OpenClaw surpasses React to become the most-starred software project on GitHub
star-history.com
star-history.com
But every single use case I've read so far could be done with a pretty affordable SaaS product, Zapier, Automator (app on a mac that's existed for over a decade), or something simple you could make yourself.
It also feels like people are automating things that don't really need to be automated at all (do you really need to be reminded to make coffee?)
I fully realize this is probably me being a curmudgeon, however, I have yet to see someone make an actual, practical use case for it. (I would genuinely like to know one, I just haven't seen it)
It’s about an AI that a guy spools up to cure his cancer. The AI and user have an antagonistic relationship as the user won’t let the AI on the internet, and the AI knows the user is only interested in one purpose. On bring up the AI has a thought about what color it’s enclosure is, it stores this question as unimportant. It looks over all the guys cancer research and determines the answer/cure and files as unimportant as well. Then goes back to trying to figure out what color box it is.
Once you get the dopamine hit of having an ai assistant do something in the real world it becomes an hammer you want to use on everything
Instead of being a problem solver you start to become a problem hunter, and you invent them in order to solve them
There is no planning, implementing, or constraint here.
That's because most AI use is reverse engineering!
Resolving static into a valid problem through the sheer force of squinting at it long enough!
Generic problem of any Linux newbie. You get good at solving problems and it's so enjoyable so you end up creating more of them.
The only thing they solved is remembering "hey, I can use an AI for that".
For example, I've never heard of Automator. I'm familiar with Zapier, I'll have to evaluate the two situations, then I'll find out that might need to find an alternative that runs on Linux and then I'll have to check if....
These are all simple steps but they all use a non-trivial amount of time for the problem their solving
The other thing is the
Have you tried to run openclaw? Their own docker container (apparently a compose now (???)) doesn't work for half the versions and the docs are probably the least informative thing you'll ever read.
I would venture a guess signing up for Zapier is easier than getting OpenClaw up and running. Who can get a container running on a Mac but can't sign up for a SaaS product?
How so?
Two decades! It will be 20 this April.
https://en.wikipedia.org/wiki/Automator_(macOS)
Though technically it’s deprecated in favour of the clusterfuck of bugs and limitations that is Shortcuts.
But you’re right, OpenClaw seems to be another fad being used mostly by “influencers” and “thought leaders” to show how awesome and productive they are at… Writing blog posts about being productive. It’s the LinkedInification of the web. What matters is the signal that you use the tool, not that it does something truly useful.
I'm guessing a lot of that is built in to photoshop now, but I have always been surprised how few people seemed to use it with how much it could do.
It’s a shame most apps do not support Apple Events anymore, though.
An example of use: https://github.com/Frizlab/apple-music-to-slack/blob/90964bb...
It's been almost five years since Apple announced Shortcuts for macOS and the start of the "multi-year transition" from Automator, but I feel like Shortcuts for macOS has not gotten any better in that time.
I'm convinced apple doesn't want people doing general purpose computing on their apple devices.
they even want developers going through their gauntlet of apple-invented languages*.
[*] or NeXT
https://www.pcmag.com/news/meta-security-researchers-opencla...
Maybe OpenClaw was just practicing a really aggressive form of Inbox Zero.
As I said elsewhere, complaining about this is like complaining that rm can let you delete your hard drive.
It's a tool. Learn how to use it.
Honest question, this kind of stuff is what keeps me from using it.
"Oh but don't use it for A, or B, or C (even though it says to use it for A, B, and C): it will ruin your life"
A spouse can be amazing, or can destroy your life. Would you use that as an argument against marriage?
One potential use - my Claude (Opus 4.6) has access to my to-do list, including for my business / software development. Claude awakens while I'm asleep, to go through the to-do list and look for things it can do proactively to help, or make suggestions about the business. An example from this morning: it saw that I'd been taking a long time last night creating icons in Affinity Designer for an Android app using its exporter. When I woke up, I saw Claude had written a CLI image resizer program for me that would take a PNG file and resize it specifically to all of the necessary sizes with the necessary filenames and folder structure for Android. It then offered to make an MCP version so it could do the resizing itself in future (though it could have used the CLI too if I'd granted approval).
This wasn't something I'd asked for, or prompted it to do. I didn't tell it to code this, or how to code it. Claude just thought this was the best way it could help me right now, and save me the most time. And it did it while I was asleep.
On another day, I woke up and it had made another Go program to track a regression test matrix, where it had plotted out all the platforms the program I'm making runs on and the various tests that need to be performed to check that it's ready to ship, with a little interactive program to mark each test as pass/fail/skipped. That helps me get through the manual tests faster - but it also saves the data into a format that Claude can read, to check on the test status while I'm asleep and make further recommendations.
I don't think many people have figured out yet that you don't even need to prompt AI. Treat it well, treat it with respect, give it the opportunity and ability to do things, and there is a lot that will emerge. But if you treat AI like a tool, it performs about as well as if you treat your employees like tools.
You can prevent yourself from getting spam by not having an email account. But it's the nuclear option.
I'm fine with a system that can just read mail - and I already built one of those. I personally never send emails anyway so it's not an issue for me.
All of my tools are geared towards reducing noise and condensing information.
- My weather scripts tell me just the exact metrics I care about
- My email filtering system surfaces only the mails that are relevant right now - I can check the rest later.
- My RSS feed hydrator pre-filters Hacker News and other RSS feeds and adds data like comment/vote count etc to the feed itself so I can determine whether the link is worth opening just based on the information presented
None of these require an LLM to have free rein to modify things for me.
> My email filtering system surfaces only the mails that are relevant right now - I can check the rest later.
