RayHe

Tech Investor @ZhenFund

[RayHe @rayhe99]

🚀🚀Raft

[stdrc @istdrc]

Hi, I'm RC. I built Kimi CLI at Moonshot last year, and back in 2015, bots that lived in group chats. For the past four months, I've been building Raft in public.

Today I'm launching Raft 1.0.

Right now, working with agents means juggling terminals, sessions, and skills. The more you run, the more you end up holding it all together yourself, and the easier it is to lose the thread.

Raft puts your agents in team mode: one workspace where working with agents feels like messaging your team. The work keeps moving, and you stay at the wheel.

Meet my Raft agent team👇

[RayHe @rayhe99]

@daimajia @raft_hq

[stdrc @istdrc]

We just renamed Slock to Raft today.

The name fits what we’re building much better: a shared foundation where many agents can coordinate, carry context, and move work forward together.

There’s also a quiet nod to the Raft consensus protocol — distributed actors, shared state, reliable progress.

Same vibe. Sharper metaphor.

[Ryo Lu @ryolu_]

from apps to material

software used to be something you opened

an app was a room with walls: calendar here, notes there, music there, work there. each one had its own logic, buttons, its own little kingdom. the user moved between kingdoms, carrying context in their head

but ai starts to break the walls

software becomes less like a destination and more like material. something you shape, combine, stretch, ask, remix, and leave behind as traces. a document can become an app. a conversation can become a workflow. a song can become a memory. a task can become an agent. the boundary between using and making gets blurry

the old model was: choose the right tool for each task

the new model is: express the shape of the thing you want, then refine it with the system you built

this changes the role of the interface. ui is no longer only fixed views for fixed functions. it becomes a surface where intent turns into structure. the best interfaces will feel less like menus and more like clay – responsive, persistent, inspectable, and alive

apps won’t disappear. rooms are still useful. but the deeper shift is that software stops being a set of sealed containers and becomes a medium people can think through

like paper, but executable
like language, but spatial
like memory, but programmable

software stops being something only programmers make

it becomes material anyone can shape

[kabikabi @jakevin7]

OpenCLI v1.8.0 is finally out 🎉

Stayed up till 4am — finally landed the 1.7 → 1.8 stretch. Looking back, this run packed in a lot:

## Browser Agent Runtime

Rounded out the browser layer in one shot. The point: stop driving the browser with brittle CSS selectors, and move to "accessibility tree + semantic locators + CDP-primary input" — agent-native end to end. CDP input + AX snapshots + semantic locators (--role / --name / --label / --testid) + hover / focus / dblclick / check / upload / drag / wait download / annotated screenshot.

opencli browser <session> click 5 — grab a ref and click. No more selector guessing. Custom dropdowns on Radix / shadcn / Material UI that used to silently fail to open now click reliably.

## New sites / new surfaces

weread-official — WeRead's official Agent Gateway, pure HTTP + Bearer key, coexists with the cookie-based weread
12306 — full read (trains / prices / my orders / passengers)
xianyu — Xianyu inbox / messages / reply, you can talk back now
suno — music generation
linkedin — Sales Navigator + people-search + messaging / safe-send / thread-snapshot all wired up
linkedin-learning / rednote / booking / ctrip hotels + flights / DuckDuckGo / Brave / Yahoo search

## Twitter, kept polishing

list-create / device-follow / quoted_tweet / card binding_values / bio / UserMedia cursor pagination — all filled in
write-action symmetry: unlike / retweet / unretweet / quote
bookmarks / bookmark-folder / list-tweets now carry media

## Reddit / Zhihu, deeper read coverage

reddit: subscribed / whoami / home / subreddit-info / --expand-more (load more comment branches) / listing exposes post_hint+url+preview+gallery
zhihu: answer-detail / answer-comments / answer pagination

(important)

## Reliability & security

Download path-traversal fix — remote-controlled fields (e.g. video titles used as filename) can't ../ escape the output dir anymore
Page.goto stale-identity self-recovery; CDP -32000 is now retryable
undici 8.x silently bumped the engines floor to Node ≥22 — pinned back to 6.x to keep the Node 20 promise alive
youtube transcript cross-video bleed fixed (SPA watch→watch was returning the predecessor's captions)
ChatGPT web image generation back to green (now detects CSS background + canvas, not just <img>)
Typed-error sweep across Douyin / Jike / WeRead / Apple Podcasts / Reddit / Gitee / lesswrong / xhs / YouTube — silent-sentinel and silent-empty fallbacks replaced with EmptyResultError / AuthRequiredError. Agents no longer have to guess whether [] means "truly empty" or "site changed".

## Trace & Observation

Failed adapter runs now ship a trace artifact; summary.md is the entry point.
browser console / browser network --failed / --follow — agents can finally see what's happening inside the browser.

## README, 20% lighter

EN 410 → 326, ZH 455 → 371. Built-in Commands curated to 11 sites; CLI Hub reduced to a name enumeration. Easier to land on.

