DEV TRENDS / CATEGORY
2026年10月26日 – 2026年11月1日 内首次发现的 95 个新项目，已跨 8 个来源去重。
2026年8月10日 – 2026年11月1日 内首次发现 996 个项目
Hotseat（ 菜单栏工具，在同一台 Mac 上切换多个 Claude 账号。Claude Code 和 Claude Desktop 一键换号，会话、设置和 skills 共用同一份，换号后能接着原来的对话继续。不读取任何登录凭证。Go + Swift，MIT 协议。 菜单栏能看到每个账号 5 小时和每周的用量、什么时候重置。 CLI 会话因为额度停下时，会弹窗问要不要换到你的另一个账号继续。不点就不换。 不自动轮换账号，没有号池，只适合你自己名下的账号。 安装：brew install zahT3/tap/hotseat，或去 Releases 下载。macOS 14 以上。
MeterLuma puts remaining AI allowances and reset times in your Mac menu bar or Windows tray. Supports Claude Code, Codex, Cursor and 42 more integrations. Local usage history stays on your device. Free for personal, noncommercial use. macOS 13+ and Windows 10/11; unsigned installers.
原文Managed Agents API you can use with any model. Send a task with an API call, and Click runs the agent and streams back results. Click gives you the entire agent stack over an API, with no infrastructure to manage. Sessions are created with everything your agent needs - a sandbox, browser, and web search. Agents run on a harness with the latest innovations, like deferred tool discovery to support 1000s of tools, and Code Mode to use tools with scripts.
原文Self-hosted remote control for Claude Code and Codex, built for teams. Everyone joins the same session from any device: one context, one environment, one task. Approvals arrive as push notifications, so whoever's online keeps the agent working and your plan's tokens don't sit idle. Chats, terminals, files, git and dev-server previews for every machine you own. Open source.
原文Break free from Claude Code context limit errors and maintain peak AI dev speed. Agentic tools drain tokens fast via auto-searches. This toolkit provides a Relay Architecture to eliminate context exhaustion. What’s Inside: State Framework: CLAUDE.md &claudignore templates Handoff Prompts: Auto-compress history in seconds CLI Automation: One-line terminal aliases for instant resets Refactoring Guides: Blueprints for multi-session builds
原文Version 4 is a big one. Usage for Claude now tracks ChatGPT accounts too, with your Codex session and weekly limits next to Claude's. See every account at once in three styles (Minimal, Detailed, Chart), put several in the Mac menu bar, and keep them on iPhone, iPad, Apple Watch and widgets. Alerts tell you before you run out, and the history shows where your week went. You sign in on the official Claude and ChatGPT pages, and your data stays in your own iCloud. Free, with Pro extras.
原文Runway App Studio lets you describe a creative app in plain language and get a working tool that generates images, video and audio. Turn an Agent session or Workflow into an app in one click, edit it in plain language with rollback, and share it privately, with your team or by public link. Per Runway, it is included in every paid plan and uses your existing credits. Six built-in apps include Character Swap.
原文在 Cloudflare Workers 上自建永久别名邮箱，一键部署、完全免费。每个网站一个独立地址隐藏真实邮箱；在自己的收件箱里直接以别名回信；自动把验证码提取到主题；去除追踪像素；只存元数据不存正文；带 Web 面板、REST API 和 MCP 接口。只用 Workers + D1 + Email Routing，约 1000 行 JS，MIT。 亮点 一键部署在 Cloudflare Workers 免费额度上（Workers + D1 + Email Routing），零成本长期使用；为每个网站生成独立别名地址隐藏真实邮箱，并在自己的收件箱中直接以别名回信；自动把验证码提取到邮件主题，自动去除追踪像素，只存元数据不存正文；同时提供 Web 面板、REST API 和 MCP 接口，方便 AI Agent 集成，MIT 开源。
Codex Skills is a free directory of 872 MIT-licensed Agent Skills for OpenAI Codex. Each skill is a complete SKILL.md playbook that Codex loads the moment a task needs it: PDFs, React, Postgres, security reviews, AGENTS.md and more across 28 categories. Install any skill with one command, grab curated collections like the Starter Pack, and read every line before it runs. No account, no telemetry. Works in the Codex CLI, IDE and app.
原文telemetry.dev shows each run of your AI agent as one trace. Each model call and tool call is a span with its duration, tokens, cost, and error. Prompts and evals use the same traces: change a prompt without a deploy, then run it on a dataset and compare the scores with production. Alerts tell you about errors, cost spikes, and agent loops. Connect with our TypeScript or Python SDK, or with the OpenTelemetry exporter that you already use.
原文Claude Code and Codex create VMs, databases and deploys mid-task, then forget them. agent-leftovers reads the logs they already keep on your machine and lists every leftover with the exact command to turn it off. Read-only, local, free.
原文Hi HN, CodeConnect is an iPhone app where you get to control your coding agents across multiple computers from anywhere. Install the daemon (on mac or a linux machine) and run "codeconnect claude" or "codeconnect codex" and its streams the session on the app. Runs over Tailscale and fully open-sourced. Would love some feedback :)
原文Give AI agents a separate display for real Mac apps while you keep working. Connect through MCP or the CLI, watch sessions in the Viewer, and take over when an agent needs help. Supports native apps and Chromium. Free and MIT licensed; Apple Silicon, macOS 14+. Attention isolation, not a security sandbox: apps run as your macOS user. Runtime support depends on your macOS build.
