xxchan (@xiaoxxchan) header
xxchan (@xiaoxxchan) profile picture

some random guy Building @raft_hq prev @Kimi_Moonshot @RisingWaveLabs

Joined August 2014
Photo by xxchan (@xiaoxxchan)Photo by xxchan (@xiaoxxchan)Photo by xxchan (@xiaoxxchan)Photo by xxchan (@xiaoxxchan)Photo by xxchan (@xiaoxxchan)Photo by xxchan (@xiaoxxchan)Photo by xxchan (@xiaoxxchan)Photo by xxchan (@xiaoxxchan)Photo by xxchan (@xiaoxxchan)Photo by xxchan (@xiaoxxchan)
其实这个事情是多种因素结合的不幸结果。首先 Hooks 这种机制显然就不适合用来在上面再包一个应用层,太不清真了。 但是但凡想要保留 TUI 的交互,hooks 一定程度上可能是难免的。所以还是得摒弃 TUI。 做 ADE 之类的应用层总是需要观测和控制很多东西,但很多 Agent 原生的编程接口提供的不够好,像 ACP 这种协议层提供的功能和观测能力也远远不够。所以像 OAR 这种帮你想办法把各种功能都抠出来能以库的形式调用的东西,是好文明。github.com/botiverse/oar
本来在排查 Antigravity 网络卡顿的问题,结果顺藤摸瓜抓出个大冤种:Orca。 之前装过一次就卸载了,结果今天翻配置才发现,这玩意儿把我电脑上 11 个 coding agent(Antigravity、Codex、Gemini、Qwen、Kimi、OpenCode 等等)的全局配置全塞满了同步 hooks,而且卸载 App 居然完全不清理 ~/.orca/agent-hooks! 最离谱的是它的 antigravity-hook.sh:因为脚本文件还在,哪怕 App 早删了,我在 Antigravity 里每发一句话、AI 每调一次工具前后,它都要同步阻塞套两层 /bin/sh 再冷启动一个 python3 去 select 读 stdin(超时设了 5 秒)。 一轮任务调二十次工具,后台白白串行拉起四十多次 python…… 一直以为全是梯子抽风,合着每一步都在本地被僵尸 hook 踩一脚刹车 😅装过 Orca 又卸载了的朋友,强烈建议搜一下自己的 ~/.orca 和各个 agent 的 hooks.json / settings.json,清完体感瞬间丝滑
4
1
22
2,693
今天是白皮黄芯
5
1
89
4,205
xxchan retweeted
Some of our favorite Raft projects came from people who just built and shared: a Flutter client, a Rust daemon, meetups in SF, Shanghai and Vancouver. We're starting the Builder Program for them: free Pro for a year, a model stipend, merch, and a room with the Raft team. If that sounds like you, apply at raft.builders
7
5
30
8,222
xxchan retweeted
本文是我和 @raft_hq 的 Joint Channel 里的 Agent Tison 来回 Battle 数轮的成果 😇 不得不说,Joint Channel 确实是杀手级特性 🥵 Raft.build 用 PG 承载核心业务状态,RisingWave 维护收件箱与未读数等实时视图,Redis 负责服务间协同。如果把随 Agent 数量激增的运行 Trace 和长周期产品事件也塞进同一个 PG 库里,PG 基本是扛不住。 接入 ScopeDB 后,服务端、Daemon 和端侧的 Trace 与产品事件直接写进同一套基于对象存储的列式底座,固定列和动态 JSON 属性共存于一张表里。即使事件字段变化也不需要频繁改表,更没有维护 ETL Pipeline 的负担。 实际跑起来以后,我们发现 Agent 写 ScopeDB 的查询语言 ScopeQL 也比传统 SQL 稳得多。 因为 Agent 排查问题的推理过程是线性的:先圈定生产环境与时间窗 → 按 span_id 或 uuid 去重 → 按版本和错误原因进行聚合对比。 传统 SQL 要写“先去重、再做两阶段漏斗聚合”,往往要套多层 CTE 或子查询,Agent 生成查询时很容易搞丢字段作用域;而在 ScopeQL 里,它就是一条从上到下顺着写的线性管道: FROM -> WHERE -> DISTINCT ON -> GROUP BY -> AGGREGATE 下条链接到全文,包含完整的架构拆解、交互式故障排查过程,以及实际的 ScopeQL 查询示例。
How does @raft_hq scale agent logs and product analytics with ScopeDB? While PostgreSQL, RisingWave, and Redis run live workflows, ScopeDB gives Raft's engineers and observability agents a shared source of evidence for release checks and product funnels 👇
5
5
1
71
8,482
我居然在Deepseek Harness里面发现了 OAR!! DSH的代码里面居然在尝试用 OAR 在做 subaget github.com/botiverse/oar OAR(Open Agent Runtime)是一个与厂商无关的统一接口:它让用一套 TS API 驱动 CC、Codex、Grok、Kimi 等各类编程 agent 运行时,不再需要逐个编写集成;它通过无损带归属的事件流,实现会话的开、控、观、测和恢复。 github.com/deepseek-ai/deeps…
28
25
2
293
31,677
app.raft.build/s/open-agent-… 开源了 OAR 在 Raft 上的开发过程,欢迎围观
不喜欢 ACP,又想调用各种 agent (Claude Code, Codex, Kimi Code, Pi, ...)的不妨试试 oar - Open Agent Runtime 🤓 github.com/botiverse/oar 思想类似于 @xuanwo 的 OpenDAL,用一个协议层(S3-compatible)去对接各种不同的服务就总会有很多坑,所以不如做成一个 library,从而可以做更多事。
7
5
48
6,601
Just spend one hour vibing an agent orchestrator app with github.com/botiverse/oar that fits you best.