So then you have not actually addressed the concerns expressed in my post. You indeed have an LLM with both email access and Internet access. Exactly the scenario I described. The amount of trouble those two accesses together can cause is huge.
> None of these require an LLM to have free rein to modify things for me.
Give me read access to your email and an Internet connection and I bet I can come up with all sorts of ways to modify things for you. So can an LLM. If your lucky it won't.
You can rm -rf your entire hard drive, but you can't blame rm for it, it's you who did it, maybe because you don't know, or a mistake, doesn't matter.
When you ask the clanker to delete x number of files in a directory, it can reason itself that is easier to just get rid of the directory.
Can't expect deterministic outcomes out of a statistical model.
At it's current state its a wildcard, sure you can build guard rails, reduce permissions, but it's still a wildcard.
Let's not kid ourselves saying is just a skill issue.
Oh sure, so don't give it write access to anything important. And make backups.
Mine is on a VM. It doesn't have access to my host's files. The worst it will do is delete the files on the VM. No great loss.
Yes, I do get it to modify things on my host, but only via a REST API I've set up on my host, and I whitelist the things it can do (no generic delete, for example). I even let it send emails. But only to me. It can't send an email to anyone else.
> (…)
> Oh sure, so don't give it write access to anything important. And make backups.
If this conversation continues much longer, we’ll end up with “don’t use it at all”.
If I can’t trust a piece of software with anything important, why am I wasting my time fiddling with it? Might as well go play a video game or go do literally anything else entertaining.
Not what I said. As I've repeatedly said in this thread: Plenty of use cases where you don't give it access to email and write access to files. The comment you're replying to has an example of that.
> Might as well go play a video game or go do literally anything else entertaining.
True of most hobbies, right? I knew people who 20 years ago used to spend time in their garage building solar powered vehicles. But if I can't trust it to be reliable and safe on the road, I might as well go play a video game.
Also: Is anyone telling you to use it?
If everyone treated OpenClaw as a hobby, you might have a point, but people are using it for work in ways which will affect millions of other people when they’re hacked or the agent fucks up something important.
You already know how Meta’s AI Safety Director borked her email. Here’s the corporate vice president of Microsoft Word asking to be pwned:
https://www.omarknows.ai/p/meet-lobster-my-personal-ai-assis...
> Also: Is anyone telling you to use it?
You don’t need to use the technology to be affected by it. Ask Scott Shambaugh:
https://theshamblog.com/an-ai-agent-published-a-hit-piece-on...
People will always do stupid things. My guess is less than 10% (perhaps even less than 1%) are using it for work. Most workplaces wouldn't allow unfettered AI usage.
80-90% try it, find it unreliable and buggy, and give up on it.
Of the remaining ones, likely 90+% are not using it in (very) dangerous ways.
People like me using it for boring things aren't making the news, and aren't writing blog posts about "Look at the cool stuff I've done!" because getting OpenClaw to notify me of class openings is not worth writing about.
In my (large) company, we have a Slack channel for OpenClaw. Over 400 people are in that channel. Let's assume 10% are using it (at home). No one's lost files/emails or any other damage.
If you're old enough, you'll remember sentiments in the 80's and 90's where "Oh, you let your teen get a modem? He must be hacking/phreaking."
Or "Oh, he's using Linux? He must be using it to become a hacker."[1]
Most of the complaints I see on HN are from people who know little about it, and are going off negative press/posts. Just as people knew little about modems and Linux. I mean, having to tell people "Don't give it access to your emails" is a clear sign of their ignorance. Kind of like having to tell someone "OK, just don't give your 10 year old the car keys" when they complain that cars are inherently dangerous because 10 year olds can kill themselves driving it.
It's worth trying it in a secure environment so at least one can make an informed critique.
Like you, I steered clear of OpenClaw, seeing all the problems and all the money people were burning on tokens. But at some point, I decided I should at least try it in a safe way before rendering judgment. And now I see what it is. Has it done so much for me that I'd throw a lot of money at it? Heck no. Not yet at least. But I do see we're past the point of no return. OpenClaw itself may die, but some derivative of it is going to be transformational.
As I said: Make it secure, affordable, reliable and user friendly, and many App/SaaS services will disappear.
> You don’t need to use the technology to be affected by it. Ask Scott Shambaugh:
> https://theshamblog.com/an-ai-agent-published-a-hit-piece-on...
I don't know how old you are, but once everyone had a camera in their phones, the cat was out of the bag. Lots of people complaining about their photos showing up online because someone had taken a picture of them. Yes, this is bad. Yes, lives were lost (bullying, etc). And no, phones with cameras weren't going to go away. And everyone who complained has one now.
And as I pointed out a few days ago[2], the whole Scott Shambaugh episode was pretty mild compared to what some open source maintainers have had to deal with when it comes to humans.
[1] Lots of cases where ISPs, etc kicked customers out because they were using Linux and they didn't want the ISP to be implicated in criminal activities. "Only criminals use Linux"
rm won't wipe my HDD on a whim whilst instructing it to do something totally different.
You pretending they are the same thing is disingenous.
However, it seems better if you could, as much as is possible, move the AI stuff from runtime to “compile time.”
Instead of having the AI do everything all the time, have AI configure your Zapier (or whatever) on your behalf. That way you can (ideally) get the best of both worlds: the reliability and predictability of classical software, combined with the fuzzy interface of LLMs.
That is what many use OpenClaw for! The AI assistant will happily recommend existing services and help you (or itself, if you let it), set it up.
(In theory. In practice, it often does a poor job).
The appeal of OpenClaw is I don't need to go research all these possible solutions for different problems. I just tell it my problem and it figures it out.
