原型仓库是 MIT 协议、TypeScript 实现,定位是单仓库部署的主动式编码智能体,监听 GitHub 事件,可接入飞书。请求进来后被记成工作项,在独立会话与目录里执行,最后回到原对话交付结果,比如一个 PR;Web 面板只列出还没人处理的事项。仓库自述的工程量指标是 4.1k star、375 fork。\n\n设计哲学很直白:尽可能少占用人的注意力,尽可能多地交付价值。规则是:被点名或被规则允许时才动手,干活时沉默,需要人决策时才开口。免责声明同样直白:这是研究原型,大量工程由编码智能体完成,可能仍有 bug,与内部版本并不相同,文章里讨论的部分设计并未实现。\n\n## 还有三件事没解决\n\n安全是最硬的一条:Agent Loop 部署在云端、端侧只留轻量 daemon,默认云端与端侧互信;一旦云端 Agent Loop 遭入侵或提示词注入被攻破,它连着的全部端侧设备都面临被控制的风险。即使云端没被攻破,凭据进入模型上下文也存在泄露风险,还需要让智能体不接触凭据也能用凭据的机制。其余是长期记忆,以及多个隔离实例的协作边界。\n\n## 所以呢\n\n这份设计没有提出新理论,作者说得很清楚:它把计算机系统里经典的工程思想搬进了 Agent 系统。真正有意思的是那个 3 到 5 个 Session 的观察——过去两年,模型能力把单任务的门槛抹平了,瓶颈上移到了人的注意力,而这次 DeepSeek 把它当成可以用系统设计解决的问题:把会话管理、上下文同步和事件路由沉到底层,让人只做决策与验收。\n\n不过实验是合成的、只跑 5 次,原型也明说不是产品。官方内部实例已在持续运行,参与问题复现与修复、代码审查响应,甚至参与 Harness 自身的迭代。","https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002Fdsh-superagent-prototype","998df6db-96e6-4b8e-8be1-cfa00a6cd177",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"e82b2d09-81b2-43d1-977e-e018443b3c14","coding-agent",{"id":19,"name":20,"slug":20,"description":14,"color":14},"b52db7e9-7c58-42c3-9536-5132cb2f8f72","deepseek",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"8ef1c52a-9555-4c90-89d2-1b693abb5854","en","Super-agent: DeepSeek Splits Session Management in Two","DeepSeek published a Super-agent design and MIT prototype: a Router dispatches, Workers execute. Ablation: p50\u002Fp95 of 12.4\u002F15.7 vs 20.4\u002F29.5 minutes.","On October 31, the DeepSeek Harness team published a design essay on WeChat, \"Exploring Super-agent Design: Towards a More Autonomous Agent System\", and open-sourced a runnable reference implementation the same day: [deepseek-ai\u002Fdsh-superagent-prototype](https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002Fdsh-superagent-prototype). The repository README states up front that this is a research prototype, not an official DeepSeek product.\n\nThe essay opens with a plain observation: a person can typically only keep track of 3 to 5 active sessions at a time. In agent designs built around a single session, the ceiling for one task is set by model capability, while the ceiling for many tasks is set by human attention. Changing your mind over IM, then reviewing a pull request on GitHub, means remembering for yourself what each session said. The proposed fix is to pull sessions out of the user's view and demote them to an implementation detail inside the system.\n\n## A two-layer split: Router and Worker\n\nThe first design choice is to split the work across two layers. A resident Router layer absorbs human interaction from every entry point, aggregates cross-task information and decisions, and acts as the global ordering point, coordination point, and memory. It does not execute concrete tasks and need not even carry a Bash tool. The team's read is that the Router's job is often less complicated than it looks, and a lighter model usually balances speed and effect well enough.\n\nThe Worker layer does the actual work. Each Worker has isolated context, mounts the full set of execution tools, and focuses on a single task; because it directly determines task results and code quality, it usually needs the strongest available model. Workers never block each other, and a Worker can spin up Subagents or be an Agent Team itself while staying transparent to the Router above. The user keeps talking to one assistant while several sessions advance in parallel behind it.\n\n## Event-driven: two-stage filtering and direct Thread delivery\n\nThe second design choice makes the agent driven by environment events instead of waiting for instructions. Triggers include group chat messages, email and calendar reminders, code review comments, and CI results.\n\nReal group chats are extremely noisy, so the team applies a two-stage filter borrowed from computer systems, the Fast Path \u002F Slow Path split. A programmatic layer checks deterministic conditions first: source, group scope, allowlists. Surviving candidates then go to an LLM that decides whether to act according to a strategy written in the prompt. That semantic layer doubles as a natural-language extension point, so behavior can be switched on and off through prompts at any time.