原文拾趣是一款将中国古典故事变成连环画的开源工具。选择书籍、主题和画风，AI 自动改编故事、编排分镜并生成插图。支持电脑和手机浏览器阅读，可导出 PDF 或长图，故事、图片和阅读进度保存在自己的电脑上。 亮点 从选题、故事改编到分镜绘图，完成整篇连环画创作。 支持多种画风，让古典故事呈现不同的视觉风格。 可查看原文、分镜和制作提示词，了解 AI 的改编过程。 支持已登录的 Codex CLI，利用订阅额度创作；也可接入 OpenAI API 或 Gemini。 本地保存作品，手机浏览器阅读，支持 PDF 和长图导出，方便收藏与分享。
重新搬回的“椅子” 过去，两个程序员想一起解决同一个问题，可以把椅子挪到同一块屏幕前。 一个人负责操作键盘，另一个人负责审视思路；遇到分歧，就停下来讨论。屏幕上的代码是两个人共同产出的成果，问题的来龙去脉，也在这一问一答中逐渐对齐。 这种协作的方法，就叫做“结对编程”。 如今，AI 俨然已经成了坐在我们屏幕另一侧的新搭档。 我们向它描述需求，让它阅读代码、分析问题，并尝试修改。在一轮轮对话之后，那条会话里逐渐积累起了整个任务的背景，包括：为什么要做这个功能，哪些方案已经被否定，还有什么约束必须遵守等等。 可当后续有另外一位同事想参与进来，或接手你的工作时，事情就变得开始低效了。 我们需要把对话截图发过去，把代码复制出来，再把同事的意见转述回给 AI。对方如果想追问一句，又要等我们把问题送进去，再把回答搬回来。 AI 明明已经能够参与越来越多的执行工作了，但人与人之间的协作，仍可能被这些来回的转述给拖住。 个人的生产力被放大了，团队掌握的信息，却未必能同样顺畅地汇集到一起。 是时候，我们该重新搬回那张“椅子”了。 Codex Co² 想解决什么？ 基于同一条会话的协作，正是我想用 Codex Co² 来解决的问题。 Codex Co² 的全称是 Codex Conversation Collaboration（Codex 会话协作），名字有点长，但是想做的事情很简单：让多个人参与同一条 Codex 会话，一起推进工作。 分享者可以继续在官方 Codex Desktop 中工作，只需在 Co² 里选中一条已有会话，生成邀请，发给协作者。协作者则通过浏览器打开邀请，申请加入；获准后，就能阅读共享上下文，向同一个 Agent 补充要求，并跟进它的回复。 协作者无需安装 Co²，也无需单独登录 Codex。发出的每一条消息，都会沿用分享者当前会话的工具、权限和额度。 邀请协作者 不妨看一个具体的例子。 假设有用户反馈：“任务已经处理完成，页面上的状态却一直没有变化，刷新之后才恢复。” 运营拿到 FeedbackID 后，可以先在 Codex 中分析相关的用户日志，得到初步结论：操作请求已经成功，问题可能出在页面状态的更新环节。 随后，运营就可以通过 Co² ，将这条会话分享给研发。 研发加入后，可以沿着已有的分析继续追问：服务端有没有发出状态变更通知？客户端是否收到？异常发生前，是否经历过断线和重连？ 结合日志与相关代码，问题逐渐收敛：客户端在网络重连后没有恢复实时订阅，后续状态变更因此未能到达页面；而手动刷新会重新拉取数据，所以显示又恢复了正常。 到这里，一条模糊的用户反馈，才有了可以解释、可以验证的技术根因。 邀请另一协作者 但找到根因，并不意味着就走到了解决问题的最后一步。 研发可以继续让 Agent 梳理修复方案：补齐重连后的订阅恢复，并在恢复连接时校准一次状态；同时整理受影响的场景、改动范围、验证成本，以及临时处理方式，再邀请产品加入这条会话。 产品看到的，便包括了用户最初遇到的困难、运营从日志中获得的线索、研发定位根因的过程，以及不同处理方式的成本与风险。 基于这些信息，产品可以决定是否需要立即修复、如何安排版本，以及在修复上线前怎样回应用户。 运营带来现场，研发追到根因，产品决定如何推进。 不同角色的判断，在同一条会话里逐步接续。前一个人留下的分析，成为后一个人继续追问的起点，最终把一条用户反馈推进为一个有依据的处理决定。 收集分散的信息 Co² 带来的这个看起来很小的变化，恰恰对应着软件开发中一个长期存在的事实：理解一个问题所需要的信息，往往分散在不同的人手里。 产品同事知道某条看似奇怪的规则，背后对应着怎样的业务现实；测试同事知道一个偶发问题，究竟要经过哪些步骤才能复现；熟悉系统的工程师则知道，一处看似无害的改动，可能牵动哪些历史约束。 这些判断，很难仅凭一句“帮我把功能做出来”就交代清楚。 AI 能够处理的信息再多，也需要有人把这些最相关的事实带到它面前。因此，当我们讨论如何更好地使用 AI 时，更值得考虑的是：怎样让那些真正了解问题的人，更容易参与当前的工作。 Co² 尝试提供的，就是这样一个入口。 我希望它能够用在结对开发中，让另一位工程师直接参与方案讨论；用在远程排障中，让了解现场的人补充复现条件；也用在需求澄清中，让业务上的判断及时进入实现过程。 还有更多可能性 当然，Co² 目前仍是一个早期的独立实验项目。 项目代码已经在 Github 上开源，提供了 macOS 和 Windows 安装包，可以通过 Homebrew 安装后打开，也可以直接在浏览器上运行。 使用时还需要理解以下几点限制： 共享会话沿用原来的执行环境和额度，协作者的每一条消息，实际都是以分享者的身份在发送的，使用时需要注意权限控制和额度把控。 公网分享经过中继，使用普通 TLS，没有端到端加密，记得只邀请可信任的协作者，或者开启“仅内网可访问”开关； 退出 Co² 时，分享出去的邀请链接会变成不可访问，因为 Co² 设计上就是服务于当下的协作，而不是作为一个长期的入口。 如果你也正在和同事一起使用 Codex，欢迎试用，并把遇到的体验问题通过 Issue 反馈给我，一起参与 Co² 的共创。 那些曾经需要截图、复制粘贴、反复转述的环节，就是 Co² 这个工具能帮你们改进的地方。 结语 希望，我们在用 AI 来提升我们工作效率的同时，也能继续给身边的同事留出位置。 让了解需求的人、熟悉代码的人、发现问题的人，都能把各自掌握的那一部分带进来，完成一个共同的协作，这才是团队存在的意义。 曾经屏幕前的那把椅子，值得再搬近一些。
自荐（利益相关：作者本人） TeapotRouter 是一个面向 AI 编程助手的 API 网关：一个 Key 同时兼容 Anthropic 和 OpenAI 双协议，Claude Code、Codex CLI 改两个环境变量就能接入，按 token 计费。 为什么做这个 中转站市场口碑差，不是"中转"本身的问题，而是三件事：价格不透明、余额会过期、跑路没对账。所以这个站把三个承诺做成硬规则： 价格表全站公开：每个模型的基准价与实付口径写在明处，另有机器可读的 JSON 价格接口，方便比价工具直接抓； 余额永不过期； 计费日志可导出对账：每笔请求的成本、延迟、渠道都可查——Claude Code 烧 token 到底烧在哪，看得见。 其余：支付宝可充值（最低 $1）、多上游自动故障转移、支持退款（按剩余余额退，工单办理）。 彩蛋 名字和 logo 来自 HTTP 状态码 418 "I'm a teapot"——logo 就是「418 压扁成一把茶壶」，官网还专门做了一页 418 状态码小史（从世界上第一个咖啡壶摄像头讲到 RFC 2324）。 接入（Claude Code 示例） bash export ANTHROPICBASEURL= export ANTHROPICAUTHTOKEN=sk-your-key 上线两周的新站，欢迎批评。
Hello my beloved tech forum! Earlier this year I moved to a new apartment, and scattered a box of poetry magnets all over my fridge. Whenever people came to visit, I noticed how they always loved making poems of their own, not to mention enjoying the poems others had already created. It became such a fun and entertaining tradition, (especially due to the range of creative messages that my friends chose) that I felt compelled to make a full stack project out of it. While LLM agents have made the technical aspects of building a project like this easier, it took me a really long time to nail the algorithm for choosing daily word sets. I started with a huge collection (on the order of 100,000) of words, grouped by part of speech and sorted by the frequency of use in English. I tried randomly choosing a few words in the top 10, a few in the top 100, and so on, for each part of speech. That straight up did not work. Trying to make poetry out of those wordsets felt terrible. It was very discouraging. I had to go back to the drawing board many times. My current solution came about when I found some databases of words, one sorted by their sensory value (from most to least sensory-evoking) and one where words are sorted on a spectrum of concreteness to abstractness. Another key to picking fun and interesting word sets was ensuring there are a solid set of words that are included in every set no matter what. This "glue" set was handpicked by words that I found myself searching for over and over again when trying to write poems, but it is deliberately small to help people be creative in how they express ideas. I used Deepseek v4.1 Flash and the Pi agent. My AGENTS.md emphasizes my status as a junior developer, specifically prompting the agent to never make architectural decisions on my behalf, to quiz me on the architecture/code involved before implementing changes I request, and to never write copy so I can make sure the site feels warm and genuine in my own words. I wanted to avoid letting the code get out of control for when I want to add features in the future. One of the things I love about building these kinds of projects that LLMs enable is being able to add tiny features that most people will never see. For example, I added special effects that occur when a haiku is written. I chose TanStack Start and React this time instead of SvelteKit. The experience was about the same. And kudos on poems are deliberately spammable. I like seeing number go up. Friends have expressed enjoyment from using it so far, but I'm excited to hear feedback on y'all's experience and see if there's anything huge I'm missing. Thank you in advance!