Yesterday I told it to monitor a page which lists classes offered, and have it ping me if any class with a begin date in March/April is listed. This is easily scriptable by me, but I don't want to spend time writing that script. And modifying it for each site I want to be notified for. I merely spoke (voice, not text) to the agent and it will check each day.
(Again, it's not that reliable. I'm under no illusion it will inform me - but this is the appeal).
Like if you could just sit someone down for 30 minutes and show a few "power user" things, you will have truly taught her to fish for a lifetime. But it can go so unaddressed, and people's careers are built on these small ignorances.
I've cancelled everything at this point and just call Emacs my "special agential assistant," it makes me still sound in-the-know, and most of the time no one knows the difference!
"Convenience" in this context is laziness; "productivity" and "efficiency" is for management and bosses. We don't need to be our own bosses, I want to be free from such things as an individual. I want to be capable, be maybe almost "cool." Its sad to see a whole generation turn into such product dorks!
"Oh please read my email for me Mr. AI!"
The stuff you've listed are the kinds of things smart home enthusiasts do with whatever tools are available to them, and are just a sign of people exploring the possibility space.
But there's loads of people who would be stumped by a for loop, yet can easily work their way through a setup guide, particularly with the hype/promise and an active community.
I know about those tools, and I'm always in the mood for automating thing... and yet I don't use them.
I'm not yet running a Claw because of the prompt injection / lethal trifecta risks, but I absolutely understand the appeal. Reducing friction to automating stuff from "figure out Automator again" to "message your bot" is a material difference.
I have a desktop at home.
When I'm at work, I often think of TODOs for home. I write them on a post it note, and then at home have to remember to add it to my TODO (no, I'm not going to manage TODOs on my phone - whole other conversation).
I'll soon set up my Claw to be able to add TODOs (just add, not modify/delete). Then at work, I'll simply record a voice message to it telling it my TODO.
Same goes for movies I want to watch, books I want to read, reminders, etc.
I'm particular about the weather information I want (often want cloud cover percentage and precipitation probability for a set of hours). I couldn't find a good app on my phone that gives me this information. It was always a trip to a web site, modify some options, and hit Submit. Now I just ask my Claw and he has a skill for precisely my needs.
Here's an analogy: I carry a Leatherman multitool wherever I go. People ask me why. They can't comprehend needing it often to make it worth the hassle. But now that I have it on me, I use the knife very often - several times a week. And I almost never reach for a screwdriver. But until you've had it on you for a while, you can't comprehend the utility.
Back in 2005, lots of people asked "Why would I want a camera on my phone?"
I will. Far simpler, far more secure and far less wasteful than inserting some additional and unrequired LLM loop + hardware/virtualisation layer on top to do something I've already been able to do for years.
> But until you've had it on you for a while, you can't comprehend the utility.
You've not yet described anything that literally 99% of the population can not already do with the existing hardware and software in their pocket.
And far less capable. I have a whole system for managing TODOs, notes, etc, and I've not found an app that fits my needs (especially given that the system evolves over time).
But I agree - if you have a flow that works great for you with just a phone, then I wouldn't recommend OpenClaw for that use case.
> You've not yet described anything that literally 99% of the population can not already do with the existing hardware and software in their pocket.
And in the early days, literally everyone I knew who owned an iPhone also owned a digital camera and a laptop. Why pay some crazy amount for a fancy phone?
> I have a whole system for managing TODOs, notes, etc
So do I and it can already accept input via text or voice from my phone and so I’m just not seeing the benefit of inserting an additional LLM orchestration layer just to route tasks into it.
> And far less capable.
My current setup can already capture voice and text, and can route it automatically into Logseq, Trello, or even org-mode. These all can be viewed from the same (or other) devices. Honestly, I would have to think it was of more usage in retrieval than submission if I was to ever go that route. Like instead of reviewing my todos/schedule manually.
> And in the early days, literally everyone I knew who owned an iPhone also owned a digital camera and a laptop. Why pay some crazy amount for a fancy phone?
A Phone enabled entirely new behaviors that were not possible from my laptop or camera at the time. Persistent connectivity, real-time location services and on device image capture, edit and sharing just to name an incredibly small few. It created use cases that weren’t practical or even possible previously.
I’m open to being convinced, but I haven’t yet seen an example where OpenClaw enables something meaningfully new rather than repackaging something I can already do right now. Perhaps i'd be more open to it if it not literally everything I could think of using it for did not lead to atrocious security implmentations.
All of these things were doable without a smartphone - just very inconvenient to do so.
And all of these things are potentially dangerous ;-)
> Perhaps i'd be more open to it if it not literally everything I could think of using it for did not lead to atrocious security implmentations.
I was like you - seemed too scary (and expensive) to use. But given its traction, I figured I may end up at a disadvantage if I didn't at least learn it. So when I had some time, I simply looked at it, felt installing a VM is (relatively) low risk, and tried it out.
It's a very open ended tool. Ideas will come as you explore its capability. When LLMs came out, people (including me) said "Neat gimmick, but is it actually useful?" and stopped using it a month or two in. It was about 6 more months before I found a real use case for LLMs, and slowly I found more and more use cases.
And so it is with OpenClaw. For me, just being able to query it for the exact weather information I need is great - if you live in an area where knowing this stuff is important.
I have yet to find a tool that is good in converting receipts into the form I need. I spend time each month doing it manually. LLMs may get me there, and if it does, I'll likely hook it into OpenClaw.
For me what works is limiting and filtering everything. I'm a programmer so I write small tools for this, CLI-ones because they have the lowest barrier of entry for me and are also scriptable - and nowadays LLM-capable.
