今天就去看34
2026-09
S mcp-extensions OpenAI官方开源的ChatGPT插件开发套件,让MCP能力以原生特性形态进入ChatGPT,是平台级入口。一天208星且由OpenAI背书,几乎必然成为ChatGPT插件生态的事实标准依赖,商业化路径(插件市场、API计费)清晰。风险是生态政策变动与跨厂商标准分裂。
Build plugins that feel like native, first-class features of ChatGPT.
S Show HN: Raven – The harness of harnesses, built for RSI Raven自称'harness of harnesses',做自进化多智能体生态以支撑RSI(递归自我改进),4807星+HN 53分是本批最强早期动能之一。若RSI叙事兑现,可成为Agent编排的基础设施层。风险是概念超前、工程落地存疑,且与LangGraph等成熟编排框架正面竞争。
S Show HN: HN.watch – Videos of all Hacker News posts 把任意文章几秒内转成 HTML 渲染的解说视频,单条约 0.04 美元,把视频从'美元分钟'压到'美分秒级',这是新的内容成本结构。118 分加上 MCP、Chrome 插件、ChatGPT 插件多渠道分发验证了需求;Scrimba 是 YC 系成熟公司,可走 PR/文档视频 B2B 订阅。风险是画面粗糙、Loom/Canva 等大厂易复制。
Hi HN, I’m Per, founder of Scrimba (YC S20). We’ve spent the last decade teaching people how to code with an HTML-based video format. We’ve now plugged an LLM into it, so that people can create explai…
S PrunedCTC 把 CTC loss 的激活内存从词表线性增长压下来,同时保持 loss 与一阶梯度等价,直接利好大词表语音识别训练;命题小、描述准确。局限是只在单一设定下给出结论,还没有与主流 CTC 实现的实测对比。
Memory-efficient CTC loss with exact loss and first-order gradient equivalence under vocabulary reduction, and activation memory that no longer scales linearly with vocabulary size.
S jevgrep 为 coding agent 提供语义化代码检索:不搜关键字而是按'这段代码做什么'召回文件与上下文,直击 agent 上下文窗口浪费的痛点。743星/1.8天是本批最强信号,说明 agent 工具链正被快速采纳。商业化可走开源CLI+企业内网索引订阅,或被 agent 框架内置成标准依赖。风险是 Sourcegraph、grep.app 及 IDE 内置语义搜索已占位,且依赖 Jev 模型热度能
Find code by asking what it does. A CLI for coding agents that uses Jev to discover relevant files and source context.
S Show HN: Reladraw – A diagram language where you decide where to place things 解决 Mermaid/Graphviz 无法控制布局与 Draw.io 手动拖拽低效之间的空白:文本化但位置可控,且专为 agent 可编辑设计。两天 232 星加 HN 151 分是本批最强信号,playground 与 Claude skill 说明已有真实采用。商业化路径是设计工具与 agent 绘图标准入口;风险是 Mermaid 生态惯性与 Visio 类厂商跟进。
I love making diagrams to help understand, plan, etc. However, the options are (A) auto-placement languages like Mermaid or Graphviz (which don't let me decide how the diagram looks), or (B) software …
S HaloBattery 把罗技、雷蛇等无线鼠标耳机和蓝牙设备的电量统一显示在 Windows 托盘,解决一个非常具体的小麻烦。两天 170 星说明痛点真实,局限是只支持 Windows、各厂商协议覆盖要翻 README 才知道。
Unified battery level indicators for wireless mice, headsets and Bluetooth devices in the Windows tray
S Show HN: Whiteboard (YC W26) – An open-source IDE for thoughtful software design 基于 CodeOSS 的桌面白板,让 agent 通过 SDK 把架构图、序列图与代码位置双向链接,并用 Rust 语义 diff 折叠噪音,Salesforce、Modal 已有人用,直击 agentic coding 的认知负债。YC W26、MIT 开源、托管版收费路径清晰,风险是要与 Cursor、Claude Code 自带 review 正面竞争。
