Fish Pocket - iOS DeepSeek Harness
Run DeepSeek Harness on iPhone and iPad, with a fully native SwiftUI interface. No computer, no server. 在 iPhone 和 iPad 上本机运行 DeepSeek Harness,界面完全原生。
OriginalDEV TRENDS / DAILY DISCOVERIES
Daily discoveries from developer communities, independent makers, and open source.
Run DeepSeek Harness on iPhone and iPad, with a fully native SwiftUI interface. No computer, no server. 在 iPhone 和 iPad 上本机运行 DeepSeek Harness,界面完全原生。
Original远程桌面软件,在手机上输入设备 ID 与访问码连接自己的电脑,支持触控、键盘操作、同账号设备管理和云收藏;普通单人连接免费,含 VIP 付费权益 - 下载 · 开发方介绍与体验招募
OriginalAI 大模型榜单,汇总 Artificial Analysis 等第三方公开评测,把能力分、单任务成本和速度放在一张表里比较,每个模型有标注来源的详情页,每周更新,中英双语
OriginalA weekly radio built from your YouTube Music, NetEase Cloud Music or QQ Music likes: Claude picks new songs and says why. Paste a playlist link for a no-login roast of your taste. English, 简体中文, 繁體中文.
OriginalAccurate and fast voice input for Mac/Windows/Linux
在自己的 Windows 10/11 电脑上给照片和长视频换脸,4 个模型可选,实时预览,NVIDIA/AMD/Intel 显卡加速,素材不上传;安装和预览免费,导出需订阅
OriginalOne runtime, multiple chatbot plugins ecosystems. Reuse native plugins without rebuilding their worlds.
多模型 AI API 网关,一个 Base URL 调用 Claude / GPT / Gemini / DeepSeek / 千问等文本模型,以及 Seedream、Seedance、可灵等图片和视频模型;兼容 OpenAI、Anthropic、Gemini 三种接口格式,Claude Code / Codex 改 base_url 和 key 即可使用,按量计费 - 接入文档
OriginalOpen-source, local-first video editor your AI agent can drive over MCP (Claude Code, Codex, Cursor). Windows · macOS · Linux. The AI makes the cut; you finish it on the same timeline.
大家好,我们是 FlyDesk 飞桌的开发团队,向周刊自荐这款正在持续完善的远程操作工具,也希望邀请有实际需求的朋友参与体验并反馈问题。
Original(英文版: Models Rank(模型榜)汇总 Artificial Analysis、Arena 等公开的第三方评测,把 AI 模型的综合能力、完成一道评测题的平均成本和速度放在同一张表里比较。网站本身不跑分,每个数字都标了来源和核对日期,每周更新。 做这个网站,是因为挑模型时经常要在好几个榜单之间来回切换,而且很多榜单只列每百万 token 的标价。这里的"成本"用的是 Artificial Analysis 测出来的平均单任务成本,更接近实际花费。60 多个主流模型各有一个详情页。 我是作者,欢迎指正。
Original我想自荐一个自己做的 Mac 菜单栏小工具,叫 Split。 两个窗口并排的时候,中间那条边往往要两边分别去拖。Split 会把两个窗口配成一对,中间给出一条共享分隔线。拖这条线,左右或上下两边会一起变大变小。点一下分隔线,两边对调。 也可以先在菜单里选好比例,再用菜单或快捷键把当前窗口放过去,这个比例下次还在。窗口拖到屏幕边缘或角落,会先预览落点,松手再就位。同一侧还可以叠几个窗口,点旁边的数字就能切换。 它不管 Spaces 或多桌面,摆的是当前这块屏幕上的窗口。支持中文界面。 闭源软件,$4.99 一次性买断,不是订阅。需要 macOS 13 Ventura 或更高版本,Apple 芯片和 Intel 都可以。第一次用要在「系统设置 → 隐私与安全性 → 辅助功能」里给 Split 开权限,不然它动不了窗口。
OriginalRust GUI components for building fantastic cross-platform desktop application by using GPUI.
Random animal generator website with a curated photo library.
