一直没时间折腾龙虾（OpenClaw），今天硬挤出一小时装上了，玩了下可能并没有网上说得那么好，因为有的案例两年前我都实现了，现在看反倒有点遗憾。说到底，这东西并不神秘：只要你有一台能跑本地模型的机器/服务器，就算不懂代码的人，好好写写提示词，也完全有机会把类似的东西拼出来。
但得承认，龙虾的价值还是在于它把“能做什么”变得更可想象、更可复用，也给了大家一个能延展的空间和思路。再加上 Skills（技能），确实会让人产生一种“很神奇”的直观感受，特别是“记忆”。某种意义上，不是龙虾这个产品神，而是我们个人的想象力贫乏。Skills 这段时间的爆发式增长，又一下子把之前想都不敢想的落地场景给整出来了，就这点来说，这个产品还是厉害的。
最后必须泼点冷水：不建议普通人装。先不说配置、安装排错、安全性这些，光网络速度就能把人折腾的想死。真想把龙虾的上限拉满，人民币也是一个无底洞。就比如说你让豆包元宝这些总结下普通的文本还行，不上贵的模型充其量就是一个加速过程。
I’ve never really had time to tinker with “Lobster” (OpenClaw). Today I forced myself to carve out an hour to install it and play around a bit. Honestly, it may not be as good as people online claim—some of the example use cases are things I’d already implemented two years ago, so in that sense it actually felt a bit disappointing.
At the end of the day, there’s nothing mysterious about it: as long as you have a machine/server that can run local models, even people who don’t know how to code can absolutely cobble something similar together—if they’re willing to spend time writing solid prompts.
That said, I do have to admit OpenClaw’s real value is that it makes “what’s possible” more imaginable and more reusable. It also gives people an extendable space—both a framework and a way of thinking. Add Skills into the mix, and it really does create that “wow, this is magical” first impression—especially with “memory.” In a way, it’s not that the product itself is magical; it’s that our personal imagination is often too limited. The recent explosive growth of Skills has suddenly made a bunch of scenarios feel practical—scenarios that previously felt almost unthinkable. In that regard, the product is genuinely impressive.
One last cold splash of reality: I wouldn’t recommend that the average person install it. Even setting aside hardware requirements, installation/debugging, and security concerns, network speed alone can be maddening. And if you really want to push OpenClaw to its upper limit, your wallet becomes a bottomless pit. For example, tools like Doubao or Yuanbao can do a decent job summarizing ordinary text—but without paying for more expensive models, the best you’re getting is basically a faster workflow, nothing more.