\n\nConsecutive events under one GitHub pull request, one IM topic, or one email chain form a Thread. After the Router creates a Worker from the first event, that Worker subscribes to the Thread and later events are delivered to it directly, no longer passing through the Router. Only events nobody has subscribed to come back for the Router to dispatch. Context is managed like a cache: Inline pulls in information likely to be needed soon, while Spill keeps content that exceeds the budget outside the window as a reference only. Unique IDs on Triggers, Threads, and Workers link these into a reference graph, supporting on-demand multi-hop retrieval.\n\n## The ablation: 12.4 minutes against 29.5 minutes\n\nThe team's self-reported experiment drives group chat and GitHub CI message streams simultaneously. A single run contains about 300 messages, roughly 20 of which are work requests; coding work is replaced by a tool that only consumes time, so the measurement targets scheduling and collaboration rather than the execution quality of any single task. Message intervals and work durations are scaled to 1\u002F4 while model call latency stays real, and each configuration is repeated five times.\n\nBy the team's own numbers in the [original essay](https:\u002F\u002Fmp.weixin.qq.com\u002Fs\u002FzoS7GlReWPhfCjENGxXCdg), Super-agent's task completion time comes out at p50 12.4 minutes and p95 15.7 minutes. In the same ablation table, a single session with the Subagent tool under a neutral prompt reaches p50\u002Fp95 of 20.4\u002F29.5 minutes, and under a delegate-first prompt, 13.8\u002F29.0 minutes. The essay explains why: under the neutral prompt the model never invoked a Subagent across five runs, so all work stayed in the main session and was repeatedly interrupted by new messages. After switching to delegate-first prompting, four of five runs delegated, and the one run where the model judged the work \"small\" and did it itself produced the p95 long tail.\n\nThe authors themselves label these results preliminary reference only: the load is synthetic, the environment is simplified, and each configuration ran five times.\n\n## The bill arrives on the same table\n\nThe same table lists the other half of the ledger. Super-agent reads more cache tokens per run than the single-session setups; the essay gives only the direction, while a [third-party reconstruction of the same ablation table](https:\u002F\u002Fagentspulse.github.io\u002Ftutorials\u002Fdeepseek-super-agent-architecture) puts Super-agent at 23.23M tokens against 6.05M and 10.82M for the two single-session prompt configurations. Removing the Thread direct-delivery mechanism barely changes completion time, but the Router has to make about 30 percent more decisions and total read-cache tokens rise from 23.23M to 31.77M, roughly 37 percent more. Direct delivery does not speed anything up or improve quality; it simply saves the Router tokens and inference overhead.\n\nThe Router has a batching layer of its own. The team's data: as the arrival rate of Triggers needing a decision climbs from about 1 per minute to about 40 per minute, average batch size grows from 1.0 to 2.8 while per-request latency rises only from 2.5 seconds to 3.9 seconds, so inference time amortized per decision falls from 6.4 seconds to 1.6 seconds. On that basis the team concludes one resident Router suffices at their current team size, with no horizontal partitioning needed yet.\n\n## What the open prototype gives you\n\nThe repository is MIT licensed and written in TypeScript, positioned as a proactive coding agent scoped to a single repository per deployment. It listens to GitHub events and can optionally connect to Feishu. An incoming request becomes a work item, executes in its own session and directory, and returns to the original conversation with a usable result such as a pull request; a web panel lists only the items still waiting on a human, and in a healthy deployment that panel can stay empty. The repository reports 4.1k stars and 375 forks.