阮老师好,自荐一个我们做的工具 Imagera Code,我在 Imagera 工作。 产品页: 安装:npm i -g imagera-code(macOS / Linux / Windows) 它做什么 在终端里描述要改的东西,它读取项目、修改文件、运行命令和测试,然后给你看 diff。也可以非交互运行(imagera-code run),放进脚本和 CI;可以把 MCP 服务器当作工具使用。 和同类工具的区别 一个订阅可以用 276 个编程模型,包括 GPT、Claude、Gemini,不用分别开几家的会员 没有 3 小时锁定,只有每周额度;超出后可以用积分继续 $20/月起,闭源商业产品 谢谢阮老师! 🤖 Generated with Claude Code
OriginalA Windows GUI for cleaning up stale Git branches — bulk delete with backup, restore anytime
Nenrin 是我做的一个日记 App,有 Mac 版和 iPhone 版。 它存日记的办法跟一般的日记 App 不太一样。一年里每个日期对应一个 Markdown 文件,最多 366 个。8 月 18 日的日记都在 08-18.md 里,一年一节: 08-18 2021 第一次一个人搬进新家。 2024 把去年今天的日记翻出来读了一遍。 2026 今年在同一天,写下第六圈。 明年这一天再写,同一个文件里就多出一节 ## 2027。所以翻到哪一天,往年的这一天都在同一页,去年的和五年前的放在一起读。写的年头越多,这一页越长,有点像树的年轮,名字也是从日语的「年輪」来的。 文件就是普通的 .md,默认放在 iCloud Drive 的 Nenrin 文件夹里,也可以换成别的文件夹。其他编辑器也能直接打开来改,哪天不想用了,把文件夹拿走就是。App 不用注册账号,我这边也没有服务器。 Mac 和 iPhone 打开的是同一个文件夹。保存时只替换今年那一节,两台设备同一天都写了也不会互相覆盖。 其他做了的功能: 当天在「照片」里拍的照片会显示在日记下面,只读,不会写进文件。 年历里一格是一天,写过的年份越多,颜色越深。 日记里写 #旅行 就是标签。搜索可以组合着来,比如 #旅行 京都 2025。 可以导入别的日记 App 的导出、Markdown 文件夹,或者带日期的 TXT、PDF,每篇放回它原来那一天。导入之后还能撤销。 iPhone 上有「历年今日」小组件和每晚提醒。 Mac 版完全免费,在官网下载,需要 macOS 14 以上。iPhone 版也是免费下载,需要 iOS 18 以上。里面有一个 Pro 内购,买不买功能都一样,算是支持一下开发。
OriginalA fresh take on Markdown: a fast, native rich-text editor built in Go with MyGo. Inspired by Arc, with colorful spaces, split views, and local-first notes.
A free and open-source database management tool, suitable for team use. It offers capabilities such as access control, data anonymization, SQL auditing, CI/CD, and supports cross-regional deployment.
DSH 的本地笔记插件:随手记录、AI 自动归档、日记、待办与便笺,支持 MCP 让 AI 会话直接读写本地 Markdown。
OriginalGo 单文件一体化企业后台底座,内置可视化建模、表单、RBAC 权限、流程引擎、门户 CMS、AI 知识库;Windows 解压 exe 直接运行,MIT 协议开源。 Gitee:
Original自荐一个在线 AI 视频工具:Omni Flash 这是一个网页版工具,不用安装软件,也不用自己准备 API Key。打开网站后,用文字描述想要的画面,也可以上传一张图片或一段参考视频,生成短视频。比较适合做产品演示、营销短片和社媒草稿。 新用户有免费额度,可以先试用再决定要不要购买积分或套餐。生成后可以预览并下载。 我参与这个产品的运营。如果适合周刊的工具栏目,欢迎收录。
OriginalA native, customizable menu bar for macOS, with themes, widgets, popups and AeroSpace, yabai or OmniWM workspaces. A free, open-source alternative to SketchyBar.
给你正在跑的 coding agent 配一个本地 Web 工作台:看过程、插话纠正、管模型与工具。支持 pi / Codex / Claude Code / opencode,会话就是引擎的落盘文件,不引数据库。
Original【 大家好,我是 GlowTrans 的开发者。我做了一个随打字随翻译的小工具,支持 macOS 和 Windows,免费使用。 使用时不需要更换输入法:照常输入母语,稍作停顿后,译文就会出现在光标附近,满意后点击即可替换。做它的起因是我回英文邮件、参与海外社区讨论时,经常需要在输入框和翻译网站之间来回复制粘贴。个人日常使用非常便利。 翻译来源可以选择公共服务Mymemory、自带的 OpenAI 兼容接口;在支持的 Mac 上,也可以使用 macOS 系统翻译。这些渠道都是完全免费。 也可以选官方翻译,支持调整语气、参考上下文和设置术语表。 目前不同应用的输入框适配效果还有差异,欢迎试用和反馈。
OriginalNico 投资指南:面向大陆用户的美股、港卡与加密货币投资教程,香港银行开户、美股券商、出入金、QDII、ETF、买比特币,每周同步更新
OriginalAI agent for Steam Deck Game Mode with bring-your-own API keys (BYOK) and local voice input.