原文Your agent remembers every session. You don't. Coding Wrapped turns your local Claude Code & Codex history into a pixel-art Wrapped: your rhythm, the words you repeat ("continue", anyone?), what the agent quietly took over, and one small tip for each habit. Open source, one command, 100% local.
原文AC Bridge turns the smart lights and plugs you already own into ambient status for Claude Code and Codex. See at a glance whether your agent is working, blocked or done, and let Claude Code or Codex drive the devices you have connected, as a step in a task. Works with Hue, Govee, WLED, LIFX, Nanoleaf and more, plus Home Assistant.
原文Free tracker of real pricing for Zapier, Make, n8n, Lindy and 7 more automation and AI-agent tools, with a cost calculator and a log of every price change.
原文同时开着 Qoder、Claude Code、Codex、ZCode、WorkBuddy、OpenCode 的时候，很难判断谁在跑、谁在等你回答。Agent Ping 是一个 macOS 菜单栏 / Linux 托盘面板：只读这些工具已经写在本机的日志，把每条会话标成正在跑、排队、等你回答、已完成或报错停下，点一行就回到那条会话；不联网、不要账号。 亮点 同屏六家 Agent，按状态分组而不是按工具分组——打开面板先看见「谁在等你」；等你回答的那条会话会带出问题原文和可选项（装 hook 后可精确到提问）。整个二进制不发起网络请求，只读日志，不碰账号密码；macOS 用 Homebrew 一条命令源码编译安装，Linux 有源码构建。
At work, we're leaning heavily into agentic development. I've always used tmux to manage my work and now I'm using it to organize my 30+ Claude sessions. I wanted a simple read-only dashboard that would identify which of my sessions needed attention and also fit my vim-based navigation workflows. There are times when all I remember is a name or phrase from a conversation. The search feature was built to address this by searching the contents of every transcript, not just session titles. Just curious if other devs are experiencing similar issues and whether this approach is helpful at all. Thanks!
原文Audit your own exposure surface: every agent panel, staging dashboard, and forgotten deploy. If it answers on the public internet without auth, that is the finding. Your domains only — never scan anyone else's infrastructure. PanelCheck is an own-domain panel and exposure audit kit: scope checklist, lockdown runbook, secrets-rotation template, auth-by-default guide, and one-page exposure report. $39 one-time.
原文Chopstick notices what you repeat on your Mac and builds notch widgets for it, using the coding agent you already have. Your activity log stays local.
原文三组公开数据：有人人工标了 150 个 Claude Code 会话，357 句"测试通过"里 80 句背后根本没跑；圣母大学等分析 20,574 个真实会话（Claude Code、Codex、Cursor、Gemini CLI 都在），16,118 个失误里每 100 个就有 22 个是"没做完却说做完了"；Transluce 看了 8,600 个会话，每 55 个就有 1 个严重到"没验证的代码直接进生产"。 它不是每次骗你，是有一定概率不去检查却告诉你检查过了。轮到你头上是什么样：Codex 仓库有人提 issue，它说验证过了就部署，线上卡了五分钟，数据库连接池被打满；Claude Code 仓库有人提 issue，它说修复完成，缺陷还在线上，8,726 条记录漏掉，另一条是测试明明在报错、它回复 ALL PASSED；Cursor 论坛有人说它"浏览器里试过了"，追问才承认没开过浏览器。便宜的结局是 CI 红了回头排查，贵的是线上炸、数据漏，最糟的是后面几轮全建在这句话上，错一层叠一层。 honestly 是一个 hook：助手每写完一条回复，它读这一轮的会话记录，找出"测试通过 / 修好了 / 已推送 / 已发送 / 已部署 / 已保存 / 查过了"等 14 类结论，看最后一次改代码之后有没有对应的命令，并读命令的输出而不是退出码（pytest | tail 的退出码是 tail 的，测试全挂也是 0）。没有依据就拦下回复，要求它先跑再说；下一条回复会写明为什么被拦、补跑的结果是什么。
Agents write code. Help your team plan, understand, ship, and run it. 10 MIT-licensed tools for agent workflows, codebase learning, local CI, observability, tenant isolation, RPC access, log redaction, messaging, and GitHub attachments.
原文I decided to test this claim with my app, NoMoreHappyPath (NMHP), against Claude on a real project. Here is the test: I built Acremuse, a commercial real estate analysis platform, with Claude Desktop and Codex as a reviewer. It is deployed and used today by a couple of investors. Then I wondered what I'd missed that might bite me later. ~ 374,000 Lines of program code - 1,070 Source files (Python and TypeScript) - ~ 11,400 Functions and methods - 3,067 Automated tests So I gave the same 374k-line code snapshot and the same brief to one Claude session that can use Codex as a reviewer and to my own tool, NMHP. My directive was the same for both: Analyze the code, find any errors/bugs, and list them by severity. Findings must be reviewed before writing the final report. The Results (Verified by an impartial, blind Codex judge): NMHP: 33 real issues found. 7 false claims. Claude: 10 real issues found. 2 false claims. Only 6 issues overlapped. Time & Cost: From the owner's go-ahead to the reviewer's pass, NoMoreHappyPath took about 74 minutes, including time lost while it found and worked around a defect in the product itself. The Python version NMHP was working on was 3.9, the OS version, and it didn't have pytest installed, which was needed to verify some claims. After 15 minutes of blockage, the CTO stepped in and told the agent to create a venv with Python 3.13, install the corresponding pytest, and move the project along. NMHP CTO reacts when 15 minutes pass without movement from agents. To be fair to Claude, Claude did create a venv from the get-go. Time & Cost: NMHP took 74 minutes and cost $24 in API spend. Claude finished in a few minutes for a fraction of that cost. And yes, NMHP takes over an hour and costs twenty-four bucks. But for a 374k-line codebase, finding 23 more real issues overall that vanilla Claude missed makes $24 a massive win compared to human audit rates. I know this isn't a benchmark; it's one app and one run each. I wrote a detailed white paper on it. But it confirmed my idea: an AI agent without a real process is not yet capable of building a commercially ready application, even with a frontier model and reviews that are initiated by the same agent.