\n\nThe design philosophy is stated plainly: take as little human attention as possible, and give as much value as possible. The rules that follow: act only when named or when a written team rule allows it, stay quiet while working, speak only when someone must decide or can use the result, and ask one concrete question at a time. The disclaimers are equally plain: this is a research prototype and not an official product, much of the engineering was done by coding agents, it may still contain bugs and its style may be inconsistent, and it is not identical to the internal version. Some designs discussed in the essay are not implemented.\n\n## Three things left unresolved\n\nSecurity is the hardest one: the Agent Loop runs in the cloud while only a lightweight daemon stays on the device, a design that assumes the cloud and the endpoint trust each other. Once a cloud Agent Loop is compromised through a server breach or prompt injection, every endpoint it is connected to faces the risk of being controlled. Even without a cloud compromise, credentials entering the model context risk leaking, and the team still needs a mechanism letting the agent use credentials without ever seeing them. The rest are long-term memory, how to present tasks to users once sessions are hidden, and the context boundaries and A2A collaboration between isolated instances inside one organization.\n\n## So what\n\nThis design proposes no new theory, as the authors say plainly: it carries classic computer systems engineering ideas into agent construction. What is genuinely interesting is the observation about 3 to 5 sessions. Over the past two years model capability flattened the barrier to any single task, and the bottleneck moved up to human attention. This time DeepSeek treats that as a problem solvable with system design, pushing session management, context synchronization, and event routing down into the architecture so that people only decide and accept.\n\nThe experiment is synthetic and ran five times, and the prototype says outright that it is not a product. The team's internal instance built on these ideas is already running continuously, taking part in issue reproduction and fixes, responding to code reviews, and helping iterate DeepSeek Harness itself. What this looks like once outside teams actually run it remains unanswered.","deepseek-super-agent-router-worker-architecture","2026-11-01T07:31:50Z","2026-11-01T07:31:56.963512Z","2026-11-01T07:31:56.963522Z",true,"agent",43,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"c10ebc48-1efe-409e-bfc3-7f00c6e04768","Velum 不是单文件 C++:它是 DeepSeek-V4-Pro 和 Claude Code 的协作样本","velum-deepseek-claude-code-collab-sample","2026-09-28T07:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"63c30bcd-3ffc-47c5-bd74-c2a9ed8f7c94","DeepSeek Harness 预览版开源:Agent 被拆成可插拔的插件栈,模型只负责想、Harness 负责做事","deepseek-harness-plugin-stack","2026-09-05T06:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"278dc1cc-ab13-4de3-9ed4-27f6732dda3c","五个27B组队靠账本编排追平Fable 5,复现只承认打平","gvs5h-ledger-orchestration-qwen-fable-rerun","2026-11-01T09:20:28+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"73e29aee-b368-423f-be27-7653f65b4775","DeepSeek DSec 公开:300 万沙盒日撑 V4.1 训练","deepseek-dsec-v4-1-sandbox-rl-training","2026-09-28T00:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"c696208b-6535-4eb9-b1ed-2e4f835d2f88","NVIDIA SoL-Pi 把 coding agent 的 token 砍掉 44%,harness 开始变天","nvidia-sol-pi-harness-token-compression","2026-09-19T03:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"32b938b6-01a3-43c9-b040-14db6c5f57c6","NVIDIA 把 Agent 装进一个 Python 类:被忽略的 NOOA,一半 token 跑出 SWE-bench 82.2%","nvidia-nooa-python-agent-framework","2026-08-23T17:20:00+00:00"]