Hello! We’re Sid, Alex, Ketan, and Milan. We’re building Whiteboard ( https://whiteboard.dev.fast/), an open-source desktop app where humans and agents can architect software together in a common work…
S golive-skill 把 agent 写完的代码怎么上线这段最脏的活封成 Agent Skill:域名、数据库、邮件、支付全走用户自有账号,零依赖 CLI 做 detect→plan→apply 的可审阅流程。1.4 天 874 star 说明生态正缺这最后一公里。风险是 Vercel、Cloudflare 官方 Skill 与内置部署能力直接覆盖,自身难收费。
Take your agent-built product live: hosting, database, domain, email, payments — on your own accounts. Open-source Agent Skill + zero-dependency Node CLI: detect → plan → approve → apply → verify. No …
S Show HN: An open-source manufacturing ERP/MES/QMS 开源自托管的制造业 ERP+MES+QMS 一体系统,补的是中小工厂被 SAP、甲骨文高价闭源束缚的空白,open-core 与按产线授权是成熟工业软件路径,51 分说明需求真实。风险是实施重、交付依赖顾问,合规模块做不厚易被既有 MES 厂商挤压。
S jev-chat-jarvis 把读屏+意图判断+候选回复做成手机上的对话副驾,解决微信飞书里不会回话的痛点,不注入不封包降低了封号顾虑;1.5天3723星来自极强话题性。商业化可走订阅制输入法式入口。最大风险是平台ToS与隐私合规,且同族仓库刷量嫌疑会反噬可信度。
装在手机上的对话副驾:在微信 / QQ / X / 飞书里读懂对方、给出候选回复、一键填入输入框,发不发由你。非侵入,只读屏幕,不 hook 不改包。
S Show HN: Foremerge – Catch intent conflicts between parallel coding agents 在 git 之上用 intent 与 scope 声明做多 agent 并行改码的冲突预判,把语义冲突从 PR 阶段提前到写码前,并自带 MCP 与验收 gate。多 agent 同仓并行是当下真痛点,451 星加 33 分证明工程圈共鸣,有机会成为协作协议层标准;风险是 git 厂商或 Cursor 自带协调能力。
At, GPTree, we run several coding agents across our team on one repo using parallel worktrees. Apart from wasted time reviewing and fixing conflicts at PR time, the failures that hurt the most are whe…
S preflight 本地代理在请求送出前扫描 LLM 输入与附件里的密钥,解决员工把凭证贴进 AI 导致泄露的高频事故。作者有影响力,企业 AI 安全预算明确,有机会做成 DLP 标准层;竞品是各厂商自带隐私过滤与 Zscaler 类安全网关。
A local proxy that scans LLM requests and attachments for secrets before they reach the model.
S Show HN: CUA-S1 – A System One Model for Computer Use 针对 computer-use 里大量局部决策(勾选、跳过、填值)训练 70 万参数小模型,把 System1 专用判断从通用大模型循环中剥离,单次 7ms 对 260ms;背靠 2.4 万星 cua 与开源驱动、评测基准,HN 56 分说明 agent 厂商在真关注;风险是 jev 系服务先成标准。
Hello HN! We're Dillon and Francesco from Cua. We were wondering how many computer use tasks actually need a full general purpose LLM (e.g. gpt-6-astra, claude-opus-5 etc.) to think through all their …
S Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash 把工具调用与结构化 JSON 输出压到 8–29MB 的 2 位模型,可在树莓派与浏览器 WASM 里跑到 4k tok/s,解决端侧 agent 的部署成本。第三次迭代拿到 156 赞,说明团队在持续兑现反馈。可走端侧推理 SDK 授权。风险是能力窄,且被 llama.cpp、MLX 等边缘运行时挤压。
Hey HN, Henry from Cactus here. We submitted Needle 2 here a few weeks ago, and the feedback in the discussion thread was incredibly valuable, thanks! Thanks to all that feedback, we’ve been able to m…
S Show HN: Swift-Qwen3.8-27B, -58.3% thinking, x1.95 speed, accuracy of xhigh 对 Qwen 27B 后训练,用 on-policy 蒸馏惩罚与过度思考相关的 token,think 长度减半而精度几乎不掉,直接降低推理成本。3 天 80k 下载并有社区量化与独立评测,动能强。变现靠 API 与授权(仅对千万美元以上公司收费);风险是基础模型自带 effort 控制会吞掉该价值。