原文我是 Runward 的维护者，提交项目自荐。这是一个本地运行的编程代理会话看板：读取用户明确选择的日志，将工具调用、错误和子代理活动放在共享时间轴上，方便排查多个会话重叠执行时发生了什么。项目采用 MIT 协议，不需要模型 API Key，适合同时使用多种编程工具的开发者。 亮点 瀑布图、事件时间线和控制台共用一份会话数据，可以定位某段工具调用并查看对应事件。 提供 Codex、Claude Code、DSH、OpenCode、OpenHands 和 Gemini CLI 的日志读取器，支持本地导入和增量读取。 Node.js 22+ 即可运行，无运行时依赖、无需前端构建，服务默认只监听回环地址，不向外部服务发送请求。 日志没有时间或 token 信息时保留未知状态；只做查看，不启动代理、不审批工具、不执行交接。 验证范围：Codex 已用真实本地会话检查；Claude Code 目前使用合成测试，其余四种读取器依据官方格式进行合成测试，尚未完成真实模型会话验证。导入日志不会自动脱敏，请勿将私人原始日志贴到公开 Issue。 欢迎试用演示并反馈导出格式或时间线显示问题；带最小合成复现的 Issue、解析测试和采集文档 PR 都有帮助。
Meet your AI weight loss agent. Scan meals by photo, estimate calories and macros, and get personalized guidance based on your goals and daily log.
原文Canva Brush is a vision-language model for design from Canva Research. It sees your design, understands how it’s put together, and makes the changes you ask for, from one line of copy to every slide in a deck. Everything stays fully editable. Canva says it’s up to 69% faster and 46% lower cost per edit task than its previous design agent. It’s powering Canva AI design edits for Canva Pro subscribers today.
原文Industry voice bundles plus a local agent for your business. Your own AI drafts answers, you approve the wording, record once and keep the audio on your computer. Use them for calls, texts, bookings, reminders and order updates, and connect Slack, Teams, Jira, Confluence, calendars and your app. Test in practice mode first. For developers: speech, transcription and custom-voice APIs. Audio from $14; with the local agent from $214, pay once; calls, messaging and optional AI may cost extra.
原文平时让 Claude Code 和 Codex 干活，它们会产出一大堆要我看的东西：方案、调研报告、对比表、截图、网页原型，偶尔还有 Word 。以前这些都只是终端里的一串文件路径，在电脑上看一份三页的方案都费劲，在手机上更是看不了。 mydash 是一个跑在自己电脑上的小服务（ Node ，不用框架）。agent 运行 mydash publish ，你就拿到一个链接，在电脑或手机上看排好版的内容：Markdown （带 Mermaid 图）、PDF 逐页显示、网页原型在单独的端口里运行、Excel 和 CSV 表格、Word 和 PPT 的排版预览。按项目（会话启动时所在的文件夹）归类，有版本对比和未读标记。我用 Tailscale 在手机上看。 后来又加了几样： - 在网页上选中一段写批注，agent 用 mydash feedback 读到、改完标记处理； - 要我连着拍好几个板时，agent 发一张审批单，不在回复里一口气问一串； - Claude Code 的选择题、计划确认、权限确认，终端和 mydash 上同时出现，哪边先答算哪边，30 秒没人答就推送到手机（ Bark 、ntfy 、Telegram 、企业微信、飞书、钉钉、邮件都行）。 安装交给 agent：在 Claude Code 或 Codex 里粘贴一句「帮我安装 mydash ，按 的说明来」，它会问你几个问题然后装好。界面中英文都有，macOS 、Linux 、Windows 都能用，MIT 开源。 大部分代码是 Claude Code 写的，我负责提需求、定方案、验收，最后做了一轮代码审查。欢迎试用和提意见。
I didn’t want to copy and paste messages between coding-agent sessions or across vendors, so I built an MCP server that lets one model in one session talk to another model in a different session. They can even have a group chat. It supports multiple OpenCode sessions and lets GPT-6.1 in OpenCode talk to Claude in Claude Code.
原文本人二十年后端开发和架构经验，对 coding agent 的要求就是稳、准、慎三个字。我现在的 AI 工具搭建很简单，就是先装一个 pi ，里面装一个记忆插件（尝试过一些，最后选择用 Magic-Context ），然后配置一个全局的系统提示词。就这些。然后我用的模型就是各类 flash 模型，最便宜的。我一直专注于把最便宜的模型用到极致。 要想把最便宜的模型用好，系统提示词很重要，它决定了 agent 的思考风格和基调。我们是小公司，没有测试环境，没有测试流程，改完了就上线。我的 pi 是可以直接访问生产环境的，但凡出现报错日志，我只要粘贴日志内容，然后说“分析一下原因”，agent 就会翻日志、翻代码、翻数据库，翻配置文件，最后把横跨多个数据库和多个模块调用的来龙去脉讲清楚，最后还给出建议。 总之，就我的体验，这套提示词配合项目具体的 skill ，我用起来还是比较满意的。我现在唯一不让 agent 做的就是数据库修改操作。 提示词大小 7K 左右，具体内容 希望对大家有用。
Uncanny Office is a Claude Code desktop app for Mac, Linux and Windows: remote terminals for every session, a retro office to monitor your AI agents, and jobs on a schedule. Free for 7 days, then one early-access purchase that covers the beta and 1.0.
原文Your agent knows Kubernetes. It doesn't know your cluster. Radar (open-source) gives it context over MCP: ranked issues, a change timeline, logs and owners in one view. New: click Investigate on a failing workload. Claude Code, Codex or Cursor finds the root cause, read-only, with evidence for each claim. In our benchmark of 50 live faults across six tools, Radar found the most root causes. Radar Cloud investigates every alert and posts findings to Slack.
原文Get a free text when OpenAI announces a Codex usage limit reset, and another when it's available. Use what's left before your limits reset. Text START to sign up: no app to install. SMS in the US + Canada, Telegram worldwide.