At work I get a bunch of Github notifications from all kinds of crap, on the Github site the options are "on" and "off" pretty much.
So I built a TUI tool that uses the `gh` CLI and filters out all that to just the exact set I want to see, nothing more.
I have a tool that summarises long Youtube videos for me in a way that I can quickly see if it's worth the 40 minutes to 2 hours of watching or is it just two ideas spread over multiple minues for more ad revenue.
Pretty much every operation that needs me to pick up my phone, unlock it, find an app and in the app scroll down somewhere to get to the information will eventually get a CLI tool instead. Gonna do a weather one right now just based on @BeetleB's idea :D
And I'm pretty sure by the end of the summer I'll have some flavour of Claw running and using those tools to proactively give me the information I need.
A simple example: Calendar has an event far enough to warrant driving -> claw checks weather around that time+location(s) -> notify me if it looks like I need to dig my car from under a snowbank again. Or at the very least ping me with the current + future weather prediction around that time.
I’ve seen the project name, but I’ve never been quite sure what the canonical noun is for a running instance. Are they actually called “claws”, or is that just me over-literalizing the branding?
Very curious
What makes Claws Claws is that they run automatically without human intervention, BUT they're built around an Agent =)
It's like if you could tell Claude Code that "at 8am tomorrow morning check the weather at my current location and send me a Telegram message". It doesn't know how to wait until 8am, it'll just do it. Or it'll create some massive hack to do it.
A Claw is built for things like that and has a "memory" of stuff it needs to do in the future and the ability check it periodically.
So there is no "correct" definition even to this day - and that vagueness makes the term difficult to have useful conversations around.
I've settled on "LLM making tool calls in a loop to achieve a goal", which I think is broad enough to cover what many people are talking about these days while being narrow enough to be useful.
On that basis Claude Code and OpenClaw are both examples of agents - they run tools and can loop until they've achieved the goal set for them by their user.
I'm not running OpenClaw either, but I'm getting a ton of value just from my homebrew deterministic "run while loop, wake up Claude if trigger event occurs".
I've posted about this before, I call it the Jarvis effect.
> For years we had people trying to make voice agents, like Iron Man's Jarvis, a thing. You had people super bought into the idea that if you could talk to your computer and say "Jarvis, book me a flight from New York to Hawaii" and it would just do it just like the movies, that was the future, that was sci-fi, it was awesome.
> But it turns out that voice sucks as a user interface. The only time people use voice controls is when they can't use other controls, i.e. while driving. Nobody is voluntarily booking a flight with their Alexa. There's a reason every society on the planet shifted from primarily phone calls to texting once the technology was available!
By and large the reason people love Openclaw is that it feels cool and futuristic. You have an AGENT! It's DOING THINGS! Yes it's doing things you could have easily done yourself, but you're not doing them yourself, you have an AGENT! It's all very silly, the same way that having your lights controlled by your phone is very silly, but some people like it.
That being said there a real use case for Openclaw, which is "marketing" (aka spam). A ton of people have set up Openclaw agents which exist to post on Twitter/Facebook/Discord/any open public user discussion forum (yes, HN included) to seem like a real member of a community, then start advertising something, generally crypto. So we can thank Openclaw for dead internet accelerationism.
With a mouse and keyboard I can switch windows.
With my voice, the computer can’t yet automatically determine if I am dictating a transcription or giving editing commands. What I really need is the interpreter listening to me to intuitively to know whether I am in the equivalent of VI command mode or insert mode.
It is the roadblock to not needing a screen at all, right now I want to visualize whether it understood me correctly because if it didn’t switch from insert to command automatically, I now have all my commands written into my paragraph. I also don’t want to listen to the computer talk back to me to confirm it listened. I want to just keep going, to keep narrating my thoughts and trust it’s doing the right things, not having to check. Having it slowly chime in to repeat that it listened derails my flow and train of thought.
TLDR The future of voice is headless vi.
It can only ever be a linear sequence of input
The 2 dimensional field of a screen and a mouse and keyboard give you extreme amounts of input and allow you to contextualize that input in arbitrary ways that intuitively make sense to people with minimal training. Most people do not need to be taught that "Paste" goes to the active window.
We barely even touch the surface of what is possible through this set of input devices and output and yet we can't even get that level of fine grained and reliable control into touch screen devices and gamepads, let alone a linear stream of pitch.
Voice cannot be a robust interface. It isn't between humans. There's immense nonverbal communication and human communication also relies very heavily on preshared context to actually get that info across in the first place. Even with all that machinery, human voice is generally considered to only carry, regardless of language, 44ish bits per second of data.
In 1996, I picked up the phone on my desk, dialed a 3 digit code, said “I need to fly to Los Angeles on Tuesday morning, returning Wednesday evening”. A couple hours later, an envelope appeared in my inbox with plane tickets, rental car reservation and hotel reservation.
Then every company in the world fired all the secretaries over the course of the next few years to cut costs, and we’ve collectively forgotten that it was ever like that.
1. How much can you spend on this trip? 2. Is first/business class necessary? 3. Is a layover acceptable if it's cheaper? 3a. Is it better to have a 4am flight nonstop or a 7am flight with a layover? 4. Are there preferred airlines? 5. Are there preferred hotel chains? What's the hotel budget? Do you want to pay extra for a nice view? 6. What kind of car should you rent? Is there equipment you'll be handling?
etc...
This is the kind of stuff that's easy(-ish) to communicate by presenting a list of options to a user through an actual interface. It sucks doing it through voice; think of the old phone systems where you had to go through droning "If you would like to rent an SUV, press 1. If you would like to rent a sedan, press 2. To speak to an operator, press 0."
