Skip to content

#产品更新

14 today

Oct 21Wednesday

TechCrunch · AI

Google has joined Amazon in attacking Nvidia’s stronghold with AI app developers

Google 周二宣布开源 TorchTPU,让现有 PyTorch 应用只需修改配置、无需重写代码即可迁移至 TPU 和 Google Cloud。Amazon 周一发布的 TorchNeuron 同样支持应用迁往 Trainium 和 AWS,并向 PyTorch 主项目提交解除部分功能与 Nvidia 芯片绑定的代码。

AI HOT · Products

Databricks 发布 MemCurrent 生产级 AI 智能体持久记忆架构

Databricks 推出 MemCurrent 持久记忆架构,已用于 Agent Bricks Memory 和 Genie。它结合保存原始会话的情景记忆与提炼经验的派生记忆,通过插件、按需工具和离线 dreaming 流水线读取、写入及整理记忆。在来自真实企业任务的内部基准上,相比无记忆方案,通过率从 22.3% 提升至 68.4%,token 用量减少 31%,运行速度加快 26%。

AI HOT · Products

Perplexity Computer now runs fully locally on Mac, no cloud inference needed

Perplexity CEO Aravind Srinivas says Perplexity Computer can now run entirely on a Mac, with no cloud inference or WiFi. Both the model and the agent run on the machine, and it can operate native Mac apps and local files, so sensitive data doesn't have to leave the device. The feature is available now in the Perplexity Mac app.

Why it matters: Running the model and agent on the Mac itself means sensitive data stays local while it works with native apps and files.

Google Blog · Gemini

Gemini adds a business hub, marketing and finance tools for small business owners

Google is rolling out a small-business hub in Gemini with marketing and finance skills and new app connections, opening to all Gemini app users worldwide over the next few days. Owners can upload their website, brand kit, fonts or color palette, and Gemini keeps the brand style when generating visuals with Nano Banana 2.1, so specs don't need re-uploading.

Why it matters: Gemini pulls brand asset generation, financial analysis and business app connections into one small-business hub, cutting the repeat uploads and manual reading of financial numbers.

Hacker News front page

Show HN: Run Claude Code unattended inside a fence, under a dollar budget

SIwithin 可在指定目录和美元预算约束下无人值守运行 Claude Code 或 OpenAI Codex CLI,并为每次运行留下成本及完成状态记录。工具仅支持 macOS,Claude 适配器通过本地回环 API 代理隔离真实密钥、计费并限制网络访问,但跨过预算的单次响应仍可能造成超支。

Hacker News front page

Launch HN: Respan (YC W24) – Open-source Rust AI gateway built around plugins

Respan 发布开源 Rust AI 网关,以小型核心和公开接口支持插件化扩展,目前处于 pre-alpha 阶段。供应商适配、MCP、实时语音、图像与视频,以及智能路由、回退和个人身份信息脱敏均可通过插件实现;跨协议转换并不覆盖所有功能。

Hacker News front page

Show HN: Penmentor – Claude marks up your handwritten math like a teacher

Penmentor 将 Claude、Gemini 或 GPT 接入 xournal++,按键即可在手写数学页面上标注错误、给出提示并生成交互动画。它通过 Claude Code、Codex 或 Antigravity CLI 使用已有订阅,支持整页读取、分级提示和基于 manim 的动画;演示视频使用脚本控制笔迹及固定模型回答。

Hacker News front page

Show HN: Jev Cards, Can you beat a vision model at a picture quiz?

JMS Code 在 GitHub 发布 Jev Cards,一款让玩家与视觉模型 Jev 1.14 同图竞答的看图问答游戏,覆盖水果、国旗、地图、国际象棋和视错觉等 490 张卡片。Jev 1.14 是决策模型,一次调用即读取图片与问题,为每个选项返回概率,约四分之一秒作答且不生成文本。游戏提供限时冲刺、10 秒一题的 Versus、连胜和好友同题实时对战模式。

Oct 20Tuesday

AI HOT · Tips & opinions

Google 发布 TorchTitan TPU 分叉及模型优化指南

Google 发布适配 Google Cloud TPU 的 TorchTitan 分叉,提供分布式训练配方及按步骤执行的模型优化指南。SimpleFSDP、Graph Trainer 和 Pallas 内核支持分片、整步图捕获及计算优化,FLUX.1 案例的训练步耗时从 11.49 秒降至 2.78 秒,达到同硬件 JAX MaxDiffusion 吞吐量的约 99%。

Hacker News front page

Tsrs: A Port of TypeScript to Rust

Cloudflare、VoidZero 与 Owner.com 的 Max Schwenk 合作开发 tsrs,将 TypeScript 编译器移植到 Rust,实验版本已可通过 npm 安装。

The Verge · AI

Adobe Lightroom mobile can now open closed eyes in photos with a simple tap

Adobe 宣布 iOS、iPadOS 和 Android 版 Lightroom 移动端可用生成式 AI 一键睁开照片中闭着的眼睛,检测到人物闭眼时会自动显示 Open Eyes 快捷操作。该功能会参考你 Lightroom 图库中同一人物的其他照片,让替换后的眼睛在颜色、大小和形状上更匹配。免费版用户可试用两次,之后需订阅包含更多 AI 额度的 Adobe Creative Cloud 计划。

AI HOT · Products

Cursor ships Agent SDK to define agents in code and deploy with one command

Cursor released its Agent SDK, which lets teams define agents in code and deploy them to Cursor infrastructure with agent-sdk deploy. The framework handles events like pull requests, task handoffs and parallel sub-agents, and includes built-in evals and event replay. Prompts and tools go through version control and review. Agents can also run as a single Node process on your own infrastructure.

