Chase Chou @janus_path
⚙️ Engineer by craft, philosopher by practice. 🏗️ Business-logic & technology architect. 🧩 INTJ-A. https://t.co/rqxmTSr9t5 Joined March 2021-
Tweets416
-
Followers248
-
Following624
-
Likes574
@FradSer 所以今天我整了一个方案,把OpenCode Go整合进codex,让gpt-sol指挥 OpenCode Go的便宜模型干活 x.com/janus_path/sta…
最近codex不再重置额度了,额度不够用了🤣 所以,尝试一套低成本 Codex Agent 方案: Codex Desktop + lidge-jun/opencodex + OpenCode Go (GPT-5.6 Luna) 我比较看重它的一点是:不需要放弃 Codex Desktop 和 Codex Harness。 最近在测试一套我觉得很有意思的低成本 Codex Agent 方案: Codex
@LinearUncle 我也是今天刚发现官方有API,然后立马尝试接入OpenCodex x.com/janus_path/sta…
最近codex不再重置额度了,额度不够用了🤣 所以,尝试一套低成本 Codex Agent 方案: Codex Desktop + lidge-jun/opencodex + OpenCode Go (GPT-5.6 Luna) 我比较看重它的一点是:不需要放弃 Codex Desktop 和 Codex Harness。 最近在测试一套我觉得很有意思的低成本 Codex Agent 方案: Codex
Setup is also simpler than it looks. If you already have Codex Desktop and OpenCode Go, you can ask Codex itself to configure the whole thing: Install and configure lidge-jun/opencodex on my macOS machine. Goal: - Keep official Codex Desktop and GPT-5.6 Sol as the main agent - Connect OpenCode Go - Use GPT-5.6 Luna as a lower-cost sub-agent - Use Codex native spawn_agent for delegation - Prefer the configuration that works reliably for cross-provider delegation - Do not replace or break my official Codex login or native model setup - Inspect my current environment before changing anything - Verify one real sub-agent invocation after setup OpenCodex: github.com/lidge-jun/open… Then let Codex read the docs, install, configure, and validate it. Next I’m going to test this on real development tasks: Sol → Luna → Sol Review The question I care about is simple: How much Codex quality can we retain while reducing expensive model usage?
I’ve been testing a low-cost Codex Agent setup that looks surprisingly promising: Codex Desktop + lidge-jun/opencodex + OpenCode Go (GPT-5.6 Luna) The idea is not to replace Codex. Instead: Keep official Codex / GPT-5.6 Sol as the manager, and use Codex’s native spawn_agent to delegate well-scoped tasks to a lower-cost GPT-5.6 Luna worker. Roughly: Codex Desktop → GPT-5.6 Sol: understand, plan, delegate, review → native spawn_agent → OpenCodex: sub-agent model routing → OpenCode Go / GPT-5.6 Luna: execution → Sol reviews the result The key idea is: Keep the Codex Harness. Lower the worker model cost. OpenCode-AI: opencode.ai/go?ref=H4NREZ3…
搭建其实比想象中简单。 如果你已经有 Codex Desktop 和 OpenCode Go,可以直接让 Codex 帮你完成: 帮我在 macOS 上安装并配置 lidge-jun/opencodex。 目标: - 保留官方 Codex Desktop 和 GPT-5.6 Sol 作为主 Agent - 接入 OpenCode Go - 使用 GPT-5.6 Luna 作为低成本 sub-agent - 使用 Codex 原生 spawn_agent 进行任务委派 - 优先使用适合跨 Provider delegation 的配置 - 不替换我的官方 Codex 登录和原生模型配置 - 修改配置前先检查我当前环境,完成后验证一次 sub-agent 调用 OpenCodex: github.com/lidge-jun/open… 然后让 Codex 自己读项目文档、安装、配置和验证即可。 我准备继续拿真实开发任务测试: Sol → Luna → Sol Review 看看相比全程使用 Sol,实际能节省多少额度,以及任务质量会下降多少。
最近codex不再重置额度了,额度不够用了🤣 所以,尝试一套低成本 Codex Agent 方案: Codex Desktop + lidge-jun/opencodex + OpenCode Go (GPT-5.6 Luna) 我比较看重它的一点是:不需要放弃 Codex Desktop 和 Codex Harness。 最近在测试一套我觉得很有意思的低成本 Codex Agent 方案: Codex Desktop + lidge-jun/opencodex + OpenCode Go (GPT-5.6 Luna) 核心不是“把 Codex 换成另一个模型”,而是: 继续让官方 Codex / GPT-5.6 Sol 做主管,通过 Codex 原生 spawn_agent,把边界清晰的任务交给更低成本的 GPT-5.6 Luna 执行。 大致结构: Codex Desktop → GPT-5.6 Sol:理解、规划、拆任务、验收 → native spawn_agent → OpenCodex:子 Agent 模型路由 → OpenCode Go / GPT-5.6 Luna:执行 → Sol 最终 Review 这样保留下来的其实是最有价值的部分: Codex Desktop + Codex Harness + Sol 的复杂上下文理解能力。 降低的只是部分执行任务的模型成本。 主 Agent 仍然可以使用官方 Codex 负责理解复杂上下文、规划和验收,而通过 opencodex,把部分明确、低风险的任务交给 OpenCode Go 中的模型执行。 换句话说: 保留 Codex 的 Harness,只降低部分任务的模型成本。 接下来准备和官方 Codex、Pi + Skills 做一些实际任务对比,看看这种组合到底能保留多少 Codex 的生产能力。 OpenCode Go首月只需要5刀: opencode.ai/go?ref=H4NREZ3…
