YuShen

Building local-first AI tools, solo & in public. Runs on your machine, not the cloud. Follow for build logs + launches ↓

[YuShen @yushen686]

Curious if people would want this.

Someone at the Codex hackathon built an MCP plugin that connects ChatGPT with Codex.

The problem is very specific: GPT-5.5 Pro is currently much stronger at planning, research, and breaking down tasks, but you can’t use it directly inside Codex. Codex is great at execution, but its planning ability isn’t quite at the same level.

Their approach was to use MCP to expose the local dev environment to ChatGPT, so GPT-5.5 Pro handles the planning and task breakdown, while Codex handles the implementation.

In other words: let each AI do what it’s best at. One thinks, the other builds.

There’s also an unexpected bonus: ChatGPT and Codex have separate rate limits. With the same account, connecting them this way can feel like effectively doubling your available quota.

I think this is a pretty interesting direction. A lot of people are still asking, “Which AI tool should I use?” But the real answer might be: use all of them, and make them work together.

The walls between AI tools are starting to come down through protocols like MCP. In the future, it may not be about choosing the single strongest AI, but about building the right AI stack for your workflow.

Would you use something like this?

[Vincent | 信号＞噪音 @VincentLogic]

有人在 Codex 黑客松上做了一个 MCP 插件，把 ChatGPT 和 Codex 打通了

解决的问题很具体：GPT 5.5 Pro 是目前规划和调研最强的模型，但 Codex 里用不了。Codex 擅长执行，但规划能力不如 5.5 Pro

他的做法是用 MCP 把本地开发环境暴露给 ChatGPT，让 5.5 Pro 负责规划和拆解任务，然后把执行部分交给 Codex 来干

等于让两个 AI 各干自己最擅长的事：一个想，一个做

还有一个意外收获：ChatGPT 和 Codex 的速率限制是分开算的。同一个账号，接上这个之后相当于额度翻倍

这个思路其实挺值得借鉴的。很多人纠结"到底用哪个 AI 工具"，但真正的答案可能是：都用，让它们配合

AI 工具之间的壁垒正在被 MCP 这种协议打通，以后可能不是选一个最强的，而是搭一个最��适的组合

[YuShen @yushen686]

some update,maybe cloud design is bestway

[YuShen @yushen686]

I’m trying to recreate this beautiful browser game by @probiex007, learning WebGL and Three.js along the way.

Still rough, but it’s already playable in the browser. Huge respect for how
@threejs @webgl_webgpu

[kk@study @kk_design_dev]

めっちゃ流行った"あのサイト"を参考にThree.jsで作ってみた！ こういったイラストスタイルでいけると思ってなかったから見た時は衝撃だった

色の面に対して、ちょい掠れがちな境界線をつけるだけで、途端にアナログ感でるの面白い

GLBたくさん使う必要があると思い込んでたけど、全然必要なくてThree.jsすごい

エンジン: Three.js
人型モデル: Vroid Studio
音: ComfyUI Stable Audio3・Ace Step 1.5

[YuShen @yushen686]

A way to use gpt to replace Lovart👀

[Cell 细胞 @cellinlab]

原作者插件还没开源，但是其实直接
在 内置浏览器 打开 https://t.co/zXwLpIOqLi
类似的画布工具就行，不用插件也能实现

本质是 快捷地给 Codex 喂入标注图

[YuShen @yushen686]

@threejs @probiex007 @webgl_webgpu 😂

[YuShen @yushen686]

I’m trying to recreate this beautiful browser game by @probiex007, learning WebGL and Three.js along the way.

Still rough, but it’s already playable in the browser. Huge respect for how
@threejs @webgl_webgpu

[YuShen @yushen686]

so cut #cat

[YuShen @yushen686]

@litsadSlitude ✌️

[YuShen @yushen686]

最近好多人问 Plasma One U卡怎么开，最大的坎就是邀请码——官方邀请制，码还消耗极快。

与其在 X 一条条搜、手动刷新，我做了个工具帮你省事：

✅ 自动从 X 抓最新邀请码
✅ 按新鲜度排序，越靠前越可能有名额
✅ 一键复制，过期的自动标灰

#PlasmaOne #U卡 #数字游民

[YuShen @yushen686]

@ybwdaisy 哈哈

[YuShen @yushen686]

拿走不谢 👉 https://t.co/e7DwezGCCR

[YuShen @yushen686]

https://t.co/I3d1aDFK6m

[YuShen @yushen686]

Spent the last few months building this — shipped in0 v1.0.0 today 🚀

A native macOS terminal multiplexer (workspaces / tabs / splits), but the part I actually care about: live AI agent status. See at a glance which split your Claude/Codex agent is in, and whether it's thinking, waiting on you, or done — right from the sidebar.

No more window-hopping to guess if your agent is stuck.

Built clean-room on libghostty + SwiftUI/AppKit.

[YuShen @yushen686]

https://t.co/QRSKVRKkAg

[YuShen @yushen686]

I built voiceBox because I was tired of uploading my voice to someone else's servers just to clone it or narrate a long doc.