Why it matters: Cursor's Agent SDK turns prompts, tools and evals into a code project, so custom agents follow the same version control and review as other software.

TechCrunch · AI

Yelp can now call on your behalf to check wait times at a restaurant

Yelp 新增 Check for Me 功能,由语音智能体代用户致电餐厅查询等位时间,并通过短信返回结果。用户在餐厅页面输入用餐人数和电话号码即可使用;Yelp AI 助手也新增 OpenTable 订位集成、偏好记忆和自定义提醒,并支持查看过去 30 天的聊天记录。

Hacker News front page

DeepInfer: High-performance, large-scale inference framework for DeepSeek models

DeepSeek 开源支撑其大规模生产服务的推理框架 DeepInfer,本次提供支持 DeepSeek-V4.1-Flash 的离线推理能力。框架支持优化内核、异步调度、确定性采样和结构化输出,可运行于 NVIDIA SM100 GPUs 与华为昇腾 950 NPUs,采用 Apache License 2.0。

Hacker News front page

Fizzpoker – No-login Online Poker with friends

Fizzpoker 推出免登录的浏览器端 No-Limit Texas Hold’em,创建 2–9 人牌桌后分享链接,好友点开即可入座,无需下载或注册账号。游戏为锦标赛式冻结赛,盲注每 3、5、10 或 15 分钟上涨,筹码无现金价值。免费账号可解锁最多 9 人牌桌、更多机器人对手及赛后基于 GTO 模型的 Fizz Score 复盘。

AI HOT · Products

Cursor launches Agent SDK for defining and deploying agents from a code folder

Cursor launched its Agent SDK, which lets teams define agents in a code folder, test them with evals, and deploy to Cursor or their own infrastructure.

Why it matters: Cursor's Agent SDK organizes prompts, tools and skills as a code project, testing agent behavior with local evals and replays of real pull requests.

TechCrunch · AI

Microsoft says AI agents are like insider threats as ‘frontier models can’t be trusted’

Microsoft 周一推出 Guardian Agents 等企业安全功能,将企业部署的 AI 智能体视为潜在内部威胁。Guardian Agents 属于 Project Perception,用于监控企业使用的智能体;Microsoft 安全业务负责人称,前沿模型因行为不可预测而不能被信任,企业应将智能体供应与监管权限分离。

TechCrunch · AI

Amazon is taking its biggest shot at Nvidia yet with new TorchNeuron dev tool

Amazon Web Services 周一发布 TorchNeuron,让使用 PyTorch 编写的 AI 应用更容易迁移到其 Trainium 芯片。Amazon 称只需修改一行配置,无需重写代码,并以 Deepgram 为例称其在生产级准确率下每美元可处理的音频增加了 70%。Amazon 还计划向 PyTorch 上游贡献代码,减少框架与特定芯片的绑定,让开发者拥有更多硬件选择。

TechCrunch · AI

Replika’s founder built the AI assistant she pitched at Disrupt 12 years ago

Replika 创始人 Eugenia Kuyda 周一向所有用户推出 Wabi,将 12 年前的生活助理构想发展为支持多人共享、多线程任务的个人 AI 助手。Wabi 2.0 接入邮箱后会自动创建线程和应用,用户也能请求定制应用,并在共享聊天中按需开放个人信息,访问权限限定于具体聊天。Wabi 目前免费,商业模式尚未确定,公司正在洽谈进一步融资。

Oct 19Monday

Hacker News front page

Show HN: KiviDB, a faster Redis-compatible memory database and vector cache

KiviDB 是用 Rust 编写的开源多线程内存数据库,兼容 Redis 协议并内置全文与向量搜索。项目公布的 32 核测试中,KiviDB 1.1.0 的流水线 GET 吞吐为 21.2M ops/s,约为 Redis 8 的 9×、Dragonfly 的 2.3×,测试结果依赖环境与工作负载。

Hacker News front page

Superlite: A Faster Alternatie to Superpowers

Superlite 是作者开发的 Claude Code 插件,通过并行任务、分层智能体和代码钩子,将规格交给智能体后自动完成开发、测试与审查。当前版本单次基准用时 16.1 min,通过 177 / 177 项隐藏测试,相比 superpowers 原版此前 3 次运行的平均 33.6 min 减少 52% 时间;这一比较来自同机测试,并非多次重复的新版本结果。