虽然我一直使用codex,但是为了降低使用成本,今天我尝试了社区方案,pi + OpenCode Go(GPT 5.6 Luna). 我不建议小白使用社区方案,推荐还是使用codex这种成熟产品。 因为社区方案会有很多问题需要去解决,这里我分享一个最快捷简单的方式,就是使用codex去驾驭pi,让codex去检查pi的一些问题,比如: - Pi 是否能正确识别 AGENTS.md - Pi 是否能发现工作区已有的 skill - 增加命令安全护栏(Pi 本身没有内置沙箱,默认拥有当前用户的系统权限,因此仅靠提示词不够。) - 建立 Codex → Pi 任务模板
@cxjwin 没有显式MQ也可以,一个小系统要啥MQ,一上来就是大而全的去造火箭,事实上很多团队自行车还没造完就玩完了,或者说只需要个自行车就够用了,可以出去忽悠了……
@dotey 我倾向于 全栈(偏基建架构)+全栈(偏前端)+全栈(偏后端)+QA+产品(UI/UE) 虽然AI极大的提高了软件工程的效率,可以取代一部分人,但是不代表软件工程中的角色会消失。 x.com/janus_path/sta…
Build a plugin once and use it across compatible agent clients. Introducing Agent Plugins, an open standard developed with @awsdevelopers, @cursor_ai, @github, @code, and @vercel that packages Agent Skills and supports MCP server configurations in a shared format.
@0xMilkRabbit 我刚也想出来一个非常节省codex token的方法。 x.com/janus_path/sta…
我看到很多博主建议大家使用codex的时候,为了节省token,教去配置子智能体使用luna模型,但是其实有一个更方便的、更节省token的方式,就是让sol自己判断分配工作去使用GPT-5.3-Codex-Spark 模型,然后让sol检查任务结果,而且GPT-5.3-Codex-Spark 模型的额度是单独计算的。 x.com/Saccc_c/status…
@Youngxxxxu 我刚想出来一个节省token的好法子 x.com/janus_path/sta…
我看到很多博主建议大家使用codex的时候,为了节省token,教去配置子智能体使用luna模型,但是其实有一个更方便的、更节省token的方式,就是让sol自己判断分配工作去使用GPT-5.3-Codex-Spark 模型,然后让sol检查任务结果,而且GPT-5.3-Codex-Spark 模型的额度是单独计算的。 x.com/Saccc_c/status…
我看到很多博主建议大家使用codex的时候,为了节省token,教去配置子智能体使用luna模型,但是其实有一个更方便的、更节省token的方式,就是让sol自己判断分配工作去使用GPT-5.3-Codex-Spark 模型,然后让sol检查任务结果,而且GPT-5.3-Codex-Spark 模型的额度是单独计算的。 x.com/Saccc_c/status…
如何能让你的Codex拥有无限子弹? 将子代理设为luna-max。复杂任务用sol去做方案设计,用luna-max去并行执行 具体设置方法: 在Codex对话中输入下面的提示词: 在 ~/.codex/agents/luna-worker.toml 创建一个名为 luna_worker 的自定义代理。 使用以下设置: model = "gpt-5.6-luna"
Agent 越强,我们越需要软件工程。 AI 让代码生成越来越便宜,但理解系统、控制复杂度、避免跑偏、保持长期一致性并没有变得免费。 好的架构,本质上是在减少 Agent 的重复推理。 Architecture is cached reasoning. 写了一篇文章,聊聊为什么软件工程在 Agent 时代不是过时了,而是价值正在发生变化。 blog.janus-path.com/2026/08/05/Age…
@cholf5 额度告急,真是捉急,不知道要不要使用仅剩的一次额度重置,用了又怕突然免费重置,不用的话眼瞅就要停工摆烂……🤣
Tibo, one piece of feedback on the current Codex model experience: With Sol, Terra, and Luna, plus multiple reasoning-effort levels, model selection is becoming a two-dimensional decision that users have to make before they even start working. For users who don't understand the differences, it's hard to choose correctly. For experienced users, repeatedly choosing the right model and effort for every thread or task is unnecessary overhead. I wonder if Codex could introduce an adaptive model routing / delegation mode. The user could choose only a high-level strategy such as: Auto Fast Balanced Max Then Codex itself could dynamically route subtasks to different models and reasoning levels based on task complexity. For example: simple file search / mechanical operations → Luna low routine implementation / tests → Terra medium complex implementation / debugging → Terra high architecture / high-uncertainty reasoning → Sol high or xhigh And if a lower-cost model fails validation or becomes uncertain, Codex could automatically escalate. Advanced users could still manually pin a model when needed. I think this would simplify the UX significantly while also reducing unnecessary token usage. More importantly, model selection feels like something the Codex runtime should increasingly optimize for the user, rather than requiring the user to manually orchestrate the model hierarchy.