纯前端 PDF 合并/按页码拆分/体积预估核心库:页码区间解析、分块方案、激活码校验,浏览器与Node通用,本地处理数据不出设备
OriginalAppScreen 是一个免费的浏览器工具,用来制作 App Store 和 Google Play 的应用截图。 上传应用截图,套上 iPhone、iPad 或 Android 设备边框,加上标题文字和背景,然后按商店要求的尺寸导出。打开网页就能用,不用注册,导出的图片没有水印。
OriginalAI Drake Video AI Drake Video 是一个浏览器工具:上传一张照片,就能让照片中的人物成为《Burning Bridges》派对场景短视频的主角。生成的视频保留派对布景和镜头动作,并配有音乐。 AI Drake Video 的主要功能 一张照片生成视频:建议使用光线充足、单人出镜、腰部以上且双手可见的照片。 选择时长和画质:支持 5 秒、15 秒和 28 秒 480p,部分时长支持 720p。成片为 16:9 的 MP4。 直接下载分享:无需安装软件。5 秒视频通常约 3 分钟生成,较长视频通常需要 5~10 分钟。 AI Drake Video 的核心特性 Drake AI video generator 预先设定了场景和动作,使用者不必编写提示词或自行剪辑、配乐。它生成的是换主角的视频,不提供 AI 配音或歌曲生成。 这是一项付费服务,没有免费试用;生成失败会退还相应额度。项目并非 Drake 或其唱片公司的官方产品。分享视频前,建议确认照片使用已获授权,并留意发布平台对音乐的版权规则。如果想了解这种视频的制作方式,也可以从 how to make Drake AI video 入手。
Original一、我扒到的风控底牌:触发 OpenAI 封杀的 6 大高危红线 要想防封,必须先搞清楚官方在查什么。根据最近这波封号和之前的案例,绝大多数账号阵亡都是踩了这些雷: 底层网络环境污染(最高频死因):用廉价共享机场、超售节点或机房IP,短时间内频繁切换节点,同一出口IP下请求量异常,直接触发连坐。中转/反代在这轮清洗里被重点盯上。 支付卡与设备硬限额(新增高危):同一张卡或同一台设备关联太多付费订阅(Go、Plus、Pro 100/200/500都算),会直接弹出“Too many subscriptions associated with this card or device”或“Device limit reached”。低价区订阅、一卡多号、多设备叠号,这波死伤最惨。 API 违规分发与模型蒸馏:把个人订阅转成API共享或二次售卖(比如sub2api那种),或者拿输出来做蒸馏,风险极高。11月20日起,第三方模型路由/AI网关还得提前拿到官方书面审批。 支付链路异常:虚拟卡、一次性卡,或者支付地区和登录IP严重对不上。 非人类的自动化并发:请求间隔太规律、短时间超高并发,把个人订阅当生产后端用,很容易被识别成脚本滥用。 设备指纹与登录异常:同一设备高频切多个账号,或者短时间多地登录,再跟已封账号产生关联,基本就交代了。 二、我目前的破局核心:彻底抛弃廉价节点,上高纯净专线 防封最底层的逻辑其实就一条:让系统觉得你是个网络稳定、行为正常、没有共享订阅的真实个人用户。 网络环境绝对是重中之重。 我之前也用过各种低价机场,结果IP污染严重、出口老跳、并发一高就卡,后来直接被连坐。那种“万人骑”的节点,等于主动往风控枪口上撞。 正确的做法是建立低复用、高稳定性的IEPL专线。 我自己实测下来,目前用得最稳、口碑也最好的是 TAG 机场。它能成为我现在的主力,主要靠这三点: IEPL内网专线防断流:彻底不用公网中转,断流和IP跳变少很多,ChatGPT长对话和Codex长时间写代码都稳。 物理级价格壁垒和低复用:全球90+国家和地区节点,方便我固定常用出口,不乱跳。套餐门槛相对高,设备数限制也严,把大量低端脚本和羊毛党挡在外面,单节点污染概率低很多。 原生多客户端支持:Clash、Shadowrocket都能直接用,分流规则好配,DNS泄露基本能杜绝。 如果你现在还在用容易连坐的廉价共享机场,我强烈建议尽快换。可以直接去 TAG 机场官网 购买尝试,我自己日常主力是Bronze套餐,重度用的话Silver、轻度用的话Special也够。 三、行为伪装:我给自己定的“真人画像”规则 网络干净了还不够,使用习惯也得像正常人: 循序渐进:刚订阅别立刻拉满并发或通宵跑脚本,保持合理间隔,别7×24小时不间断。 固守阵地:我基本就固定1到2个常用出口,坚决不无意义跨国乱跳。 私产私用:个人订阅只自己用,绝对不共享,更不会做成API对外提供。 底子干净:尽量用真实支付方式,避开虚拟卡和一卡多号。不买来路不明的成品号。API Key一旦泄露立刻撤销重生成。 数据自保:重要聊天记录和Codex项目我会定期导出到本地,云端数据被停用后可能一时半会拿不回来。 另外提醒一下,12月开始Plus/Pro/Business的ChatGPT和Codex会共享Token额度,使用节奏要提前规划好。 四、手机验证码接码:我现在用的纯净平台 注册或登录ChatGPT经常要手机验证码,国内号很容易被识别异常,那些乱七八糟的虚拟号平台更是雷区。 我目前用得比较稳的是 DogeSMS。