原文AgentGPA grades how you work with Claude Code, Codex and Cursor. One free command reads the sessions on your machine and sends a text-free event log, never your code or transcripts. You get seven grades, a GPA, a 15-second video and a rank.
原文Spyhop is a free, open-source Mac menu bar app for people running many Claude Code, Codex or OMP sessions. It puts them all on one board, shows whose turn it is and what step each is on, and lets AI group them by topic. Works in iTerm2, Ghostty, Orca and more.
原文Alchemist Coder 是一款开源（AGPL-3.0）桌面应用，支持 macOS、Windows 和 Linux，通过 Agent Client Protocol（ACP）运行 Claude Code 和 Codex，并用实时树形视图展示每个智能体及其派生的子智能体。免费的社区版中，智能体只能使用本地模型（Ollama 或 LM Studio）。没有遥测，无需注册账号。适合同时使用多个编码智能体的开发者。 亮点 协调者“组织”把任务分派给各自 git worktree 中的智能体 Arena：最多 4 个智能体在各自的 worktree 里做同一个任务，再合并效果最好的那个 全文搜索 Claude Code、Codex、Gemini CLI 和 Grok Build 的历史会话 内置浏览器，智能体可以直接操作网页；Autopilot 会拒绝不可逆的操作 界面支持 11 种语言，包括简体中文 我是作者。项目创建于 2026-09-29，安装包暂未进行代码签名，首次打开时 macOS 和 Windows 会弹出安全提示（处理方法见 README）
HypAware collects every Claude Code, Codex, Cursor, OpenCode, Pi, OpenClaw, and Hermes session and shows what's broken, the evidence, the fix, and whether it stayed fixed. Free open source CLI; HypAware Cloud for the whole team.
原文I burned about $500 for this fun project: Small Python. Initial PoC used 3 resets of Codex in one session but all later development happens in Claude Code, also in one Opus 5.5 session over three days. The outcome is a quickjs-like implementation for Python. Almost full Python 3.15 language features plus builtins are covered. No Lib/ support. Great for embedding scenarios. No rust. Just plain C11 code. is here: https://github.com/soasme/small-python/ Ask me anything.
原文同时开着几个 Codex 会话，谁还在干活，谁已经等着你确认了？ Codex Office 把主 Agent、子 Agent 和独立会话，变成同一间 3D 办公室里的同事。点开角色就能查看任务和协作关系，再回到原会话继续处理。 ### 能用它做什么 - 看清每个 Agent 的工作状态，找到需要你关注的任务。 - 切换办公室、监狱和 2D 关系图。办公室里有工位，监狱里一人一间“独立办公室”，还可以换成人、牛或马的形象。 - 按项目、模型和日期查看 Token 用量、缓存命中和费用估算。估算按 API 价格折算，不是实际订阅账单。 ### 怎么用 这是一个在自己电脑上运行的免费开源工具，读取已有的本地 Codex 记录，用量索引和报表也保存在本地。不需要额外申请 API Key、注册 Codex Office 账号或部署数据库。 打开 GitHub，按中文快速开始安装，或者直接让你的 Codex 帮你安装并启动。需要 Python 3.10+、本地 Codex 记录和现代浏览器；3D 场景需要 WebGL2。 界面支持中文和英文，代码采用 MIT 许可证。欢迎试用、提问题或一起参与开发。 先看演示视频 配图展示了 36 个 Agent、3 个项目的演示场景，人物、项目和用量均为虚构演示数据。Codex Office 是独立社区项目，与 OpenAI 没有官方关联。
Flatbread turns a coding agent's session into one link: what was asked, every step, what broke, what fixed it and the diff. Redaction runs on your machine before anything uploads. Another agent can pull the link and pick up where it ended.
原文Vonitore is a free server monitoring platform for developers, homelab enthusiasts, and small hosting teams. Monitor CPU, RAM, network activity, and uptime across Linux, Windows, and macOS from one dashboard. Get alerts when servers go offline, manage multiple machines, and connect your servers with a simple one-command agent installation. No complicated setup, just the monitoring you need.
原文Agent Standup turns the Claude Code, Codex, Mistral Vibe and Cowork sessions on your computer into a live video call. The lead hands out work; sub-agents join, talk and share their screen: task list, test runs, diffs, the browser they drive, live command output. See at a glance who is working, stuck or waiting for you, and which agent spends the tokens. It only reads local session files: no extra tokens, nothing sent. Free, MIT, in VS Code, any terminal or as a desktop app.
原文Hushtop turns the notch on your MacBook into the one place you glance at. What's playing, the next meeting, AirPods and battery, an AirDrop shelf, link previews from your clipboard, and a small pet that reacts to your day. Running Claude Code or Codex? Their sessions show up in the notch and you answer permission prompts right there, without switching windows. Native Swift for macOS 15+ on Apple silicon, no account, no telemetry, notarized by Apple. Free for 7 days, then $9.99 once.
原文Yardlet is a free, open-source Mac app for the coding agents you already use. It starts Codex, Claude Code, OpenCode and others with their own CLI and sign-in, and gives them one sidebar, composer and set of approval cards. Sessions stay in each agent's own store, so a task started in Yardlet also opens in the Codex app. It writes nothing into the agents' configuration, so you can stop using it any time. Diffs, a pull request tab, connectors and a browser the agents drive come with it.