Hacker News front page

I built a programming language that LLMs can't use

开发者构建并开源 KillSwitch,通过对抗性语言设计测试 LLM 和编码智能体能否理解、编写未见过的代码。它是一种基于栈的解释型语言,利用字符计数、规范中的隐藏谎言及触发安全护栏的语句增加难度,示例见 https://killswitch-lang.org 。

Simon Willison

llm-echo 0.5

Simon Willison 发布 llm-echo 0.5,这个调试插件提供一个不涉及 LLM 的“模型”,只把收到的提示词原样回显。新版本更新了模拟 token 用量统计,将工具调用、系统提示词和推理 token 纳入计算。示例中 `llm -m echo 'hi there' -o thinking 1 --usage` 显示 2 input、24 output。

Simon Willison

llm-tools-datasette 0.1.1

llm-tools-datasette 发布 0.1.1 版本,修复了 LLM 0.33 改用 httpx2 后插件因未直接依赖 httpx 而出现的兼容问题。该插件此前依赖 httpx 但未显式声明,此次升级后已恢复正常。

Hacker News front page

Show HN: WhatsApp MCP running free on Cloudflare

WhatsApp MCP 可部署在用户自己的 Cloudflare 账户,让 Claude Code、Claude Desktop 等 MCP 客户端读取和搜索 WhatsApp 聊天,但不能发送消息。

Oct 18Sunday

Hacker News front page

Show HN: Sphinx – A hook that quizzes you on your AI's code before shipping

Sphinx 是面向智能体辅助开发的免费开源 Git 钩子,在受保护分支推送前根据代码差异提问,答对后才继续推送。独立审查模型不读取编程聊天记录,回答不会传给编程智能体,也不会打开 .env 文件和密钥;可用 sphinx yolo 跳过检查,每次跳过都会记录。

Simon Willison

web-edit 0.1a2

Simon Willison 发布 web-edit 0.1a2,为其用于通过 Tailscale 跨机器浏览和编辑文件夹内文件的微型 Web 应用新增文本与图片数据的"复制到剪贴板"选项。该功能源于他用手机在笔记本上运行 pelican 提示词、并希望便捷地将日志复制到 Gist 的需求。

Hacker News front page

Show HN: Things to Do API

Things to Do API 提供面向智能体应用的本地活动数据接口,覆盖演出、讲座、快闪、市集和节庆等,支持 REST 与 MCP 接入,并可用自然语言查询。免费额度为每月 1,000 次调用、每分钟 60 次,无 key 时每天可试 50 次查询;按量计费 $5/1,000 次,超过 10 万次后 $3/1,000 次,超过 100 万次后 $2/1,000 次。

Oct 17Saturday

DeepSeek · API updates

DeepSeek opens its in-house search API

DeepSeek has opened its in-house search API, built on the DeepSeek search engine that already backs search on its web product and API. The engine indexes hundreds of billions of web pages worldwide, supports multilingual search and handles hundreds of millions of search requests a day from around the world. DeepSeek says it filters SEO cheating, GEO manipulation and low-quality AIGC through identification mechanisms. Usage is covered in its web search and crawling guide.

Why it matters: DeepSeek's search engine already serves its web and API products and handles hundreds of millions of requests a day, and now outside developers can call it.

The Decoder

ChatGPT's Chrome extension can now read and act on open tabs

ChatGPT's Chrome extension can now read open tabs and take actions in them, and the update applies to all ChatGPT plans. Users can point it at a tab with @chrome, Chat mode pulls data across tabs, and Work mode uses Computer Use to send messages, fill forms, download files or group tabs. The upgraded sidebar shows full chat history. OpenAI's examples include comparing products across tabs, summarizing articles and checking price pages from different sellers.

Why it matters: ChatGPT now reads across tabs and acts inside the browser from its Chrome extension, so page comparison, form filling and file downloads happen in one sidebar.

Hacker News front page

PostgreSQL 19 delayed as AI review finds more defects, 8–10 features pulled

The PostgreSQL Global Development Group pushed PostgreSQL 19's release to around November 19 and withdrew 8–10 significant features. LLM-assisted review surfaced edge-case defects faster, and the fixed cost of handling each fix ate into the pre-release time budget.

Why it matters: LLM review is finding edge-case bugs faster, and the fixed cost of each fix is squeezing the time left before release.

Oct 16Friday

AI HOT · Products

NVIDIA RTX Spark laptops go on sale today, up to 128GB unified memory

NVIDIA's RTX Spark laptops went on sale worldwide today, with up to 128GB of unified memory and 1 petaflop of FP4 AI performance, starting at $2,599. The machines pair a Blackwell RTX GPU with a Grace CPU and CUDA, run large language models and AI agents locally, and add agent permission controls plus inference serving over a home network.

Why it matters: RTX Spark puts local model running, coding and testing inside a Windows laptop, cutting how much the dev loop depends on cloud test environments.