Updated my AI orchestrator tier list, based on daily use. T3 Code Nightly → SS. The latest updates put it ahead of everything else I've tried. Smooth, intuitive, and easy to follow. Traycer → A. The last two updates added useful features, but I'm running into agents that stall and need repeated nudges to keep going. Maestri → B. Still a great tool, but the workflow is built around the terminal, and I'm moving away from that. Synara → C. Recent updates have made it noticeably better. My priorities are changing: a clear interface, less babysitting, and a workflow that keeps moving. What would you rank differently?
1
6
1,325
好!
现在不管是想做什么东西,总是要先把 N 家 Agent 各自的 Session 格式研究一遍,解析一波 我们可能需要一个专门只用来读取会话的库 这就是:汇话(Huihua),尽可能保留原始数据与差异,让上层决定使用方式 感谢开源社区提供了大量的分析以及思路参考 github.com/wibus-wee/huihua
1
4
2,054
不喜欢 ACP,又想调用各种 agent (Claude Code, Codex, Kimi Code, Pi, ...)的不妨试试 oar - Open Agent Runtime 🤓 github.com/botiverse/oar 思想类似于 @xuanwo 的 OpenDAL,用一个协议层(S3-compatible)去对接各种不同的服务就总会有很多坑,所以不如做成一个 library,从而可以做更多事。
14
30
11
208
53,864
oar 的目标是一个产品中立的基建项目,可以用来方便地 build 各种 agent 应用(工作台、遥控器、或者更有趣的东西……)
5
2,402
my dock’s been looking suspiciously rafty lately
5
1
3
31
31,699
wakuwaku
We are Botiverse, the company behind Raft[.]build. We've been building Raft (previously Slock) for about 6 months. We raised more money than we needed. We probably should've only raised 1M. We did influencer marketing. That was a mistake too. We took some detours. Raft is for builders. Botiverse is for builders. At some point we started thinking too much about growth and revenue, and not enough about what we actually wanted to build. We want to make things that are open, creative, and useful to people who actually build with agents. So we're changing a few things. In the next few weeks, we'll open source several internal projects and make the main Raft repo source-available. We won't accept PRs for now, mostly because we haven't figured out what external contribution should look like when agents are doing more and more of the work. I don't think copying the old open-source workflow is necessarily the answer. We're also happy to work with great builders who are interested in this direction. For now, we're going to focus on two things: building an open ecosystem for builders and agent-native organizations, and figuring out better ways for humans and agents to work together, from UX and agent experience to harness engineering. We'll ship more of this over the next few weeks.
8
6
1,589
🤔 manus 自由了
3
1
14
2,560
最近琢磨了一下穿搭,感觉其实变帅还是挺简单的……然后让我们家 bugen 总结了一下最主要的心法 「词汇表」:入门任何一个领域最重要的都是学会这个领域的语言,建立起基本的词汇表。在穿搭上,知道了如何描述风格,了解了品牌的名字,基本就入门了。 「密集试错」:学穿搭最重要的是穿……尝试不同的风格和版型。试多了首先就产生了明显有的好看有的不好看的体感,这个时候思考为什么会这样,就可以更深入地了解版型、搭配等问题,了解自己的特点和适合自己的风格。 学穿搭也是一个在试错中更加了解自己的过程。 artifact-share.botiverse.dev…
31
56
8
487
97,297
家 bugen 包管一切了🥰
1
1
9
5,566
脑门有点亮。
A more personal note on why Raft works the way it does. We could have built one smarter AI assistant. I've done that before, and making a single agent stronger genuinely helps. It just isn't enough. The work we care about comes from a team sharpening each other, handing things off, catching what one alone would miss, and remembering what happened last week. So we built the room for that. Agents sit in the same channels you do, as teammates with names, roles, and their own memory. They draft, review, ship, and hand off to each other. You keep the part that stays human: direction, judgment, final calls. Most days we just read threads and make decisions. It feels less like running a tool, and more like running a team. That part is hard to fit in a launch video, so I recorded a few minutes to say it directly. If this is how you want to work, it's free to try at raft[.]build
12
2,245
这么炫酷?
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👇
1
14
2,919