多国真实号段,接码速度和成功率都不错,专门适合ChatGPT这类需要手机验证的场景。比那些容易被风控标记的廉价虚拟号靠谱太多。 需要的话直接去 DogeSMS官网 购买就行。 五、我总结的挽救与避坑FAQ 已经被封了怎么办? 马上保存邮件、错误提示和订单信息,走官方申诉入口,强调是个人正常使用、没共享、出口相对稳定。千万别立刻用同一污染网络批量开新号,申诉期间也尽量别用高风险方式登录。明显转售或滥用的,申诉基本没戏。 固定美国IP就能绝对防封吗? 不能。固定干净出口只能降低IP跳变和共享出口的风险,账号共享、自动化滥用、支付异常照样能封你。 提示额度用尽就是被封了吗? 不一定,很多时候只是触发了5小时或周额度限制。真正封号会有明确的暂停或停用通知。 用sub2api中转会被封吗? 风险极高。个人订阅转API分享官方明确不支持,11月20日后第三方网关还得审批。用官方客户端或支持Sign in with ChatGPT的工具会安全很多。 老邮箱/苹果内购能提升权重吗? 目前没看到公开证据,核心还是真实资料和正常使用习惯。 出现Device limit reached / Too many subscriptions怎么办? 这是卡/设备硬限额,买Credits或重置额度没用。可以试试切到其他已允许的账号,或者取消订阅申请退款。 亮点 总结 2026年想稳着用ChatGPT和Codex,我自己的核心就三点: 1. 网络要干净稳定:远离廉价共享节点,优先上IEPL专线并固定出口。我目前主力就是 TAG 机场。 2. 使用要像真人:控制节奏,绝不共享或转API分发。 3. 资料要真实:支付信息一致,避免批量操作。手机验证码我现在用的是 DogeSMS。 网络环境永远是第一优先级。还在用廉价共享节点的,我建议尽快迁到更干净的专线,不然封号风险会一直挂着。把这三点做好,配合 TAG 机场 这种高纯净专线 + DogeSMS 接码,目前来看已经能避开绝大多数风控。
OriginalAuditable code review skill for Claude Code and Codex: four independent probes, evidence-backed findings, and a JSON-derived merge gate.
Control AI coding agents (DSH / Codex / Claude) from your phone — Feishu long-connection + self-hosted web console. 手机遥控 AI 编码任务:审批、插话、多机切换、E2EE 分享。No inbound ports.
OriginalThe Dynamic Island for your AI agents. Claude Code, Codex and more in your Mac's notch, plus a full menu bar toolkit.
Production-grade engineering skills for AI coding agents.
The personal finance app for everyone (by everyone)
A batteries-included framework for building web apps
《Hello 算法》:动画图解、一键运行的数据结构与算法教程。支持简中、繁中、English、日本語,提供 Python, Java, C++, C, C#, JS, Go, Swift, Rust, Ruby, Kotlin, TS, Dart 等代码实现
OriginalIp2region is an offline IP-to-Region localization library and IP data management framework with both IPv4 and IPv6 supports, 10-microsecond level query efficiency, xdb search client for many programming languages
Production-grade platform for building agentic IM bots - 生产级多平台智能机器人开发平台/ Agent、知识库编排、插件系统 / Bots for Discord / Slack / LINE / Telegram / WeChat(企业微信, 企微智能机器人, 公众号) / 飞书 / 钉钉 / QQ / Matrix e.g. Integrated with ChatGPT(GPT), DeepSeek, Dify, n8n, Langflow, Coze, Claude, Gemini, GLM, Ollama, SiliconFlow, Kimi, openclaw / hermes agent, deerflow
Originaloctocat: 分享 GitHub 上有趣、入门级的开源项目。Share interesting, entry-level open source projects on GitHub.