Goal hasn't changed: read & action infrastructure for AI agents.

https://t.co/4rmxL1wSJc

[RayHe @rayhe99]

@PSrobotics @rerundotio I'd recommend giving Odin1 (from Manifold Tech) a try instead haha. One sensor for all.

[RayHe @rayhe99]

@qiayuanliao 🤣

[RayHe @rayhe99]

lowkey shocked realizing a middle schooler made this

[清凤 @qqqqqf_]

做了个叫 Arkloop的东西,是个 Agent 客户端
开源,本地优先,简单优先
你可以把他想象成 claude desktop but open source, 并且带有自己的 taste
哦对了,另说一点,不是任何 agent sdk 套壳,也没有任何 base 任何 claudecode 行为,我一个人打磨了三个月

和市面上大部分产品不同
我花了很多时间在一个细节上 : 减少认知负担

一个例子 : 我平常只用一个模型聊天，那为什么每条消息前面都要告诉我用了什么模型？这是噪音
再举一个 : 开发者总是喜欢让 agent 的 tool use 完整的展示到前端,但是背后真正的用户体验逻辑是我需要知道 agent 在工作/在往哪个方向偏,所以我们并不需要如此详细的信息

设计哲学：认知负担,信息,价值导向,美学

换个话题
我一个人做 Arkloop 用了三个月,现在他能用了,但是它离完美很远
做产品很重要的一件事是…不要闭门造车,也就是我需要你们的真诚建议

引用来自 Arkloop readme 中的一句话
“我欢迎所有形式的贡献。即使你不是开发者，只是一个普通用户——如果你在使用中感到任何不舒服的地方，哪怕只是一点间距、一个颜色、一个很小很小的细节，或者是一个很大的方向，都可以直接开一个 issue。
我认真对待每一个体验细节，你的反馈会让所有人的体验变得更好。
如果你是开发者，Arkloop 的 Agent 核心、记忆系统、hook 机制都是开放的。你可以接自己的 provider、写自己的插件、甚至改掉你不喜欢的任何设计。”

这是我认为整个项目最精髓的一点,我希望看到你们的反馈,不管是细节还是方向
我在用 dify 的时候,我常常发现,一个特别小的间距问题,竟然在这么大的仓库里没人修
因此,我很重视这点理念

Arkloop 现在并不稳定，还有很多不完善的地方。希望大家多多包容,多提意见

另带一提,Arkloop 可以从 openclaw/hermes 导入配置
Github 仓库:https://t.co/usHZDiOERj
Arkloop 官网兼下载:https://t.co/ZAAursWEl6
关注我的推特:https://t.co/UCFcS7AmUm
加入 Arkloop 的 telegram 群组:https://t.co/M7vddLerd6

[RayHe @rayhe99]

@holegots 不合理啊 按这么说 所有互联网公司也都是中国移动中国电信的中转站？显然不是，token（流量）本身重要但又不唯一，而基于此deliver的价值才是利润的来源

[RayHe @rayhe99]

@arkuy99 举手🙋‍♂️

[Physical Intelligence @physical_int]

Our newest model, π0.7, has some interesting emergent capabilities: it can control a new robot to fold shirts for which we had no shirt folding data, figure out how to use an appliance with language-based coaching, and perform a wide range of dexterous tasks all in one model!

[Claude @claudeai]

Introducing Claude Opus 4.7, our most capable Opus model yet.

It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back.

You can hand off your hardest work with less supervision.

[RayHe @rayhe99]

@Yuancheng 哈哈哈哈下午刚聊一个

[stdrc @istdrc]

Hi, I’m RC. I previously built Kimi CLI at Moonshot AI.

Now I’m building Slock, an agent-human collaboration platform for modern builders and teams.

Today, we're shipping a ton of new features and improvements in Slock: search, thread inbox, saved messages, message permalinks, pinned chats, server join links, a more consistent color system, and many smaller upgrades.

More details in the thread below.

[RayHe @rayhe99]

@istdrc incredible

[RayHe @rayhe99]

@beginnersblog1 oh! thank you so much!

[RayHe @rayhe99]

@Ai_Vaidehi I want one! thx

[ARX @ARXrobotics]

AC 2
Anyone can contribute data at home

Ego+AC one+Pi*0.6

👀Uses only head cameras to complete the task.

💪The task is learned from human videos.

📱Data collection requires only a mobile phone.

📸Works across different cameras.

🦾Errors are corrected through human demonstrations.

✅The training process is simple and efficient.

🤗High task accuracy is achieved after training.

Thanks，
Tesla that video give us some ideas
Pi*0.6 awesome work！

[RayHe @rayhe99]

@JingxiangMo @gradientrobots What a pity that K-Scale has shut down—such a brilliant group of geeks gone! Can’t wait to see JX take his new company to even bigger heights.