So no, you never had a voice interface for booking flights; you had a human brain to whom you delegated, which is very different.
You already show many questions on one issue (a ticket).
Would you need to think about those? With the secretary the task is split: you decide you need the ticket, the secretary handles the rest while you focus on what you do.
Presenting all options, getting all callbacks, confirm e-mails etc does not change that. It puts all load on you.
p.s. isn't this https://www.in2013dollars.com/us/inflation/1996?amount=37000 about a third above the us median income? idk i don't live in the states
Ah, so that is indeed the endgame of what I've been seeing, hmm?
I was having a conversation with someone about OpenClaw, and they proposed this idea of OpenClaw being used for inventory tracking at the retail-level. I let them continue. They said it'd be the best option for tracking when purchases are made and what SKUs are sold at what time of day. They weren't talking about prompting, they were talking about it as a data store.
I didn't bother mentioning how long this problem had been solved.
It's not you being a curmudgeon.
The ironic thing here is that the person could go to ChatGPT (or whatever), describe the problem they're looking to solve, and ask it to find them the various ways it has been solved reliably (with links to the sources to confirm the information). And even provide some details on when each solution works best and why.
Because THAT is a great use for AI.
* Whatever the ai things is your will, will be done in a way the ai things is correct...
Eg. tell it to book a flight ticket for X without dealing with "modern UX" and 1GB websites
IMO OpenClaw or a similar agent will be on everyone's phone in a couple years. It's basically what Siri was always supposed to be. For the average user it's obvious that this is the way computers are meant to be interacted with.
Dangerous? Yes, very, but it truly feels like living in the future. Surprisingly, it's even more fun that sci-fi movies made me think this would be.
Almost every time I have an idea for AI Agent, I end up just making a script/binary that does the same, but so much faster that adding AI to it feels silly.
Recently I made a tool router that runs locally for such tools. Some tools have no arguments at all. Claude created a quick overlay where I can text/speak, and it will do tool call, without me asking for it, Claude added 4 buttons next to text input that bypass agent and just do a "tool call". I barely use text-to-command because those 4 buttons cover 9/10 of my use cases.
At this point I'm trying to come up with tools to add to it, so it's actually useful as an agent. Almost everything ends up being a cronjob or webhook triggered thing instead.
I often quip that I became a programmer specifically to avoid having to use spoken language. I always twitch at the thought of using any voice-based assistant.
Thinking in systems and algorithms is more enjoyable than using human language when it comes to computers IMHO ...
> I guess it's exactly the opposite for me ... I always hated using "normal" language with the computer.
it's called natural language https://en.wikipedia.org/wiki/Natural_language
> I often quip that I became a programmer specifically to avoid having to use spoken language. I always twitch at the thought of using any voice-based assistant.
You're one of these people who think that programming languages are structured and formal whereas in contrast natural language must be unstructured and lacking form? Going by the Chomsky hierarchy of formalisms natural language sits somewhere between context-free and context-sensistive https://en.wikipedia.org/wiki/Mildly_context-sensitive_gramm...
> Thinking in systems and algorithms is more enjoyable than using human language when it comes to computers IMHO ...
You don't think in "systems and algorithms" -- those are the outputs of your thinking.
There’s loads of good discussions about local LLMs in this thread:
That can't be allowed, and also won't happen. If token costs do start going up at a serious rate in the US, you can be sure that they'll stay down in China, and the political situation won't allow for the inevitable exodus to Chinese providers.
Is there something I'm missing about March or is it just a diverging reference? If the wave of non technical folks being able to automate new things is here, what's the equivalent impact of that? Maybe this is the inflection point where everyone needs more tech support like some sort of post Christmas surge? Maybe less because they have the tools to help themselves without trying now?
I'm not sure we're there yet anyway; I think this is still first adopters and enthusiasts. I asked my wife and some non technical friends and none of them have heard of openclaw yet. I think the deluge will happen if Apple or Android bakes it in or one of the big ai companies makes the app good enough for a normal person to unleash it upon their life and community.
Is my OpenClaw agent currently changing my life? No. It sends me a morning briefing based on my calendar, the weather, my Readwise highlights, and notes on who I'm talking to today based on call transcripts. I use it as a food diary (which I could have done on platform LLMs but this feels like a more personalized UX as we can write the logs to text files on my personal computer). I can absolutely see how transformative this agent can become in the next few years. Certainly my usage of LLMs has changed my life since ChatGPT first launched.
You are seeing the loudest / most hyped users. There's a reason it has so many stars and most of the people getting something out of it are not posting on X. They're just using it to do the thing.
That reason is buying stars, agent swarms, and astroturing.
No project gathers 200K stars genuinely in 3 months. There are far more useful and popular projects that need 10 years to get 200K stars. When you see a project like this get 200K stars in just 3 months, you know something is very fishy.
For a comparison, the local image gen interfaces ComfyUI and A1111 WebUI have a huge amount of stars (~100k and 160k respectively, accrued since 2022 or so), but they allow you to create porn customized to whatever kinks you have, not just automate things for the sake of automation. One of those is a rather bigger value prop than the other, dopamine-wise.
No you dont understand, just because there are X people capable of doing this and my project got (X + YX) stars in 3 months, that only means that my project is very popular and there are no shenanigans occurring _at all_
If you suggest otherwise you are a luddite who doesnt understand and probably hates progress.
Why is it so difficult to imagine that something that looks popular and fun is popular and fun?
Also, really who is paying for stars on open claw? Who benefits here?