我开源了 KeptNear,一个本地优先的密码与令牌管理器。它不只服务于人,我还在构建本地 Broker,让 AI Agent 和本地工具只能执行用户明确授权的凭据操作,避免原始 Token 出现在对话或工具输出中。 当前预览版优先完善 macOS 密码与令牌管理体验。Broker、MCP 和 CLI 仍处于源码级开发者预览,尚未作为最终用户能力激活。KeptNear 当前未签名、未经过外部安全审计,请暂时不要存储生产环境凭据。 github.com/chasechou007/K…
The current preview focuses on the macOS password and token experience. Broker, MCP and CLI access remain a source-level developer preview. KeptNear is unsigned and unaudited, so don’t use production credentials yet.
I open-sourced KeptNear, a local-first password & token manager. Beyond human use, I’m building a local Broker so AI agents and tools can perform user-approved credential operations without receiving raw tokens in chats or tool output. github.com/chasechou007/K…
Kendeng | AI Visuals @kendeng_me
457 Followers 745 Following I create AI videos and images—and show how they’re made. 🎬 Breakdowns · 🧪 AI model tests · ✍️ Prompts & workflows
eLongZ @eLongZq
1K Followers 3K Following Chủ/Nhà sáng lập Web & App, quan tâm đến AI, SEO, thị trường Việt Nam và tăng trưởng sản phẩm độc lập (indie product). Đang mài giũa các sản phẩm hướng đến nội
Swift Anna @AnnaSwift41811
0 Followers 38 Following
Ajexonlin @AjexejA
8 Followers 516 Following
Bob @Barryw10000
61 Followers 2K Following
Marcell Havlik @cviklihamar
2K Followers 2K Following I'm using AI in 24/7 and sharing what I learn! AI dev at @todoforai Father of two
daoping @daopingxu
55 Followers 2K Following
yiyi gao @yiyi44990
2 Followers 91 Following
Gang Liu @GangLiu2013
23 Followers 768 Following
mo mo @memewu_0002
2 Followers 118 Following
Starock @MaoXingyan75512
3 Followers 107 Following
张月 @zhangmeinvdetui
0 Followers 73 Following
fulanliu @liujiangtao8
63 Followers 3K Following
Sky @Sky1814469
1 Followers 76 Following
shi zhang @shi_zhang73425
5 Followers 340 Following
Kabochar @kabochar6
30 Followers 1K Following Social university student. I built: https://t.co/uMMMr5eyBd, https://t.co/2ApCj61NFm
404 @FallenFairy404
16 Followers 635 Following
JohnfiSh @moneymaker586
6 Followers 204 Following
duox1ng @duoxi1ng
5 Followers 404 Following
平凡OnlOOker @onlookersh
6K Followers 7K Following 🐧 前腾讯产品经理 💼 现某500强 AI战略发展总监 🤖 AI Coach(服务 OPC 与个体) 🏥 专注医疗、工具及流程管理领域 💻 Arch Linux / Vim 爱好者 ❤️ 西装/雪茄/威士忌/健身/宠物/自驾游/后摇/Tango 🔞 nsfw小号不喜勿看 @021onlooker
野猪队长阿当Ada... @AdamWang1186123
3K Followers 4K Following 野猪队长Adam Official: [email protected] | Verify on https://t.co/sYaRfKKbFz 辉煌过,也归零过 专注:交易所合作·量化团队对接·交易量起飞 已帮多个项目做到日活/量级翻倍 DM:https://t.co/R9mqc6FTqs 蓝V互关
Monkey D. Luffy @LuffyLibertario
8K Followers 8K Following KUKA ES MALA PALABRA | EL PERONISMO ES FASCISTA | LUFFY ES LIBERTARIO PAPÓN!