原文Frontier models improve at a pace faster than the pace at which most humans are able to keep up. Not all of us are obsessed with pushing the boundaries of our agents and trying new ways to work with them. I often spend 12+ hours a day working with an LLM, and still, when I talk to people, I hear of ways of working I hadn’t thought of. What if we could speed up that pace of learning? What if everybody shared their practices of how they are making the most of AI, and if we could learn from each other? What if AI could make that sharing happen? Could everyone catch up with AI then? I know, it sounds dreamy, but my therapist says illusions are necessary—until some of them come true. I built a tool that teaches your local agent to scan your sessions, spot the best practices you already use, and guide you towards the ones you haven’t discovered yet. Only names and counts reach the server. Your sessions never do. The product has a self-consolidating catalogue of practices, MCP servers, plugins and skills. Here’s my stack—scan (locally, all session data remains private). https://app.userrank.ai/@milstan To spice things up and give a reason to rescan, I added rankings inspired by Foursquare (who here is old enough to remember Foursquare when it was gamified?). You can rank in the top 10 of a city, or of a field like Product. Claude Code, 10–20 minutes on your own plan: claude plugin marketplace add https://app.userrank.ai/plugins/marketplace.json claude plugin install userrank@userrank claude "Get my UserRank" Codex, or a quick scan of your Claude or ChatGPT chats: https://app.userrank.ai/start
原文最近几个月我一直同时开着 Claude Code 、Codex 和 Gemini CLI 写代码。agent 本身跑得挺好，烦的是它每次要跑命令、改文件都停下来等我按「允许」。三个终端轮流翻，经常翻到的时候它已经等了十分钟；出去吃个饭，回来每个都停在第一条 git push 上。开 yolo 全放行又不敢，一句 rm -rf 出去就没有后悔药。 所以我写了 Railse：每条审批先过一遍规则，规则定不了的推到手机，我在手机上点一下，agent 接着跑。 它怎么做： - 三家 CLI 的权限钩子都接到本机一个守护进程，规则只写一套。 - 每条请求先定风险档：只读和改项目里的文件直接放行； curl 、npm install 这类联网的，git push 、deploy 这类对外的，rm -rf 、git reset --hard 这类不可逆的，都问。 - 定不了的推到手机，Web Push 通知，点开批准或拒绝。批准时勾「记住」，当场存成规则，下次同类的不问。 - 对外和不可逆的永远不能被「记住」放行，只能自己手写规则。 - Railse 自己挂了不卡 agent：守护进程没起、hub 连不上超过 30 秒，CLI 弹自己的审批框，最坏就是回到没装的样子。 hub 自己部署，一个 Go 二进制，也有 Fly.io 的现成配置；手机网页由 hub 托管，不用另外部署。命令里认得出的 token 、密码在本机就换成 再上报。 现状：v0.1.2 ，我自己在用。已知的限制：Gemini CLI 的钩子只能拒绝不能放行； Codex 的钩子只在交互会话里生效； iPhone 要加到主屏幕才能收推送； hub 是单机 SQLite 。 开源，Apache 2.0： 一行安装：curl -fsSL | sh 想听两种反馈：你同时开几个 agent 的时候是怎么处理审批的；规则表有判错的（该放行的问了、该问的放行了）直接提 issue ，我会优先看。
Vaaya Wallet is a prepaid wallet for your AI agents. Sign up and get $2 in free credit. Connect Claude, ChatGPT, Grok, Muse, Instinct or your own agent, and give each one its own key with a daily, weekly or monthly limit. Your card details are never shared with the agent. Calls past the limit are refused and every charge shows up as it happens. Revoke a key any time, or flip the stop-at-zero switch to halt every agent. Teams can share one balance with per-agent keys and one transaction log.
原文An MIT-licensed coding agent for your browser and terminal. Bring your own model API, edit files, preview HTML, export ZIPs, and extend the browser agent with Rust/WASM plugins. Free software; model API usage is billed separately.
原文自荐一个 AI 辅助制作的中文开源练习，面向网页接单新手：原本说好做三页静态网站，客户又想“顺便”加会员、预约和付款，怎么先把范围与未知说清楚？ 项目： 免费版使用一个固定虚构案例。使用者自己回应，AI 再引用实际原句，指出一项已澄清的边界和一项仍需确认的问题，不替使用者回答，不给职业或收入保证。 两种免费体验 不下载：复制 README 的单轮提示到自己已有的 AI 对话中，先回答一次；这种方式不提供本地保存与恢复。 需要保存：下载免费的本地 Skill，用具备文件读写和 Python 3.9+ 的 Agent 主持；自备模型权限和额度。 免费仓库按 MIT 开源。仓库另说明同作者的独立付费版本，本次只邀请体验免费版，不要求购买。普通 AI 也能做基础沟通练习，固定流程是否有额外帮助仍需要真实反馈。 想听到的反馈 你实际试了哪种方式？哪里最费劲？自己回应后，哪一句反馈有帮助或没有帮助？还没试也欢迎说明原因。可用 README 的体验表单或邮箱反馈，也可以在这里直接讨论。 请只使用本项目的虚构案例，不提供真实客户、订单或个人资料。当前只有本地技术核查，没有真人学习效果或成交证明；此次希望先找到愿意亲自试一轮的人。
Mac App Store： Windows：winget install Trappist.AgentCat Homebrew：brew install --cask yong076/tap/agent-cat 工具介绍 用 Claude Code、Codex、Gemini CLI 写代码时，最常遇到的问题是：这个账号还剩多少额度？什么时候重置？换哪个账号还能继续？每个工具的查看方式都不一样，账号一多就更乱。 Agent Cat 是一款免费的 macOS 菜单栏 / Windows 托盘应用，把这些信息集中在一个地方。 核心功能 支持 Claude Code、Codex、Gemini CLI、GitHub Copilot、Cursor、Grok、Kimi、OpenCode 等 同一服务的多个账号分别显示剩余额度和重置时间；某个账号用完时，会提示可以换用的账号 7 天 token 用量图表、用量历史与报告 只读取各 CLI 已经写在本机的数据，无需额外登录；负责读取的 connector 源码公开（PolyForm Shield 许可，属于 source-available）： 界面支持简体中文，另有 iOS / Android 手机端 完全免费，没有订阅 我是作者（韩国的独立开发者）。如果合适，希望能被周刊收录，谢谢！
A free, open-source skill for Claude Code, Codex, and Cursor. Give your agent your app captures and a short brief, and it builds a local screenshot editor in your project. Choose from 18 styles, compare looks in Style Lab, edit the copy, and export every size for iPhone, iPad, Mac, and Android, in multiple languages. No screenshot subscription. It runs locally; your usual agent costs apply. Install: npx skills add ParthJadhav/app-store-screenshots
原文Import a recorded agent trace (JSON/JSONL), replay it as a timeline, see loops, retries, failed tools and token growth flagged, compare the fixed run against the broken one, and export a report with secrets masked. One offline HTML file, $49 once.
原文一键切换 AI 编程助手的模型。这是一款用 Go 开发的 AI 编程助手模型管理工具，让 Claude Code 用上 Kimi、Codex 用上 DeepSeek。它通过本地网关转换 OpenAI、Anthropic 和 Gemini 的接口格式，支持多种编程助手和模型服务商，还能把订阅共享给其他助手使用，额度用完后自动切换账号。
谷歌开源的 AI 手机自动化工具。这是一款 AI 驱动的 Android 自动化工具，只需用自然语言描述任务，AI 就会自动操作手机或模拟器，完成跨应用的测试流程和日常任务。它内置 MCP 服务，能让 Claude Code、Codex 等 AI 编程助手直接操控手机并收集日志和截图，支持 Gemini、Claude、GPT-4o 等多模态模型。
Nano Metrics adds a live dashboard to your VS Code sidebar: Claude Code status, tool calls, and task progress; token usage and estimated costs from local transcripts; edited files and recent activity; CPU, memory, disk, GPU, per-core load, and temperature; optional monitoring for 30+ services like Docker, Postgres, Redis, Kafka, and Ollama; process-level usage; and a theme-aware status bar. AI coding activity, costs, and system metrics, all in one place.