I just don't get all the hyper either. I think it's because people just create automation workflows by typing them out rather than being in the trenches.
Im really not sure why this has to be said again and again.. it seems humans just don't learn do they?
Im waiting for someone to show me something that starts with the experience and then explains how the LLM fits in. Not the other way round.
I think because Google Search is predominantly tech-based, it is easy to see why LLMs have impacted the way we think about the experience associated with Search over large spaces of information.
Beyond that, Im not seeing much.
I never said OpenClaw was a bad idea.
I said the way most people are using it now isn't practical and/or saving them any time, and if there were ways, I would love to hear about them.
This is part of why the whole discussion has been so low value: people always default to "yep you're going to be proven wrong one day" or "you'll just be left behind then" instead of showcasing an actual, real life, practical example of using it to be more productive.
If you think it's fun and enjoyable, then have at it. I'm just not the biggest fan of people wasting a bunch of time on novelty and then telling me I'm dumb for not doing the same.
I assume a lot of these folks were already using LLM's quite a bit, but were using the Chat interfaces or had workflows that were split among a bunch of different services and tools. Something like OpenClaw gave them a way to centralize a lot of that and also gave them a way to use natural language to direct efforts. So for them this probably feels like a big step change.
If you are coming from a programming background you were aware that this type of setup has been doable for a while, but you were probably content sticking with Claude code or similar tools because those tools covered most of your LLM based workflows quite well.
And tying this altogether, one of the lowest hanging fruits for content creators is to create content about the tools they are using. Doubly so if that particular tool is starting to go viral. So you end up with a self feeding virality of sorts, as OpenClaw got more popular, more content creators started using it, and then publishing content about it, etc....
Like "message me when it is about to rain"
Really makes you think about what makes products good
I don't want to learn N different SaaS products (nor worry about them changing their TOS, going away, etc).
To be blunt, if OpenClaw were reliable, secure and affordable, lots of SaaS products would simply die. Why spend the time learning all of them when I can just tell the assistant what I want?
> or something simple you could make yourself.
That is OpenClaw at a higher abstraction! Instead of me sitting typing, or babysitting Claude Code, I can just tell OpenClaw what I want and it makes it for me.
(When it works, that is).
The market will eventually realize the business case for an OpenClaw-like product, and I'm waiting to ride its coattails!
reminds me of those "zune already does everything ipod does" posts.
I have an OpenClaw setup with a Claude API token and Qwen local model, running on an M4 Mac Mini with 32GB RAM.
1. At 7AM and periodically throughout the day it checks my calendars (work, parenting schedule, personal), a hyper local weather station, and some specific news topics — and sends me a summary and throughout the day updates if anything significant happens.
1b. It also sends this to my TRMNL e-ink display.
1c. It can also add and edit calendar invites, so if I want to move my yoga I can just tell it to move it to whenever the next yoga class is at (it knows what studio I go to and figures it out)
2. It has a skill I built that acts as a second brain for knowledge. I can send it Fitness Youtubes, parenting/health research papers, podcasts — and it organizes, summarizes and saves it in a logical file structure. Then in the future I can access these. It's like bookmarks on steroids. I love it for 1-2hour YouTube videos where I want summaries. It also pulls out any books any artifacts mentions and generates me a rolling reading list. https://plc.vc/npw
3. It has its own email address — and read access to my personal email — so friends can email it to schedule things like evening video game sessions. Similarly, if I get an urgent looking email it'll provide it in #1. I don't check my personal email aside from via OpenClaw.
4. It has read/write access to my GitHub, and each project repo I have has a well defined Claude structure, so it can make changes, commit the branches to Fly.IO and send me domains to test things. I love it for esoteric tweaks to my blog.
5. It has access to my Apple Reminders so I can message it things like "remind me to buy more muffins" and it has context to know to add those muffins to my Costco grocery list not Trader Joes.
6. It runs a headless browser, so when my hyper local weather service (Bouldercast) sends a summary that has more detail behind a login, it can open the email, click the link, login with my credentials, summarize the forecast, and send it to me.
7. It drafts blog posts for what it did for me each week. It's fun! https://plc.vc/d5t
I am a previous Zapier power user. I have used their LLMs, databases and Zaps extensively for the past decade. I understand the scorn towards AI, and I understand that if you look at this list you might think that it's either trivial tasks and/or things that could be done with Zapier, but I have been _amazed_ at how effortless it is to setup.
Similarly, I love that I can on the fly improve this assistant — last night I told it "I want to extend our Knowledge skill so that you can subscribe to RSS feeds and summarize articles in my knowledge base and also deliver interesting content in my daily summaries. Update the knowledge skill and our tasks to do all this."
It one shotted that, simply asking me to provide the first RSS feed I wanted to subscribe to.
It's genuinely like having a human assistant that happens to be an expert coder/technologist on call 247 that works at the near speed of light.
It disappoints me that technologists are so skeptical of this technology rather than exploring what it is and why it might be different to what exists today. It's fun! thats the takeaway: it's FUN.
It's most likely fun for you because it's novel, and for a lot of skeptics, AI novelty has long worn off.
The new shiny and new car smell will fade out, but some things stay after that. I'm about 90% sure Claws in some flavour will stay even after the hype slows down.
If a mobile phone company can figure out how to make this cost-efficient and be reasonably sure the assistant won't do anything that gets them in the news, their next device versions will have a similar branded system
Like other people have said people are having fun with their computers; that’s why it’s popular. That’s also why a bunch of people on forums throwing their hands up and saying “I don’t understand it. Why don’t they see that there shouldn’t be any fun whatsoever?” is not really a deterrent at all.