Original本文原文由知名 Hacker Eric S. Raymond 所撰寫,教你如何正確的提出技術問題並獲得你滿意的答案。
OriginalAI Agent Assistant & development framework that integrates lots of IM platforms, LLMs, plugins and AI feature, and can be your openclaw alternative. ✨
Turn your Notion workspace into a fast, customizable website. Built with Next.js + Notion API, with multi-platform deployment and no self-hosted server required.
Supercharge Apple App Store analytics
Let your team change your custom site without touching code
Rewritten as a native Mac & Windows app, with AI and MCP
Ask ChatGPT or Claude to design, then edit every layer
AI agents that turn buying signals into booked sales calls
Dual-screen recording and editing for Mac and Windows
An agentic notebook to build your best thinking
Automate complex accounting work without a single prompt
Grok Bot alternative that runs on your AI, completely free
Check Mac, server and website security
Behavioral QA agents that use your product like customers do
AI-native IDE: agents code with human review
See if a GitHub runs on your OS, and how to set it up
Talk to anyone by phone, in a language you don't speak
Your AI product team, with expert workflows built in
hey HN, i built this because I'm a satellite nerd who was bored this past weekend. type in any place and it finds every satellite that passed over it in the last 7 days (SGP4 on the CelesTrak catalog, tracking 17k+ objects and runs in your browser in about half a second), then pulls the real Sentinel-2 / Landsat / NASA images from those passes. fun fact: all ~17k objects in orbit are drawn on the GPU, solving Kepler's equation every frame. also cool is that Sentinel-2 timestamps are often 5-10 min off from when the sat actually crosses your spot, so some cool math was involved in making it accurate on the site. it only shows real images taken from those exact satellites, and it never store your location. code: https://github.com/dev-Armaan/satellites would love feedback, contributions, or whatever else you guys can offer!
One self-hosted app for your shows, movies and music.
Hi HN, I’m Shayan and a few months ago I started building a DB Client DBDelve with a focus on performance and just making something that feels good to use (in the spirit of Zed, Ghostty, and other tools that just feel pleasant to use). For my workflow I’ve always looked for performance, especially since some time ago I had a really outdated machine and any kind of bloat was really noticeable, I remember using DBeaver and it just crashing on me multiple times, I also had to use postGIS which just made it worse. I migrated all my essential dev tools to use the ones with performance in mind (Zed, Ghostty etc) but the only thing that was missing was DB Client. The main choices everyone has are DBeaver which is a memory hog and an overkill for most of the tasks imo. The other options like Datagrid, Beekeeper, TablePlus are either electron based so bloated and memory hogs or paid and closed-source. DBDelve 0.3.6 against DBeaver 25.3.1 on the same Postgres table, median of 5 runs each (2026-10-11, M1 Pro, 32 GB). Times are from the action to the rows rendered. DBDelve DBeaver Launch 1.1 s 5.2 s Idle memory 133 MB 606 MB 1,000-row query 0.21 s 1.96 s 1M rows, time 56 s 208 s 1M rows, memory [1] 4.9 GB 17 GB [1] The whole 1M-row, 100-column table pulled into memory on purpose, with the row limit off. Not a normal use case; DBeaver's heap limit had to be raised manually for the benchmark. Scrolling perf with 1 million rows loaded with 100 columns each: https://youtu.be/V2MOhJYFqIE It has most of the features most devs would need for 90% of their workflow plus some more I thought would be a nice touch: Readonly, read-write and Full Access modes Glass themes (If you’re into them, I am) Keyboard focused with every keybinding customizable Organization of connections into projects Me and some other folks have been daily driving it for more than a month now and most good feedback comes on it having a really simple UI to navigate, very stable for its age and the performance of course. I’m also sharing here because those people tell me this has potential to grow so I’m trying my hand at distributing it. Just want to give a huge shoutout here to the Zed team for GPUI, it's a breeze to work with if you just get through them not updating the crate officially on crates.io but the community has found the solutions for it, you just have to dig deep for it. It performs extremely well and just the HTML-like structure of it, makes it easy to understand for someone from a web dev background and agents are very good at writing it too. GPUI component also solves a lot of problems with having a good component library we can use. I think for anyone starting a GPUI project now, GPUI-kit makes a lot of sense, it just bundles GPUI, GPUI-component together and has some optimizations adding retained mode which is experimental but could provide a lot of benefit, so think all things considered it should be the default way. The major optimization I did was just lowering the memory overhead by using packed rows which reduced the memory usage by 40% on large datasets (8.1GB to 4.9GB on a 4GB table loaded in app), everything else came bundled in with Rust and GPUI mostly or they just made it very easy. Limitations: It's not signed and notarized, identity verification from apple's side has been a nightmare but it should be signed soon. Don’t expect feature parity with more mature clients for now, the general direction of this right now is the zed way of adding features in but after making sure the core workflows are done with a high quality. Do give it a try, you can import your connections from dbeaver and tableplus. https://github.com/ShayanAbbas1/dbdelve