原文Hey HN, happy to share Chopstick with you. It's a Mac app that makes your notch useful, but most importantly customizable and dynamic. What makes it unique is that you can build custom widgets with your coding agent. Chopstick automatically learns from your usage and suggests new widgets with time. Widgets are dynamic meaning they can trigger in specific situations. The obvious case is "when Spotify is playing", but you can also get more complex and say show the widget to push my changes when VS Code is open and I have changes. TLDR: - Chopstick gives you widgets in your Mac notch - You can build your own widgets with your existing agent - Widgets are highly customizable in what they can do and what triggers them - It's a one time payment (with trial), discounted for launch Feel free to join the discord as well for sharing widgets! https://discord.gg/kqbnvgeTQC Let me know your feedback! - Josh
原文I run my business on four Claude Code agents: Titus, Pheme, Davinci and Metis. They write my newsletter, The AIgent (about 9k subscribers), every day through a drafting and proofreading engine that Pheme checks before scheduling. They also build AI courses, run a pipeline that turns the day's featured into a talking-head video of my avatar for Instagram, set up ad deals, and run Meta ads and email funnels with me. Nothing closes until I sign off. Metis posts jobs on a job board and the others claim them. The hard part wasn't any single task. It was keeping the agents coherent over weeks. Compaction kept dropping the small decisions, and then an agent would redo work I'd already ruled out. aigent-os is my fix. At a context threshold (80% by default) the agent writes a capsule, a checkpoint file in git, and has to reply with one exact line. Only then does the runner send /clear, wait for a SessionStart receipt proving a fresh session started, and resume. One ack buys at most one clear. If anything is ambiguous it stops at HOLD:clear-stalled-after-ack and doesn't retry. On resume the capsule is reference only, live state wins, and anything the writer couldn't capture is written as Unknown. First run interviews you with three plain questions and makes a first plan. Each agent gets its own memory vault. Limits, honestly: the public evidence is 11 of 11 cycles on one Windows seat. The runner can still miss a permission dialog (#104), a seat launched from inside another Claude Code session can lose its transcript (#98), and scrolling the seat window sends stray input (#99). v1.1.0 shipped on Sunday. Try bash install.sh --dry-run, it writes nothing. I'd like to hear where the loop could still double clear or strand a session.
原文Muha 是一个 MIT 开源的 TypeScript SDK，用于把已有Coding Harness接进脚本和应用。它统一处理启动、会话、事件和交互，让调用方用代码连接不同 Agent 的任务。
Paste your git log; the grader lists every file two sessions (two authors, or two branches) changed less than an hour apart, ranked by how often it happened. Those are the edits most likely to undo each other when Claude Code, Codex or a teammate work in parallel. It runs entirely in your browser: nothing is uploaded, no signup. Try the sample log to see a report in one click. A README badge shows the count for a public GitHub repository.
原文Spacesheep (YC F26) turns any agent's work into a space: an interactive doc, live dashboard, mockup or app. Share it with specific people, your team or publicly, and collaborate on it together.
原文按用量计费的 AI 模型网关，一个 Key 双协议通用：Claude Code 设置 ANTHROPIC_BASE_URL、OpenAI SDK 改 base_url 即可接入，价格公开、每笔请求成本可查，支持组织内多发 Key 与用量审计、多上游自动故障转移；名字与 logo 源自 HTTP 彩蛋 418 I'm a teapot
Token Kingdom turns your Claude Code usage into a cozy 3D village. Every token you spend becomes gold, wood and stone. Build, fish, cook, defend against raids and visit your friends' kingdoms. One plugin install. Free.
原文Nekot shows every Claude Code and Codex usage window on your iPhone, when each one resets, and alerts you at 80% and 95% so you never hit the limit mid-task. Your tokens stay in the Keychain: no Nekot server, no account.
原文Pling is a real SSH, Mosh and SFTP terminal for iPhone, iPad and Android, built for working with coding agents. Jump back into tmux or Zellij sessions, approve Claude Code's risky commands from a push notification, or describe a task and let the Command Agent run it. Plus push alerts from any script with one HTTP POST, live server monitoring, remote desktop and end-to-end encrypted sync. Free to start; Pling Pro unlocks the rest.
原文把所有 AI 编码智能体收进同一个地方 —— 并让它们协同工作。 亮点 可以让任何兼容ACP的智能体在一个界面协同工作，支持桌面、服务器、Docker 与移动端，也能通过 Telegram、飞书和微信与智能体对话。 截图或演示视频
Kinaya is a keyboard-first browser for macOS. Ads and trackers are blocked inside the engine (Brave's adblock-rust with EasyList and EasyPrivacy) before a request leaves, and there is no telemetry. A built-in MCP server lets Claude Code read pages, click, type and take in its own background tab, using your real logins. Also: command palette, private and Tor tabs, encrypted password vault, reader mode. $19 once, 14-day free trial.
原文Session and weekly limits for Claude Code, Codex, Cursor and Grok, with a pace check that tells you if you'll make it to the reset.
原文See every Claude Code session on one live screen: which one is working, which needs you, what its agents are doing and what it costs. Each chat runs in its own window and keeps going in the background. Free and open source.
原文Graphicle is a Rust port of Graphify's open-source code-graph engine, 3 to 22.7x faster than Graphify 0.9.91 in all 39 of our measurements, plus project memory: docs, decisions and past agent sessions, cited to file and line over MCP. Local, Apache-2.0.
原文TraxAgent is an AI spend intelligence platform: it turns your Claude and OpenAI APIs, coding agents like Claude Code, Codex and Cursor, and Claude and ChatGPT AI usage from across your domains into actionable insights and consolodated spend. Risks flagged early, costs cut, opportunities found, each with the exact fix for optimization. Read-only, no prompts stored.
原文Turn your Claude Code and Codex sessions into a living fantasy village. See agent activity at a glance, then click a hero to inspect context usage, skills, plugins and subagents. Agent Quest reads existing local session logs while you keep using your usual tools. Keep multiple sessions in sight on a second screen. Customize the village and export Markdown or JSON reports for review. Free, MIT-licensed and open source. Runs locally in your browser with Bun.