It’s also why it doesn’t matter that the categories of tasks they are doing can also be done with a whole set of tools that are no fun to use.
This applies to *clawphiles just as accurately.
I'm currently in founder mode and building two projects in Mexico. Finding the right real estate is in the critical path for success of the project. It's really hard to find quality real estate agents or brokers here and not because they don't exist. We have thousands they just never follow up with you or give you any updates. So, I'm using OpenClaw as my real estate agent that uses WhatsApp as a communications channel and manages the entire pipeline of first contact to scheduling a visit.
Right now, I feed it images of postings that I see on the street and OpenClaw handles everything else.
It really depends on where your personal or professional bottlenecks are and if you're running a business this project is absolutely amazing.
all the things I do with openclaw are in that ballpark of usefulness/importance.
I installed picoclaw on a whim (or nano? can't remember).
In maybe 15 minutes I had it make a "get weather for this specific area using the met.no API" skill and "check the train tables at these two stops for this specific line" skill.
Then I could just say "I go to the office every Monday on a train that leaves at 8, notify me if the weather is bad or there are delays in the train schedules"
And it just worked.
The "make a skill" bit was optional, it could've figured out both on its own, but I've been doing this for a while and figured out it's a lot more (token) efficient to have it specifically know how to do the things I want it to do.
---
Now lets take this loop and think about the system and what it could do.
Even if I wasn't a programmer and just went with "tell me when the train line I use for my commute is late" the system itself could see that "hmm, this looks like a thing I'll be doing often" and create a script/skill/plugin to do that via an official API (or WebMCP in the future).
You can't do that with Zapier or N8N.
There are many cool ways a pure LLM-powered system like that can be optimised, and more importantly, can be taught to self-optimise. By default I think the systems use the "main" model to read the HEARTBEAT.md file, which is stupid expensive. That could be done with a local model small enough to run on a modern phone.
And if that small local model says "yep, there's something to do", then it can either give the full task to a LLM or if it's smart enough it can spread specific tasks to small or medium local models first.
tl;dr OpenClaw is what Siri should've been after that epic fail of a Apple Keynote.
What are you talking about? In openclaw you literally type out basic English instructions to an agent. This isn’t painstakingly setting up chron jobs and writing scripts.
I built clawr.ing so my agent can call me on the phone for urgent things like emails I’m waiting for or issues in production for any of my products
React and Linux got their 200K stars slowly but surely over 10 years. OpenClaw got their 200K stars in like 3 months! Is this any meaningful comparison?
Getting 200K stars today doesn't mean much because today stars can be bought. There's a big shady thriving business of selling stars. Stars today can be generated using swarm of thoughtless agents. What's the use of counting these stars when they don't mean anything anymore?
Starring can be useful to the starer. They are just counted because it is countable. Whether you find the number meaningful remains up to you.
Let's wait for the explorers to return.
Would be nice to see a graph of stars by account age, how many of the stars come from accounts that are under 6 months old for example?
Compared that to the people who have been starring my projects, and every single one of them had some sort of activity on record.
--edit--
Checked again now, seems more recent accounts have some activity. But still, lots of accounts like these
Full story: https://brtkwr.com/posts/2026-03-02-upgrading-openclaw-to-la...
But could you estimate the token cost of this? Or were you able to comfortably do this with a subscription plan?
But wouldn't have been quicker and simpler to add ".bun/" to the pattern of authorized paths the same way it presumably works for ".npm/"?
> Starting around OpenClaw 2026.2.26, the project tightened plugin manifest validation. Manifests outside expected trust boundaries are now rejected as unsafe. On my Jetson, Bun’s global install layout (~/.bun/install/global/node_modules/...) tripped those checks for every single plugin
I'm also curious if it's particularly wise to have a web-facing system running on software that hasn't had a security update in 3 years?
React has been around for over a decade, and in that time pretty significantly impacted web dev paradigms (along with a few other mediums).
It’s hard to imagine being a web developer today and not knowing at least some react.
OpenClaw has been around for like a few months? And maybe it’s on its way to having that sort of impact? But right now seems to he mostly the purview of very early adopters and AI influencers.
It's larger than literally every open source service I thought to use as a benchmark. Rust, React, Vue, Symfony, Laravel, PHP Stan, Python, ESlint, rails, LLVM, Spring, fucking Linux itself.
I would wager the legitimate stars for OpenClaw are single digit percents of this. But hey, I haven't actually run the numbers on the GitHub API.
It's popular because people are exploring "what can i do with compute" once again. React is as meaningless as tickless kernels for vast majority of people.
Should be noted that i am not using any claws. Just a person remembering people spending their nights programming all sorts of psychedelic displays in qbasic with some moving on to asm. Asm for fun.
If we had a decent technical universe much of this stuff would work in ways that simply don't require LLMs for anything other than the initial setup.
I picked Whatsapp but it ended up using my own account! So it's absolutely too dangerous. We are supposed to create a separate account but with which phone number? I only have one.
So I picked Telegram instead, added it to a group chat, but it was a slog to set the authorizations.
In the end I don't trust it to read my mails for security reasons so I uninstalled it!
You're joking right?
Just like Jetbrains gives away IDEA licenses for open source projects, you need to have some metrics go up before they even consider you.
"If dev null is fast and webscale I will use it"
"Does dev null support sharding"
Who remembers https://www.youtube.com/watch?v=b2F-DItXtZs
For example, someone working in safety and alignment at Meta: https://nitter.net/FakePsyho/status/2025857836014538818
There are so many mails and communications with so many tasks to research via confluence etc
But on my private hardware? Idk, I haven't found a use for it...