皮克文 @Stark1Wen
2K Followers 7K Following
saxon •M• @saxon_musk096
236 Followers 3K Following Son of Elon Musk. Self Media. MEGA 🇺🇸❤️I'll never speak ill of my Dad. Stay safe, God Bless 🇺🇸
David Tsai/蔡慎坤 @cskun1989
462K Followers 7K Following 中文撰稿人、独立时评人。 油管地址:https://t.co/jctvsg54i8 邮箱地址:[email protected]
Alex Chen @AlexChe31659937
120K Followers 89K Following 从纽约看世界,提供价值和底层逻辑。欢迎“订阅油管字幕频道:纽约二少爷 https://t.co/P11TSGO90z ” 如果你也有油管频道请告知。
The InvestiGator @InvestiGator182
120 Followers 345 Following Professional overthinker. I turn data into opinions and opinions into charts. Markets, geopolitics, whatever keeps me up at night. Not financial advice.🐊
Doge Father @elon_mus0137
317 Followers 6K Following
freemoney @haaahaaahaaa9
185 Followers 870 Following Other people get free money so you should too! I find the best class-action settlements to get you paid.
Max For AI @MaxForAI
26K Followers 3K Following 本科辍学创业,没啥好看的,也就发点AI相关的内容🫡 Head of growth @lobehub(81K🌟) Prev @listenhub @Sapient_Int 同名小红书4万粉丝,公众号01Founder(长文首发) 观点仅代表个人 更多请访问网站⬇️(欢迎商务、合作
Marina Budarina @marina_uiux
85K Followers 241 Following running https://t.co/UZqdX8dQEa - all in one design solutions for startups | 10k+ students at https://t.co/4EiuArBCeu
Om Patel @om_patel5
27K Followers 221 Following 🇨🇦 growing https://t.co/Tr6rRuOdX9 ($10k/m) + https://t.co/K8xdwRDHqX building cool stuff
Codex Releases @CodexReleases
22K Followers 3 Following Unofficial Codex release updates creds @OpenAI
鱼总聊AI @AI_Jasonyu
58K Followers 2K Following 👋 AI & 出海& Saas & APP & 海外手机卡eSIM实操干货分享 📈 8年产品经验 | 单款产品营收 $200k+ (全自然流量 )💻 代表作:@PaywallPro1 付费墙Agent:https://t.co/YJBXCadlR1 🍼 2胎爸爸 | 私信聊合作/咨询:jasonyu110
Sawyer Merritt @SawyerMerritt
1.1M Followers 480 Following EVs/space/tech. Bringing you the latest news in a single, easy-to-read feed. $TSLA & $SPCX investor, Model Y owner.
嫣敏🍥 @yanmin1105
2K Followers 472 Following 偶尔犯点病,其他时候很好的一只小杂鱼(划掉)小药娘丨07丨MTF丨家长党丨INFP-T丨天津人丨小八单推人丨速速来fo我喵丨已🔪🥚丨不抽烟不喝酒丨家有手台,喜爱出勤丨
Bob McGrew @bobmcgrewai
34K Followers 1K Following Learning new things. Former Chief Research Officer at OpenAI, early exec at Palantir, early employee at Paypal.
Mark Gurman @markgurman
468K Followers 2K Following Breaking News on Apple & Tech. Bloomberg Managing Editor. @UMSI Board Member. Send secure tips on Signal: markgurman.01 or email [email protected].
Jan Leike @janleike
135K Followers 334 Following AI research @AnthropicAI. Previously OpenAI & DeepMind. Optimizing for a post-AGI future where humanity flourishes. Opinions aren't my employer's.
Kevin Gu @kevingu
5K Followers 414 Following co-founder @dexbythirdlayer // prev research @jumptrading, @meta, @harvard
Codex Changelog @Codex_Changelog
28K Followers 1 Following Unofficial Changelog for OpenAI's Codex CLI
X Freeze @XFreeze
256K Followers 2K Following
文部科学省 MEX... @mextjapan
460K Followers 44 Following 文部科学省の公式アカウントです。 教育、科学技術・学術、スポーツ、文化に関する情報を広く発信していきます。 Ministry of Education, Culture, Sports, Science and Technology(MEXT) 公式サイト:https://t.co/O94D1M8l91
NHK科学文化部 @nhk_kabun
343K Followers 22 Following NHK報道局科学文化部=「かぶん」のツイッターです。「科学」と「文化」の専門記者たちが取材した様々な話題や、東日本大震災、原発事故に関するニュースをツイートしています。 ▼利用規約はこちら→https://t.co/x16pgRMKMc ▼フォローの考え方はこちら→https://t.co/LCciRdyr7b
OpenAI Developers @OpenAIDevs
391K Followers 1 Following Official updates for developers building with Codex & the OpenAI Platform • Service status: https://t.co/kZwnwdYYEq
Roelof Botha @roelofbotha
126K Followers 399 Following Investor. Formerly at @Sequoia and @PayPal. Board member at @SpaceX @nateragenetics @MongoDB @Square @Unity3D and @Ethos. Invested in @YouTube and @Instagram.
