Two weeks after mynd shipped, I sat down with cbvivi (Wang Xiaoguang) to record an episode about his new app, mynd. In his own words, it is “a chat-style front end for writing Markdown files.” cbvivi was one of China’s earliest vloggers, and a veteran podcast host. Further back, he was an editor at a magazine, and ran a long-running iPhone homescreen observation column on his personal blog. mynd is not the product of a developer who decided to make an app. It is what happens when someone who has spent most of his life watching how other people use software finally decides to build one himself.
Ten Years of Quick Capture
cbvivi told me that from the moment he started using a smartphone, one of his most constant use cases has been capturing ideas quickly. For years he was a heavy Drafts user, and later moved to Obsidian. The plugin that made the biggest impression on him there was Daily Notes. Every day it auto-generates a dated document. It can be empty, or packed full, but there is always a new one waiting the next day. By contrast, every note in Drafts is its own document, and once you have a few thousand of them, searching or organizing becomes almost impossible.
The Action Button on the iPhone 15 Pro completed the loop. He bound a shortcut called “Genius Idea” to it, so one press wrote straight into his Obsidian vault. He built an entire workflow around this setup, and came to depend on it deeply.
The trouble was that none of it was something he could recommend to a friend or a family member. Obsidian plus plugins plus Shortcuts plus a Markdown vault is a setup only a specific kind of person is willing to spend a weekend configuring, and Obsidian’s tutorial ecosystem has exploded again in the past few months, which only raises the bar. Earlier this year, he started pointing AI tools at the notes he had accumulated. We had talked a few times before about how Claude Code had become genuinely usable at processing files across folders, and he fed his vault into it. Somewhere in that process, he began seriously wondering whether he could build a product that did not require that entire stack underneath, something people could simply pick up and use.
The Moment He Decided
The moment he actually decided to build it was when he thought of combining Daily Notes with a chat interface. “Once I had that, everything just fell into place. The idea felt brilliant,” he said.
Looking back, he says two things went his way. The first was the blind confidence that only comes from not knowing anything. He had done no competitive research at all. If he had, he would have discovered that chat-with-yourself note apps are everywhere. China, Japan, and the US all have dozens of them, and some have far cleverer interactions. Japan has an app called 一人会議 (Solo Meeting). Another chat app splits the send button in two, so pressing the left one sends a message on the left, and the right one sends on the right. If he had seen any of that up front, his personality would have convinced him the idea was not that special, and he probably would not have built it.
The second piece of luck was that Claude implemented a gesture he did not expect to work. He wanted to let users swipe a bubble from one side to the other. When he put in the request he was nervous, since an interface-level change like that seemed like the kind of thing you could only hope to get a rough implementation of. What Claude produced had natural elasticity built in, and was arguably better than the version that eventually shipped. “Wow, I thought, being able to get to this level completely blows past what I imagined was possible.” That was the moment he felt the product was real.
The left-and-right bubbles in mynd are simpler than most people assume. Messages from the current session sit on the right. Once you put the app into the background long enough for the process to be reclaimed and then reopen it, those messages move to the left. They have become history, in a sense. Messages on the left can be freely moved, and dragged back to the right, because history can be rearranged. Messages on the right cannot, because the conversation is still in progress.
Don’t Lose a Note
cbvivi and Claude defined mynd’s highest-priority rule together: do not lose data. He would rather the app crash repeatedly than drop a single note. Every note lives in three places at once: iCloud, local storage, and memory.
Since the beta, two users have confirmed losing notes. One was writing when a test build froze before any of the protections could kick in. The other ran into an iCloud multi-device conflict where one device overwrote another.
Bug reports make him anxious, in large part because he cannot verify the code with his own eyes. As a humanities student with zero technical background, he spent two months building mynd inside the Claude desktop app, learning and embarrassing himself as he went. In his third week, Claude suggested, “would you like to roll back?” He replied, “what is a rollback? Sure, roll it back.” That cost him an entire week of work. That was the day he learned what a commit was.