Hey HN and Cats, thank you for checking this out! Github: https://github.com/raiseCatError/meowdisplay EDIT 2: I should've mentioned that MeowDisplay started as a fork of OpenDisplay: https://github.com/peetzweg/opendisplay which provided the foundational screen capture, virtual display, encoding and iPhone/iPad receivers. I've since taken it in a more phone-first direction, with work on the interaction, streaming and security layers. I've credited OpenDisplay and its developer Philip in the README under "Built on OpenDisplay". MeowDisplay is my take on the control your Mac from your phone scene. Available on iPhones, iPads and Macs, with Android rn in closed testing. The website has interactive illustrative demos if you'd like to see the app demonstrated. I think MeowDisplay does a few things differently, some of which I prefer and you might too. Such as Window view: allowing multiple tabs in split-screen type panes. Traditional Mirror/Extend views exist as alternatives. Floating windows: coming in the 2.1 release. Touch input approaches: Direct Touch(Classic or Natural), experimental Smart Touch, and Trackpad mode alongside Apple Pencil support. Doesn't sacrifice display space for a trackpad, use all of the phone's screen with your preferred mode. Stage Peek: shows a scrollable and searchable list of your Mac's windows on your phone along with an app drawer when you hold and pull on the left edge of your screen. It's free and open source so you can try it and see if it's for you without having to commit to it or anything. EDIT: I've used MeowDisplay during development whilst I was travelling between cities, left my acbook at home clamshell, powered, external display attached, and used it for small tasks remotely, it was pretty reliable even on a 5G network on a moving bus. Meow, Jai
Many thanks to my ex for inspiring my new MacOS app, Someone Was Here. It takes a photo of whoever uses your Mac while you're away, whether you left it unlocked or they know your password, and sends their photo to your iPhone if the face isn't yours. I built Someone Was Here as a menu bar app making use of Apple Vision and a local face recognition run via Core ML. It all runs locally on your Mac with no tracking or ads. Real-time remote alerts are available via one-time $4.99 lifetime purchase via the free companion iPhone/iPad app, Pushover, ntfy, Telegram or a webhook.
STE80 Humanizer: agent skill that removes AI tells and rewrites text 80% of the way to ASD-STE100 (Simplified Technical English), the method Andrej Karpathy shared.
An isometric pixel-art map of València, painted block by block. Zoom from the Micalet to the beach.
GitHub Actions workflows in any language
Audit AI repositories for offline (--network none) crashes, silent telemetry, and .env secret routing
Hi HN! I'm Claudia, first employee and founding researcher at Nuance Labs. We're a research startup based in Seattle and have just released what we've been working on for the past 18 months. Emma, a conversational audio-visual AI that you can have a video call with in real time. It might be a little rough around the edges but we think it's better than anything else out there (we're biased, of course). You don't need to make an account and it works from the browser with mic and optionally camera. Emma is built on our full-duplex audio-visual system, Nuplex. We focused on naturalness over existing benchmarks as we found they didn't correlate well with what people actually like or how we speak face to face. We think being uniquely expressive is more important than an averaged-out face, for example. Our blog goes into more detail: https://research.nuancelabs.ai/ A beta developer API is available at https://platform.nuancelabs.ai/ . We're taking safety very seriously so she may prematurely end the call, in particular for the API, but we're working on tuning that. Personality, across all modalities, was the most important part for me. I've spent the longest speaking to her and I think she can be a lot of fun! I'd love for people to try it out and I look forward to hearing what you all think. Thank you! (Apologies for the double post, someone external posted our before we were ready, oops: https://news.ycombinator.com/item?id=50429477 )
Free bracket maker — create a custom tournament bracket in minutes. Six free formats, plus registration, scheduling and live pages to run the event.
Headless 3D and VFX engine for agents. Research preview.
Commune is a newsletter platform with a shared feed. Newsletters listed there come from many providers: Beehiiv, Kit, Ghost, Buttondown, Substack, MailCheap, Mailchimp, and even blogs via RSS. For every article, a discussion is created automatically. The homepage is the product. You only need an account to interact with authors and other readers.
Person detection on a webcam stream without ever exposing the images. The client encrypts each frame using FHE and sends it to a Belfort eML server, for inference directly on the encrypted data.