So I made it run entirely on your Mac. Apple MLX + Qwen3-TTS/ASR, native on Apple Silicon. Drop in a reference clip, generate speech in that voice, narrate long text — and nothing ever leaves your machine. After the first model download, it works fully offline.

No cloud. No API keys. No accounts. Your voice stays yours.

macOS 15+, Apple Silicon. Free download:
https://t.co/L1LVWVKygd

[YuShen @yushen686]

这个开源数字人项目，真的有点像 AI 分身的雏形

[YuShen @yushen686]

Google（Chrome开发团队）官方出品。

能做什么呢👇
・AI能打开真正的Chrome进行操作·检查
・自己调查显示错乱·延迟原因·bug
・可从Claude/Cursor/Copilot/Antigravity使用

“让AI建站”的下一步是“让AI直接调试网站”。
Web开发的工作量将大幅减少。

[オープンソー���研究所 @opensourcelab9]

【衝撃】
AIに「Chromeブラウザを丸ごと操作・点検させる」公式ツールが出ました🔥

Google(Chrome開発チーム)公式です。

何ができるかというと👇
・AIが本物のChromeを開いて操作・検査できる
・表示崩れ・遅い原因・バグを自分で調べる
・Claude/Cursor/Copilot/Antigravityから使える

「AIにサイト作って」の次は「AIにサイトを直接デバッグさせる」。
Web開発の手数が一気に減ります。

https://t.co/Ki9UGdjdEN

[YuShen @yushen686]

把"美股喊单王" Serenity 的来龙去脉做成了一张图 📌

匿名前 AI 研究员，年内回报 +4502%，一套"瓶颈点理论"专挑供应链卡脖子的小盘股——$AXTI
单只 +1000%，还提前两个月精准预测了树莓派营收。

一图看懂：他是谁、凭什么火、怎么选股 👇

[YuShen @yushen686]

spent part of yesterday rewriting a PRD because /goal changes the unit of work.

before: break the feature into 12 tasks and hope the agent doesn’t drift.
after: define the finish line, the guardrails, and the 3 things that must never change. the agent can chew through the middle on its own.

that feels much closer to how i already manage a good engineer: not micromanaging steps, but being strict about acceptance. codex is starting to look more like that than a chatty autocomplete.

[YuShen @yushen686]

karpathy 去 anthropic 这件事，我最在意的不是“又一家赢麻了”，而是他自己把优先级讲得很清楚：先回到 R&D，再回教育。对 PM 来说，这种人才流动比 launch 更像一个 lab 的真实产品路线图：它到底在押研究、工具，还是未来 18 个月能落地的东西。

[Andrej Karpathy @karpathy]

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

[YuShen @yushen686]

mckay 这句我觉得很准：选 coding agent 不是在选 benchmark，更像在选你愿不愿意换一整套工作流。

模型提升当然重要，但真卡团队的通常是迁移成本：习惯、上下文、评审方式、和那点“我知道它大概会怎么犯错”的肌肉记忆。codex 如果真能让工程师少切几次工具，很多时候就已经赢一半了。

[Mckay Wrigley @mckaywrigley]

some random ai thoughts:

- for code, i went from 80/20 claude/gpt to 80/20 gpt/claude in <3 months. surprised by this tbh, and interested to see where the split is at in another 3mo.
- claude still mogs gpt for non-coding agent stuff. codex feels like an engineer (which is great for coding!), whereas claude still feels like a general purpose coworker. gpt still lacks that coworker magic
- i’m pretty meh on opus 4.7. my experience hasn’t been *bad*, but it certainly hasn’t been good. sideways if anything.
- anthropic has got to figure out the compute thing. you can feel it as a user. vibes are all out of whack bc of it. my opinions above are all likely downstream of this. it’s an issue.
- anthropic labs continues to be the goat of ai product. claude design is another hit. it’s fantastic. idk why it’s not talked about more? a+
- updated claude code app is great. i finally switched out of the terminal for it. very well done.
- how are people STILL sleeping on the claude agent sdk? i feel like i’m going insane.
- gpt 5.5 is incredible. the level to which i trust it for engineering is amazing. if i could only have one model rn, it would be this one just bc of strong need for the coding use case.
- codex team is killing it. app has been the gold standard since 5.3 release (buuut i credit conductor team for the ui innovation that everyone is using now). though i could do with a little less passive aggressive shots at ant from the codex team. TARS, dial up class by 30%. it’s a long race guys haha
- i uninstalled cursor this month and am now back to vs code for my ide. composer just can’t hang with claude/gpt, and the product feels a bit all over the place. pretty stoked about the xai thing though, because their team is absolutely stacked and i’m excited to see what they might be able to do with that compute. codex and claude code are t1, cursor is t2. i would love if this deal got xai/cursor to t1 for a real trio there.
- gemini…? seems like this is 2-3 models now where the model seems like a great release and then nobody ever uses it? i’m bullish google/deepmind but weird it hasn’t translated to product use in any form. kinda disappointed still
- no open source models have hit the opus 4.5 level. was hopeful the new deepseek would get there, but nope. good oss agents will have to wait a few more months it would seem…