Tools, and the Price of Cutting Costs
In his first week, cbvivi burned through both his 5-hour and weekly Claude limits. In the second week, he bought Antigravity as a sidekick. After running them in parallel for a while, he decided Gemini was noticeably weaker on real product work, with a few changes he genuinely could not live with. He later used it for smaller jobs like building websites, and unsubscribed when the term ran out.
He also installed OpenClaw, but never really got it off the ground. He tried pointing it at customer support tasks and at traffic monitoring, but could not get even the basics to run. It was supposed to report in every hour, and stopped after three. At the time he did not have the bandwidth to debug, so he set it aside. Looking back later, he realized that at mynd’s stage, most of the work he is actually spending his time on is not the kind of thing you can hand off to an agent.
He opened the command-line version of Claude Code exactly once, saw the terminal, and went back to the desktop app. At this point I don’t think learning the CLI on its own is worth it either. The desktop and CLI are absorbing each other’s features, and the best parts of the CLI are being pulled into the desktop one at a time.
After launch, to save money, he downgraded from Max to Pro, and a week later regretted it and upgraded back. Even now, he does not come close to exhausting his Max quota. On the last reset, he still had more than half of it remaining. Which is exactly what I keep telling the rest of our team: for an indie developer, the single most important thing is to use the strongest model with the highest thinking budget enabled. Two hundred dollars a month is some of the best money I have ever spent.
Letting the Product Fade Into the Background
From day one, cbvivi placed mynd firmly on a “low presence, no design” path, leaning heavily on native iOS components. Part of the reason was practical, since pulling every piece of UI by hand would have taken far longer than he had. Part of it was that he believed the product should feel this way in the first place.
The change that went furthest in the product was “Document Mode.” It used to be one of mynd’s core concepts: users would switch freely between a chat view and a document view. It was dropped fairly quickly. In the latest version of the architecture, he is considering moving the underlying data source off Markdown. He has observed that users care less about the file format than he expected. What matters more is front-end fluidity and the kind of atmosphere that makes a person want to keep writing. He is learning to follow the signals his users actually send.
mynd has no built-in AI. Markdown files sit in iCloud Drive; power users can open the folder as an Obsidian vault, or hand it to any AI tool for analysis. He is aware that most users today expect a notes app to have AI baked in by default, which makes mynd mildly countercurrent. He believes the absence of it is what lets people feel the value of actually writing and thinking by hand.
What Comes Next Is Content, Not Features
With two weeks of emergency patches behind him, cbvivi’s next move is content. He just finished a text and image tutorial, and has a few videos planned: using mynd on a Mac through Shortcuts (there is no Mac version yet), running an AI tool across a folder of notes, and logging daily meals in mynd before asking a model to plan next week’s menu from the log.
He thinks the product can feel “a bit like making a magazine used to feel.” mynd is a deeply personal medium. The product itself is extremely private, with no networked sharing features. But people will end up sharing pieces of what happens inside it.
What can people do with it? Will a loose, naturally forming community grow around it? His job is not to keep adding features, but to give people more reasons to return to the medium. He wants mynd to become a quiet substrate that other things can grow on top of.
It is also interesting to watch how mynd moves through its users’ hands. Everyone is using it for something different. One person’s posted screenshot shows stocks they are tracking. Another is keeping a travel diary. Someone else is archiving voice memo transcripts. The app itself prescribes none of this. There are no tutorials, and no onboarding. What you say in this conversation is between you and yourself.