Jessica Livingston @jesslivingston
364K Followers 111 Following Cofounder, Y Combinator; Author, Founders at Work; Host, The Social Radars podcast.
Alfred Lin @Alfred_Lin
132K Followers 359 Following Partner @sequoia. Working w/ founders from idea to IPO & beyond: @airbnb @doordash @citsecurities @kalshi @clay @foundforbiz @Nominal_io @zipline
Unitree @UnitreeRobotics
145K Followers 314 Following High performance civilian robot manufacturer. Please everyone be sure to use the robot in a Friendly and Safe manner. https://t.co/hI6LafokVm
WTF Footage 👀 @brutalfootages
84K Followers 11 Following Follow for the wildest and craziest footage on the internet
White House History @WhiteHouseHstry
386K Followers 1K Following Founded in 1961 by First Lady Jacqueline Kennedy, the WHHA is the nonprofit, nonpartisan organization entrusted with preserving and sharing White House history.
Stephanie Zhan @stephzhan
33K Followers 2K Following Partner @Sequoia. Investing in AI, from company formation. 1st partner to @linear @mach_industries @middeskhq @reflection_ai @ricursiveai @skildai @tavus
ClaudeDevs @ClaudeDevs
647K Followers 2 Following Official updates for developers building with @ClaudeAI
Jeff Bezos @JeffBezos
9.6M Followers 602 Following Six thousand years ago, someone invented the plow, and we all got wealthier. A gentle reminder that all civilizational wealth is driven by invention.
Nicolechan @stark_nico99
16K Followers 831 Following 分享出海增长历程和干货 | Building YouMind | User Growth Leader | Ex-CMS Securities UX | Ex-Tencent Alumni Content Lead | Community & Content Expert @Youmind_AI
Peter Steinberger �... @steipete
572K Followers 2K Following Polyagentmorous ClawFather. Came back from retirement to mess with AI and help a lobster take over the world. @OpenClaw🦞 + @OpenAI
知识分享官 @knowledgefxg
204K Followers 2K Following 热爱知识,没事分享点有趣硬核的东西,包含英语学习,AI编程,科技软件,资源网站等等。来都来了,点点关注😘。个人一直在用的静态住宅IP:https://t.co/WEcFXoKWVp
Addy Osmani @addyosmani
408K Followers 3K Following Engineering & DevRel Leader, Prev: Director @GoogleCloud AI/Gemini ✨, @GoogleChrome • Author • Known: @ChromeDevTools, Agent Skills, Core Web Vitals, TodoMVC
The Kobeissi Letter @KobeissiLetter
2.4M Followers 591 Following Official X account for The Kobeissi Letter, an industry leading commentary on the global capital markets. Email us: [email protected]
Google Gemma @googlegemma
96K Followers 0 Following The official home of Google's Gemma. Lightweight, state-of-the-art open models by Google DeepMind, built on Gemini tech. What will you build? 🚀💻
Pirat_Nation 🔴 @Pirat_Nation
341K Followers 97 Following Tech & gaming analysis with the latest updates | TV & Films | alt: @piratnation | Official @skinscom Affiliate
Craig Weiss @craigweiss
52K Followers 7K Following founder & ceo at https://t.co/0UQAAqAfoR (yc w22) | @ycombinator | software engineer | early @scale_ai | prev: @google, @meta, @snap, @lyft, @nasa
Google AI Studio @GoogleAIStudio
191K Followers 2 Following The fastest path from prompt to prototype to production with Gemini
Omar Sanseviero @osanseviero
72K Followers 3K Following MTS at @GoogleDeepMind Building Gemini, Gemma, AI Studio and more. My views ex-Chief Llama Officer @huggingface 🇵🇪🇲🇽








![[Pop music liker][Elva Hsiao adorer]](https://pbs.twimg.com/profile_images/1992898669289652225/IwDmDNDy.jpg)
