Hi everybody, we post-trained Qwen 3.8 27B to be more efficient by figuring out which tokens were linked to overthinking and penalizing them without "attacking" the reasoning length directly then fixe…
S knoxmap 把真实地点转成《僵尸毁灭工程》B42 可玩地图,并明确致谢上游 Knoxify;对 PZ 模组作者和想在家门口打僵尸的玩家是少见的真需求,局限是依赖上游工具链、产出质量随地理数据波动。
KnoxMap: turn any real place into a playable Project Zomboid Build 42 map (built on Knoxify)
S Show HN: Capsule – Single-file web apps that save their data into SQLite 把 HTML 应用、资源与 SQLite 数据打包成单个可分发文件,解决静态页面无处存数据的长期痛点,还配权限模型与本地 AI。267 赞为本期最强动能,1.0 开放格式有望成为标准,商业化可走应用市场与企业分发,风险是 WASM 沙箱方案替代。
Hey HN, I always had the problem that building HTML pages is really simple now, but trying to save data required hosting it somewhere, and sharing it afterwards was not easy. Over the last few months,…
S DeepSelect DeepSeek 官方开源的稀疏注意力 TopK 与采样 CUDA kernel,直接决定 DSA 推理的算力与显存开销,任何复现或部署该模型的团队都要用;官方背书易成事实标准,可能被 vLLM 等框架吸收,风险是只适配自家架构。
DeepSelect: TopK kernels for DeepSeek Sparse Attention (DSA) and Samplers
S maskit 本地网关在请求发往大模型前自动打码、流式回复再还原,任何能改 Base URL 的工具都能接;少数做到「装上就用」的隐私工具,局限是脱敏靠规则,复杂实体仍会漏。
数据面具 Data Maskit — 专为大模型打造的本地隐私脱敏网关(请求自动打码,回复流式还原;支持 Cursor / Claude Code / Codex / Pi 等任意可配 Base URL 工具)
S Show HN: Geiger – See every AI agent on your machine and what it can touch 解决AI agent激增但难以追踪的问题,提供只读命令清点agent、MCP服务器和插件。44分HN和68星表明安全圈高度关注。商业化路径是成为安全审计或资产管理工具的一部分,或提供SaaS服务。风险是AI生态变化快,或操作系统内置类似功能。
S Show HN: Self-hosted company OS, Claude Code and Codex agents in departments 提供自托管多租户AI agent平台,让团队用Claude Code/Codex协作并限制在沙箱内。发布0.3天获95星和37分HN热度,动能强劲。商业化路径类似'公司OS',可做团队订阅或私有化服务。风险是市场竞争激烈,且依赖Anthropic/OpenAI订阅,受平台政策影响。
Hi HN, I am Dimitris. This is a company OS that I built and use to run my business and anyone can install it and self host it for free. Think of it as Claude Code, Cowork and the cloud sessions in one…
S Show HN: Give your AI agent on-screen guides that show users where to click 解决AI agent不理解自身产品UI的问题,用浏览器代理学习应用结构并生成屏幕引导。17分HN显示关注,来自YC初创。商业化路径是作为SaaS的assist API销售,按调用计费。风险是RAG方案不断改进,或企业倾向自训练,浏览器代理的安全性和维护成本也需考虑。
Hey HN. I'm Christian, one of the founders of Frigade (YC W23). I've noticed that a lot of in-app AI agents struggle to actually understand the products they exist in. For instance, let's say a user a…
S glooprint 虚幻引擎插件,一键把 Blueprint 节点图自动排版并重新布线,按 F 就清爽整张图。蓝图凌乱是 UE 开发者的日常痛点,切口小而准;局限是只服务蓝图,稳定性还得进编辑器实测。
An Unreal Engine plugin for automatic Blueprint graph formatting and clean/performant wire routing. Press F to tidy up your entire graph.