Mentioned in this episode
- mynd’s App Store page and website
- cbvivi’s Weibo, Bilibili, and Twitter
- Drafts, Obsidian, Numi, Antinote, mymind
- Warp, Antigravity, Remotion, Screen Studio
- Poche: our team’s link-capture tool
- Fiche: chat with your Drafts notes
- @mal_shaik on what leaked source code reveals: CLAUDE.md is re-injected on every turn
上线两周之后，我和 cbvivi（王晓光）坐下来录了一期节目聊他的新产品 mynd。按他自己的说法，这个产品就是「用一个聊天式的前端界面去写 Markdown 文件」。cbvivi 是中国最早的 vlogger 之一，也是资深播客主播。再往前回溯，他在一家杂志做过编辑，也在自己的博客上连载过很长一段时间的 iPhone homescreen 观察栏目。mynd 不是一个想写 App 的开发者做出来的产品，而是一个花了大半辈子时间观察别人怎么用软件的人，最后决定亲手去做一个的结果。
十年快速记录
cbvivi 跟我说，从用上智能手机开始，他最主要的使用场景之一就是快速记录想法。之前很多年他都是 Drafts 的重度用户，后来转到了 Obsidian，一个对他冲击最大的插件是 Daily Notes。Daily Notes 每天自动生成一个带日期的文档，它可以是空的，也可以写得很满，但第二天永远有一个新的。相比之下，Drafts 里每条笔记都是独立的文档，累积到几千条之后再去检索和整理几乎不可能。
iPhone 15 Pro 的 Action Button 完成了这套流程。他在那个键上绑定了一个名为「天才的想法」的快捷指令，按一下就能直接写入 Obsidian 笔记系统。他围绕这套系统建起了一整个工作流，也变得非常依赖它。
问题是，这套系统没办法推荐给朋友或者家人。Obsidian 加插件加快捷指令加 Markdown vault，这是一个只有特定人群才愿意花周末去搞的东西，而且 Obsidian 的教程生态在过去几个月又爆发了一轮，门槛只会更高。今年年初，他开始把 AI 工具指向自己积累的笔记。我们之前聊过几次 Claude Code 在跨文件夹处理文件这件事上已经变得相当可用，他把笔记库喂给它。那个过程中他开始认真想：能不能做一个不需要那一整套底层，让人拿上手就能用的产品？
决定性的那一刻
真正决定要做的那一刻，是他想到可以把 Daily Notes 和聊天界面结合起来。「有了这点以后，就越想越顺。这个想法太天才了」，他说。
他说回头看，有两件事运气很好。一件是那种什么都不懂时才会有的盲目自信。他根本没做过所谓竞品调研。如果做了，就会发现自己跟自己聊天的笔记产品到处都是——中国、日本、美国都有一大堆，有些交互还更聪明得多。日本有一个 App 叫「一人会议」。还有一款聊天产品把发送键做成两个，按左边的发左边消息，按右边的发右边。如果一开始就看到这些，他的性格会让他觉得「想到这个好像也没什么了不起的」，可能就不做了。
第二个幸运是 Claude 实现了一个他不抱期望的手势。他希望用手势把气泡划到另一边。提请求的时候非常忐忑，觉得这个界面层面的变化，能得到一个粗糙的实现就不错了。Claude 做出来的版本直接带了弹性，甚至比后来上线这一版更好。「哇，我当时觉得，能做到这个程度完全超越我的想象。」那一刻他觉得，这个产品真的可以做。
mynd 里的左右气泡比大家想象的简单得多。当前会话里发出的消息在右边。等你把 App 放到后台足够久、进程被回收之后再打开，消息就跑到左边去，它们某种意义上已经成了历史。左边的消息可以自由移动，也可以拖回右边，因为历史是可以被操控的；右边的消息不行，因为会话还在继续。
不能丢笔记