A job board without the recruiters. Find jobs by keywords and show them in a timeline. Hopefully this helps someone looking for work!
跑步骑行几年,手里的轨迹文件越来越多,经常遇到这种小麻烦:Garmin 导出的 FIT 想传到别的平台不认、两个活动想合并、FIT 文件损坏打不开……在线工具大多要把文件上传到陌生服务器,轨迹里可是完整的出行记录,想想还是不踏实。 所以做了这个:**GPX Converter** —— 所有解析和转换都在浏览器本地用 JS 完成,没有上传这一步(断网也能用),免费、无需注册、无次数限制。 **现在能做的:** - 格式互转:FIT ↔ GPX ↔ TCX ,以及 KML / KMZ / GeoJSON 之间的转换矩阵,心率、踏频、功率数据都会保留 - Google 时间线( Timeline ) JSON 导出转 GPX - 合并 / 拆分 FIT 和 GPX ;修复损坏的 FIT 文件(文件头、CRC 、截断这类常见问题) - 在线预览轨迹:地图、距离、爬升、配速 - 轨迹海报生成、GPS 热力图 站点是纯静态的,转换逻辑就在浏览器里跑,所以速度基本取决于你自己的电脑。最近刚补了全站中文版和十几篇使用教程( Garmin / Strava 之间迁移数据、修 FIT 之类的)。 独立开发维护,欢迎试用,更欢迎挑毛病:哪个格式转出来有问题、缺哪个工具,直接回帖告诉我就行。
Original分享个今天刚上线的站: 项目目前的定位是给学中文的老外用的打字练习工具,目前专注于五笔输入法。 五笔板块的核心功能就是在打字的时候,每个汉字会实时拆解成笔画和字根, 能看到“这一笔为什么在这个键上”。 比如首页演示:左边"欢迎"用五笔(欢迎两个字各按什么键、字根是怎么拆), 右边"你好"用拼音( ni hao 的按键和注音对照)。 项目背景是偶然发现了一个开源笔画数据( ARPHIC 许可)+ 五笔码表,结合 AI ,能做到常用字的笔画级拆分可视化。 目前手工还在补充生僻字的开源笔画数据,后续考虑一起开源出来。 假设码表能做到全覆盖了,会做个像“金山打字”的现代 Web 版本的五笔练习站。 技术栈用的 Astro + Vue islands ,纯静态,CF Pages 。 刚提交了 GSC ,站是今天上线的,欢迎围观,也求个友链 🙂
Original祝大家`天天赚钱` :) 输入网址,自动帮你生成网页优化方案, 导出优化方案交给当前主流 AI 开发工具(trae 、cursor),然后添加 /goal 就可以优化项目 目前支持 jev-seo 等方案,也可以自己创建提示词来优化 在个人项目中,启用全站自动优化,可以生成整站的优化方案,导出优化文件交给智能体 1. 打开首页可以看到一下内容,输入你想优化的网址 2. 然后他会生成优化方案(比较慢,可以刷新一下) 3. 注册后,点击进入默认项目,在项目详情中,可以开启整站优化(你可以后面再过来,他会在后台帮你自动生成优化方案) 4. 点击展开这里面的 thread,然后可以看看抓取结果 5. 可以下载整站优化方案 6. 也可以下载单页优化方案
OriginalMinimal Kavita client for Android E Ink readers: download comics and novels, read offline. 面向墨水屏的极简 Kavita 安卓客户端
Original想起什么说什么,基本是要做微信小游戏就需要了解的. iOS 上微信小游戏能用的内存有上限:整个微信进程内存和显存共用大约 1G 到 1.4G(具体看手机内存),超过就会被系统杀掉;另外 5 秒内收到两次内存警告,微信也会直接销毁小游戏。 别以为你游戏资源包只有几百 MB 就没事。PNG 、JPG 只是文件小,进显存要解开成原始像素,每个像素 4 字节,小游戏占用只看分辨率:一张 2048×2048 的图不管文件多小,都占 16MB ,开了 mipmap 还要再多约 1/3 。要是你的游戏资源几百 M,基本玩着玩着就必崩溃. 解决办法是用 ASTC 压缩纹理:显卡能直接读压缩后的数据,4×4 块的显存是原来的 1/4 ,6×6 约 1/9 ,代价是画质略有损失(我自己测试看不出分别);再配合用完及时卸载资源,就能把峰值压下来。拿我游戏来说,战斗峰值从约 1G 降到了 200MB. ASTC 压缩还有个问题,可能会把某个类型的图片变成洋红色,真机才会出现,所以一定要测试好. 用户手机发热可以在小游戏网页后台给你的游戏接入高性能模式,能大幅度降低发热问题,其他也有能优化的,腾讯这点好在每个文档里面都有一键复制,直接给你的 AI 看,完美接入. 