S Show HN: Copperhead – Cursor for circuit boards 用AI辅助电路板设计,类似Cursor但面向硬件。HN 218分说明硬件工程师渴望。商业化可做订阅制PCB设计工具,市场大。风险是EDA巨头竞争,竞品是Altium/JLC的AI功能。
S Show HN: Sparrow-2 – Noise cancellation isn't designed for conversational AI 解决语音 AI 在噪声环境下的 turn-taking 与对话流理解,将呼吸、环境音等非言语信号纳入建模;HN 11 分且声称 Turnbench 第一,来自 Tavus 专业团队;可直接以 API/模型商业化,嵌入客服、语音助手;风险是 OpenAI Realtime、Deepgram 等竞品及推理成本。
Hey there, I’m Brian. I've been shipping conversational models at Tavus for the past two years. I want to tell you about our latest audio-understanding/turn-taking model: Sparrow-2! It’s a new categor…
S okf-agent-memory 为 AI 编码智能体提供 Git 原生持久记忆,实现 OKF v0.2 与亚毫秒 BM25 检索,内嵌 MCP 服务器;348 星/天说明踩中 agent 记忆痛点;可成为 Claude Code 等工具的标准依赖,但竞品有 Mem0、Letta,且 OKF 标准未定,需观察生态。
Git-native persistent memory for AI coding agents. Implements Google OKF v0.2 with sub-300µs in-memory BM25 search, embedded MCP server, and progressive disclosure. Slashes token bloat by 80% with zer…
S Show HN: ZSvirt – A lightweight, scalable open source virtualization platform 轻量可扩展的开源虚拟化平台,替代重量级方案,满足中小企业IaaS需求。进入Hacker News首页并获71点,1563星表明社区强认可。商业化路径明确:可做私有云方案或云服务商集成。风险是架构复杂度高,与KVM/Proxmox竞争激烈,需要生态支持。
S commerce-agents 提供用Claude构建购物和商家agent的参考蓝图,覆盖零售、电商等场景。Anthropic官方背书,加上Shopify示例,使其快速获得300+星。商业化路径明显,可作为企业定制agent的标准依赖。最大风险是早期版本不稳定,以及来自其他agent框架如LangGraph的竞争。
Reference blueprint for building shopping and merchant agents with Claude. Examples in retail, commerce, telecom, and entertainment included.
S claude-for-commerce-examples Shopify对Commerce蓝图的官方实现,提供商店购物和商家agent的参考代码。与Shopify生态深度绑定,对企业实用。34星虽少但增长良好。商业化路径清晰,可成为Shopify应用标准参考。风险是依赖Shopify API和Anthropic模型演进。
Shopify implementations of Anthropic's commerce-agents blueprint: a storefront shopping agent over UCP and Sign in with Shop, and a merchant agent over the Admin API
S Show HN: Aura – a Rust agent that investigates and fixes production incidents Rust编写的SRE agent平台,能调查并修复生产事故,有HITL和权限控制。真实生产验证,解决上下文溢出和审批疲劳。成长逻辑是DevOps自动化需求旺盛,Rust性能优势明显。商业化路径清晰,可作企业级SRE产品。风险是切入市场慢,与PagerDuty等协作工具有竞争。
We run a SaaS that handles petabytes of data. Our SRE team experimented with using claude, openclaw, langchain, etc. within our incident response workflows. We struggled with overflowing context, leth…
S agent-fleet-manager 为大规模重复信息收集提供通用引擎,可管理多个worker agent并行执行采集。168星增速显示需求真实。商业化路径清晰:可作SaaS或企业内部数据采集平台。风险是与其他agent编排框架竞争,且需解决数据质量和反爬合规问题。
A general-purpose engine for large-scale, repeated information gathering by a fleet of workers.
S data-deidentification 纯前端跑的文件去识别化工具:本地检测 PDF/Word/Excel 里的个资并遮罩复核,再用 CSV 编码表还原,数据不出机器。对不能上云的 HR、医疗、法务场景很实用;局限是识别规则覆盖面与大文件的前端性能。
純前端文件去識別化工具:PDF/Word/Excel/TXT/Markdown 自動偵測個資、遮罩預覽、人工覆核、CSV 編碼表可還原,資料不出電腦
S Show HN: 49 IDE – 2D Canvas for Agents 解决并行多AI CLI的上下文碎片化问题,用2D画布统一管理终端、git与issue。HN热帖+411星说明切中真实痛点。可做成标准AI开发工作台,商业化路径为订阅制。风险是AI CLI生态碎片化,竞品如Cursor/终端多路复用器。
Track agentic CLIs across multiple providers and repos, from any device and across multiple machines, see git trees, terminals, usage, issues, and UI all on one workspace. I have been trying to parall…