DeepSeek unveils Super-agent architecture for cross-session multi-task coordination
DeepSeek publicly shared its architectural vision for a next-generation agent harness, proposing a 'Super-agent' concept that maintains context across sessions to coordinate multiple tasks concurrently. The approach aims to overcome the limitations of single-session agents, which constrain multi-task progress by human attention and require manual session switching. Internal instances built on DeepSeek Harness reportedly continue tasks across IM and GitHub, and can proactively identify and fix defects from public chat feedback without direct mentions.
Coverage timeline
机器之心机器之心
本文是一篇面向工程实践的架构思路探讨，并非对 DeepSeek Harness 现有产品功能的介绍。 1 . Super-agent 过去一年里，以 Coding Agent 为代表的单任务 Agent 已经证明，模型能够使用工具解决复杂问题，且越来越少需要人类持续监控和逐步干预。与此同时，模型的多 Agent 协同能力也在提升，使跨会话的组织与调度变得可行。随着人类在底层执行中的参与减少，人的角色正逐步上移到更高层的决策与验收。 模型能力的这一提升，也催生了新的应用形态。近期个人助手（Personal Agent）开始集中涌现，它们直接接入日常的消息、日程、邮件与工作平台，持续观察信息流并协助处理事务。这种持续在线、事件驱动的工作模式并不局限于个人场景，个人助手只是其中一类应用；在软件开发等商业场景中，它同样很有价值。 然而，现有的主流 Agent 设计还难以在多任务和团队协作的规模上支撑这一模式。Agent 的工作模式仍然以单一 Session 为基本单位，不同 Session 之间彼此独立。当用户并行推进多项工作时，必须手动切换并记住各个 Session 的上下文，而据我们观察，一个人通常只能同时跟进 3 到 5 个活跃 Session。在这种设计下，单任务的上限取决于模型能力，多任务的上限取决于人的注意力。 如果只在现有的本地单会话 Agent 上借助 CLI、MCP 或外部连接器做简单扩展，Agent 虽然也能接入外部消息，但仍受制于单会话同步阻塞、多任务上下文混乱、本地设备离线即停止运行等问题，需要大量人工维护。 要突破这些限制，需要打破 Agent 与单次手动会话的绑定，使其成为一个能跨会话保持上下文、同时推进多项复杂任务且互不干扰的连续实体。我们把这种 Agent 形态及其系统架构称为 Super-agent。 例如，在我们内部基于 DeepSeek Harness 运行的 Super-agent 实例中，同一个开发任务可以在 IM 和 GitHub 之间连续推进；这个实例还会持续观察公共群聊中的反馈，在无人 @ 的情况下主动定位根因并提交修复 Pull Request： 同一个任务在 IM 群聊与 GitHub Pull Request 评审之间连续推进 持续关注公共群聊，在无人 @ 的情况下主动识别缺陷并提交修复 Pull Request 具体来
Related stories
DeepSeek releases V4.1-Flash-1017 open-weight model with faster, cheaper agent tasks
DeepSeek released V4.1-Flash-1017, an open-weight model checkpoint with 763B parameters, now live on its website, app, and API. The update claims stronger benchmark performance, faster task completion, and shorter outputs than its predecessor, achieved via reinforcement learning with self-feedback for end-to-end project development. DeepSeek also opened its Search & Fetch APIs to everyone, priced at $0.75 per 1000 successful requests, and requires DeepSeek Harness v0.2.1-rc.1 or later for optimal use.
9 reports · 6 sources955 points · 717 comments
DeepSeek open-sources DeepInfer inference stack, reports 82% gross margin
DeepSeek released its internal inference framework DeepInfer under the Apache 2.0 license, along with the token codec component deepseek-recipe. The framework, which supports DeepSeek-V4.1-Flash and targets Nvidia SM100 GPUs and Huawei Ascend 950 NPUs, reportedly achieves a throughput of 155 trillion tokens per day per cluster, with API revenue of $38 per GPU-hour at off-peak pricing, implying a gross margin of 82% after GPU rental costs. The release sparked discussions about the profitability of AI inference services.
3 reports · 3 sources2335 points · 890 comments
DeepSeek global outage hits chat and API for nearly 5 hours; token compensation issued
On October 15, 2026, DeepSeek's web chat and API services experienced a global outage starting at 8:11 AM Beijing time, affecting DeepSeek V4 Pro API and V4.1 Flash API. The issue was resolved around 12:50 PM, lasting nearly five hours. As compensation, DeepSeek credited tokens to developers based on their API usage over the past 7 days, with some users reporting credits of over 300 yuan.
Google Researchers Propose StreamVibe for Proactive Video Agents to Understand Evolving User Intent
Google researchers and collaborators introduced StreamVibe, a framework for proactive video agents that observe continuous video streams to identify explicit, implicit, and dynamically evolving user needs in real-world interactions. The team also built an evaluation and training system covering these intent types, aiming to move agents from passive instruction-following to proactive assistance that adapts as tasks progress.
News updates every hour. Subscribe in your RSS reader to get new stories automatically.
Subscribe