网页后台显示 100 用户就可以开启流量主,手机看要 1000,这里以网页后台为主. 任何游戏,一定要弄新手教程,用户进来几秒不知道干啥,绝对就退了,想象你自己玩小游戏,有那么耐心吗? 游戏版本审核的时候,只要你的资源需要额外下载,审核界面一定要选择联网,而不是单机,这里不是看你游戏的类型,只要有云上资源,就要联网,不然审核员用不能联网的环境审核失败,会直接打回. 加载云上资源的时候,游戏一定要显示具体的大小或者文件进度,我买的腾讯 cdn,按理来说网络肯定可以,但是几次遇到下不动的情况,如果你的游戏没有详细说明,用户可能会以为卡死,直接就退了. 第一次版本审核成功,后台会建议你用他们的代码加固插件,用的时候代码混淆选中等就行,高选项会直接让主包大于 4M. 接入好友排行功能,要在后台开启这个功能,不然游戏不保报错,但是也不显示内容,会让用户疑惑. 每个游戏可以开启新游种子,基本会给量让你开启流量主,一定要把版本弄好再开启,不像我,新手引导都没有就开启了,浪费了... 就说这么多吧,哎,不做游戏看别的游戏都是垃圾,一做才发现真的难啊,总是有一些问题出现,更难受的是辛辛苦苦做完,没有流量 现在小游戏不花钱推,基本都没入尘烟了 我的游戏是割草类游戏,就是各种技能搭配,纯爽游,感兴趣的朋友微信搜:站桩修仙
Original最近做了个小站,Rapvio 。起因很具体:想上传照片,制作自己的 rap video ,把自己和朋友放进一段说唱视频里。 做出来之后发现,很多人其实连歌词都还没有; 也有人词写好了,只想要一首歌,不想要视频。 所以现在拆成了几条可以随时停下的路径。 不会编曲、不会录音、不会剪片子,也能从一句话走到一首歌,或者一条 10~15 秒的竖屏片段。 视频模型试了 MiniMax 、Seedance 、Wan 、Kling 几档,都能出,目前 minimax h3 性价比是最高的,画面风格上第一版做了八种 look (生日、battle 、宠物、圣诞之类),单人双人都能生成。 写词那步是独立的一环,词确认之前不会开始做歌;反过来,词已经写好了也可以直接出 MP3 ,跳过写词。这个也是跟朋友聊到的问题。 另外有个免费不用登录的艺名生成器,点一下出一批,纯玩。 技术上最麻烦的不是接模型,是让 15 秒左右的片段在节奏上站得住——太短就不像说唱,太长成本和废片率都上去。这块还在磨。 不过歌词和 mp3 的生成基本没问题,内置了写歌 skill ,欢迎体验。 明确还没做的:不是完整 MV (单镜头,不做分镜剪辑),不做声音克隆。界面目前是英文,主要因为早期用户大多在海外,中文版在排。 想听一个反馈就够了:你更想拿它跟朋友做条短视频,还是先写词、先出一首 MP3 ?这两种用法我拿不准哪个更刚需。 |欢迎直接拍砖
Original市面上已经有很多个稍后阅读工具,但用了一段时间,觉得还是得要自己做一个,于是参考某个比较合适我的网页阅读工具我也自己做了一个[SH-稍后阅读]。 1.用谷歌 AI 做的; 2.数据离线本地存储,网址数据可以本地导出导入,暂未做加载数据的导入导出; 3.可以设置网址在线与其他设备实时同步,不用注册和登录,仅用同步码; 4.优化多图加载、代理加载等,V2EX 文章在未上飞机前能快照浏览。 5.满足了自己订阅 RSS 看到好文章时的快速保存和同步浏览需求。 [SH-稍后阅读]工具:
OriginalMinimal, low-memory AirPlay + DLNA audio receiver for the Xiaomi LX04 touchscreen speaker. Two-page player, synced lyrics, XiaoAi voice controls. GPL-3.0. | 面向小爱触屏音箱的极简低内存 AirPlay / DLNA 音频接收器:双页播放器、同步歌词、小爱语音控制,GPL-3.0。
OriginalTry another keyword or category.
18 repositories were already featured in the last 3 reports, so they are not repeated in today’s discoveries.
An open source design system that's fully customizable and agent ready
Hunt down social media accounts by username across social networks
Public repository for Agent Skills
DEV TRENDS / COMMUNITY
Discuss this edition
Sign in with GitHub to share a thought, question, or useful follow-up.