cbvivi 跟 Claude 明确定义了 mynd 的最高原则，不能丢数据。他宁可 App 一直闪退也不能丢一条笔记。所以每条笔记都同时存在三个地方：iCloud、本地存储，以及内存。
从内测开始，明确丢过笔记的用户有两位。一位是写完没发出去时测试版卡死，所有保护措施都没来得及生效，另一位则遇到了 iCloud 多设备互相覆盖的问题。
他遇到 bug 反馈会很紧张，一个重要原因是他无法肉眼验证代码。作为一个毫无技术基础的文科生，他用了两个月时间，在 Claude 桌面版里把 mynd 做出来，边学边丢人。做到第三周的时候，Claude 提议「要不要回滚」，他说「什么是回滚？好吧，回滚。」于是丢了整整一周的工作成果。那一天，他知道了什么是 commit。
工具，和省钱的代价
第一周，cbvivi 很快把 Claude 的 5 小时会话和周会话都用光了。第二周，他买了 Antigravity 当辅助，一起用了一段时间之后，他觉得 Gemini 在真实产品上的表现明显要差一些，有几处改动让他实在受不了。后来主要拿它做做网站之类的小活，时间一到就退订了。
他也装过 OpenClaw，但一直没真正用起来。他试过让它跑客服、监控网站访问数据，但是连最基础的都跑不通——本来应该每小时汇报一次，三次之后就停了。那段时间他没有余力去调试这些问题，就放下了。后来有时间再看，他意识到在 mynd 这个阶段，他真正花时间的事情大多不太能交给一个代理去做。
命令行的 Claude Code 他只打开过一次，看见终端就回到了桌面 App。在当前这个节点我也没觉得命令行值得单独学：桌面版和命令行正在互相吸收，命令行上的好东西已经被一样样搬到桌面里。
上线之后为了省钱，他从 Max 降到了 Pro，一周后后悔了又升回来。哪怕现在 Max 的额度根本用不完。重置前的最后半天，他的额度还剩一大半。也是我对我们团队其他人一直在说的：对一个独立开发者来说，最重要的事情就是用最强的模型、开最高的思考预算。一个月 200 美金，是我这辈子花得最值的钱之一。
让产品退到后面
从第一天开始，cbvivi 就把 mynd 定在「低存在感、不设计」的路线上，大量使用 iOS 原生组件。一部分是现实的考虑——每一块 UI 都要自己抠的话，时间根本不够；另一部分是他本来就觉得它应该是这种感觉。
实际产品走得最远的一个改动是「文档模式」。那曾经是 mynd 的产品概念之一：用户可以在聊天视图和文档视图之间自由切换。后来很快被放弃了。在最新版本的架构里，他正在考虑把底层数据源从 Markdown 换掉。他观察到用户没有那么重视文件格式，更重要的是前端流畅性和让人想多写多记的产品氛围。他在学习跟着用户真正的信号走。
mynd 没有内置 AI 功能，Markdown 文件放在 iCloud Drive 里，想深度玩的用户可以把它当成 Obsidian vault 打开，也可以把它扔给任何 AI 工具去做分析。他知道现在大部分用户对笔记 App 的预期是 AI 功能默认在里面，所以这款产品稍微有点反潮流。但他觉得这能让用户感受到亲手书写和思考的价值。
接下来是内容，不是功能
两周的紧急补丁期过去后，cbvivi 的下一步计划是做内容。他刚写完一篇图文教程，又计划了几个视频：在 Mac 上通过快捷指令用 mynd（目前还没有 Mac 版），用 AI 工具分析笔记文件夹，或者用 mynd 记日常饮食、然后让模型根据记录排下周菜谱。
他觉得这个产品也可以「有点像以前做杂志的感觉」：mynd 是一个个人化的载体。产品本身极其私人，也没有联网分享功能。但大家会分享其中一部分。
用户可以用它做什么？围绕它会不会形成松散又自然的外部社区？他的工作不是一直加功能，而是给大家更多的理由回到这个载体。他希望 mynd 能变成安静的底层，让别的东西可以长在上面。
观察 mynd 如何在用户手里流动这件事也很有意思，每个人都在用它做不一样的事情。有人贴出的截图里是他在观察的股票走势。有人把它当成旅行日记。有人用它做语音备忘录的转写归档。App 本身不规定这些。没有教程，不需要 onboarding，在这场对话里说什么，是你跟你自己的事。