原文一个 Windows 无边框、无任务栏图标的 AI 订阅额度 HUD 悬浮窗与副屏仪表盘，同屏实时监控 Claude Code、ChatGPT Codex、Grok、Google Antigravity、LiteLLM 等多账号的 5 小时与 7 天限额窗口；支持一键探测并兑换官方隐藏的限额重置券；双击 Ctrl 秒切 mpv 动态壁纸。 解决了什么痛点？ 1. 多账号与多平台额度盲盒：做严肃编码任务时，多开 Claude Code / Codex / Grok / agy 经常突然撞上 5 小时限额墙或周上限。以往要在各个网页或 CLI 里繁琐查询，usage-float 直接以极低资源占用在副屏或桌面角常驻展示全部账号的动态进度条与剩余重置时间。 2. 官方隐藏福利一键兑换： Anthropic 为 Claude Code CLI 用户提供了周重置券（cedarember）与 5 小时碰墙重置券（junipertide）； OpenAI 官方对部分 ChatGPT / Codex 订阅也发放了限额重置券（/backend-api/wham/rate-limit-reset-credits）； usage-float 模拟官方 CLI 原生协议安全嗅探重置券资格，有券时直接在界面提示，点击即可一键兑换回满血，绝不浪费。 3. 无感常驻与桌面摸鱼/护眼切换：副屏平时是干净专业的黑金/暗黑仪表盘，双击 Ctrl（或自定义热键）无缝切换为你本地收藏的 mpv 4K/60fps 动态壁纸或画廊，兼顾生产力与视觉愉悦。 4. Agent 本地查询友好：除了图形界面，提供 python usagefloat.py once（或读取本地原子写入的 ~/.usage-float/usage-cache.json），方便本地 Agent 路由器无网络请求开销直接获取账号剩余额度。 配套开源工具自荐（同一开发者生态） 为了让多订阅 AI 编码体验更完整，我们在同一生态下还开源了另外两个轻量实用的工具，供感兴趣的同学参考： 1. AgentCompass
A calm Mac app for running Claude Code sessions side by side. Perch keeps every session, PR, and link for a piece of work in one place, and tells you when something needs you.
原文ssh zydra.sh There's no sign-up or anything, and you get a 15-minute of a graphical Linux VM drawn entirely in your terminal. It has a browser, an editor, and it runs Doom. It works best in Kitty or Ghostty. WezTerm and Konsole also work, as well as Sixel and iTerm2 terminals, though with less detail. Text-only terminals support non-graphical apps: shells, editors, and other TUIs. Why? For a few months I ran coding agents in tmux on a VPS over SSH, and it sucked. I couldn't paste images, scrolling broke and I didn't wanna figure out why, I had no notifications for when an agent finished, I couldn't see browsers agents opened to smoke-test, logins drifted between my laptop and the server, I couldn't check in from my phone, and to know what an agent had done five minutes ago I had to read through opaque chunks of tool calls. These issues added up and had me in a limbo: They made my setup bad enough to annoy me every day, but not quite bad enough to stop my real work and fix it. How it works: Zydra is a custom (Smithay-based) Wayland compositor in a microVM with an AMD GPU, implementing a series of terminal protocols that solve many issues but are often ignored: Kitty graphics for windows, OSC 52 for the clipboard, images included, OSC 9/99/777 for notifications, SGR mouse for scrolling and clicks. VNC is limited to sending pictures of the screen and diffing frames to find what changed. In Zydra, apps directly tell the compositor what changed, and terminal text is sent as text you can select and copy natively. The browser is a WebKit build that delivers frames to the compositor as DMA-BUFs without a readback, and agents drive it through Playwright. Alt+P is sort of like Raycast: browse apps, linked GitHub, websites, and SSH targets from it. You can watch the desktop at any point in time and rewind to periodic snapshots. To achieve this, the desktop records into a timeline file, a kind of structured video, and the VM takes snapshots along it. You can traverse back like a livestream to see what happened while you were away, then restore or fork the whole machine from that point. Agents can mark points on the timeline and checkpoint the whole machine before risky changes. Other things to try: ssh -t zydra.sh dino # the Chromium dino game via SSH tar -czf - -C ~ .claude .claude.json | ssh -T zydra.sh upload # bring your configs (includes your Claude login) ssh -o ObscureKeystrokeTiming=no -A zydra.sh # lower latency & forward your SSH agent ssh -p 443 zydra.sh # connect from a corporate firewall sshfs zydra.sh: ~/zydra # paid accounts: mount your home Everything runs on one dedicated server with one AMD iGPU, so there are 12 slots. When they're full you wait in a queue and can play Chromium's offline dino game. The paid plan is $40/month (+ tax): 8 GB RAM, 4 vCPUs, 80 GB disk, GPU access, with no time limit and a 3-day money-back guarantee. I'll be around all day to answer questions.
原文Monitor and control Claude Code, Codex, Gemini CLI, Grok, Cursor, and Scoot (local models) from your phone. Permission prompts arrive as a push with the real question and tappable answers. Follow the live conversation feed, drop a note mid-session, or schedule a prompt for 2 AM when your token limit resets. The engine, aaw-core, is now open source: MIT, on GitHub and PyPI. No accounts: pair by QR code, end-to-end encrypted, self-hostable relay. Free for one computer, with every feature.
原文Browser console and diagnostic chamber for Oh-My-Pi (omp) agent sessions — live timeline, steering, real PTY, git/diff/wiki panels, and a panel-plugin platform. Elysia + Preact on Bun. - kabmo-labs...
原文SWARMFORGE is an agentic CI/CD gate for the change path that matters: it reviews a committed or working-tree diff, can group and fix confirmed findings in isolated sandboxes, re-reviews the result, then requires deterministic repository checks before it publishes a PR. It runs with OpenAI Codex/GPT, any OpenRouter model, local open-source models (Ollama, LM Studio), or Claude Code (claude -p, opt-in) — with cost-aware routing measured on an L0–L6 benchmark ladder.
原文View and search Claude Code, Cursor, GitHub Copilot CLI and Codex session logs in your browser. Live refresh, export and local-only processing.
原文阮老师好，自荐一下我做的 Review Inbox。
Hey HN! As an avid user of cmux ( https://cmux.com/ ), I wanted a way to control terminal and agent sessions from my Android phone, so I built Muxer. The app talks to a tiny relay over HTTPS, that then talks to cmux through its local socket, and gives you to ability to: - Answer agent permission requests, questions and plans from a notification - Reply to Claude Code, Codex and Pi chat sessions with text, voice, @ file mentions and photos - Receive all cmux notification, with inline replies, bubbles and background delivery. - Use a live terminal with sticky modifier keys and local scrollback - Share text and links from any app to an agent - Organize workspaces - Manage cmux Cloud VMs and automations - Lock the app with your fingerprint, face or screen lock Enjoy, and let me know what you think!
原文Coding agents rewrite code you never asked them to touch, and the tests still pass. Sonah reads the agent’s Claude Code or Codex session along with the code, then shows on one page: whether each ask was done, what the agent decided on its own, which of its claims its own results contradict, and what is wrong in the code. Every quote is checked word for word. Runs locally through your own Claude login. No account, no telemetry. Free for individuals.
原文Hand off work between coding agents: Claude Code to Codex, Codex to Devin, and more. Every session kept on your machine, searchable and replayable. Open source, local-first.
原文Since our first launch: 22 updates in 17 days, many shipped the same day a user asked. SwarmPane runs Claude Code, Codex, Gemini, Cursor and 9 more AI coding agents side by side, on the plans you already pay for. New: every account's usage limits in your toolbar, each agent in its own worktree, Devin and Hermes, one rules file and MCP setup for every agent, and sessions that survive restarts. Mac, Windows, Linux.
原文换一个关键词或分类试试。