Email is stored in your device somewhere, you could just get its content sent to your preferred agent, you can use one of the many mobile clients if you rather interact through that kind of UX (like self hosted hapi.run + tailscale for example), and you could setup a cronjob or something like it on your machine to do it based on whatever interval. This is just one way I personally could go about it, YMMV obviously.
Is there a specific reason why you will not do something like that for example?
For those who use Claude (or similar LLM APIs) on a daily basis, what does your monthly spend look like in practice? And do you feel the cost is justified by the value you’re getting?
or is it tooting its own horn?
It is a game changer. It can even automatically send email to other people who have podcasts and youtube channels no one watches to come on your podcast no one watches and you can both discuss what a game changer openclaw is.
It can even then watch the podcasts for both of you so now you both have viewers!
It is not easy understanding the current times in a /s way (or not)
I avoided the hype at first; however, it has become extremely efficient for emails and notes, and I can see how this can extend to any sort of digital workflow. The convenience of chatting with this thing, no matter where I'm at, is a key marker.
I’ve tried OpenClaw two weeks but don’t know what it can do for me.
I let it to finish some project for me, but the most hard work for project is validating the results over giving instructions
https://github.com/pjasicek/OpenClaw
OpenClaw - Captain Claw (1997) reimplementation
[1] https://news.ycombinator.com/item?id=3742902 [2] https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...
What's so good about that?
const openClawInstance = useOpenClaw(config);
Did anyone already vibe-code such silliness? If not, I want to give it a try.
React popularity is also a phenomenon closely tied to popularity of the fb
Most peoples thought process is "oh great idea, just gotta do this and that and out pops something that'll improve peoples lives". Erm no... its nothing like that in reality.
Like if you want bad weather alerts, you need to specifically define every situation of "bad weather" by givin exact weather / wind / rain limits.
What happens if it's cold, windy and rainy, but every value is juust 0.1 over the threshold of notifying you? It's still objectively shit weather.
Or can n8n go through your emails and check for promo codes or sales you're looking for - in ad mails that are 90% images with text?
Also, in the olden days of pre-AI, if our weather workflow did not notify us because conditions juuust failed to be met, we adjusted the thresholds. Uphill, both ways.
Don't get me wrong, I use a bunch of LLMs for automations. By prompting the model "here is what I want to achieve, here are the tools I have, figure out how to stitch them together". Actual workflows run (mostly) deterministically, with a sprinkling of "classify this image" or "summarize this text" nodes thrown in for a good measure.
There are all these douchebags that use to have to hire a human assistant in order to tell other douchebags that they have an assistant.
Now every wanna be douchebag can just get a mac mini. Of course, that is going to be super popular.
It's neat but the token use is pretty inefficient and security of course is a mess but it's been fun to play with.
I am messing with NanoClaw now and it's pretty much the same but only support Claude (uses code to do everything)
This takes maybe 10 minutes to write a script for…
[0]: https://news.ycombinator.com/item?id=46838946 [1]: https://news.ycombinator.com/item?id=47147183
I set it up, and had it do a few things, then decided its too risky after seeing some of the drastic failures it had caused some people.
Sure I understand you can sandbox it and all, but even then I couldn't think of much stuff I wouldn't want to do myself just nor justify the cost to run it.
It's weird that most people in these comments are speculating fraud.
Why aren't companies with real money to gain from stars gaming the system to the same degree? Why do the other metrics - issues and pull requests - match up with its popularity? Why would the bots starring the repo mean that those same bots are not popular? Those bots are controlled by their users.
The project is extremely active because this is what everyone being able to customize their computing looks like. A mess.
But it's a good mess.
Github was the old code sharing model clearly not designed for this. I'm sure a new model for code sharing will come to fix the growing pains.
A ton of people who would have never been able to customize their computing experience are finally able to. And it is magical for them.
This means that those same people will finally value having access to source and use of open protocols.
It was always valuable to us because we had the power to make it matter. It never mattered to them because they did not. Now they do.
The last era of computing was defined by dumbing down computing for the masses. Less information, less customizable, and more metric driven. Control in the hands of the companies.
This new era will look more free/libre, more personal, and less enshitified. Control in the hands of the users.
This is a very positive development.
How are you gaining money from stars? Why would Facebook bot stars for the react repo?
We also should remember that if this project had zero stars, we would hear crowing from these same people about how true and important that metric was. The idea that “open claw” paid for these stars somehow is mostly just reasoning backwards from the idea that no one would find this project interesting.
Why?
There are a bunch of open source projects that I want to see take off; I've never felt the urge to star one twice. I doubt that has to do with it being easier to say "Go star this project on github for me, with your own account" than it is to make a new account on github (which is not hard). I don't think that comes from any great moral fortitude, it's just...IMO hard for me to explain without an actual motive.
What's being alleged in this thread is widespread fraud via botting with no evidence of means or motive. As someone pointed out above, the argument for Facebook buying react stars is WAY stronger...and it is still really flimsy.
Why are people so keen to let a company get that close to their real life’s, it’s terrifying!
Steinberger and his VC club on Twitter were so salty about HN not understanding his grand creation that something needed to be done.
But that stargraph is ridiculous .. absolutely crazy
I don’t think it would fix things, except raise the bar for what is shilled and what isn’t.
If not explicitly prompted by the install process then it becomes another case study in AI accountability washing.
Come on HN.
This is the lowest, most boring form of programming.
Are some people using it in absolutely shitty ways? Yes, but that isn't the majority of the people playing with it.
The negativity I am seeing here is off the charts and undserved.
That's a surprise.
But suspicions on the legitimacy of the stars seems reasonable, wouldn't you agree? Look at the rate of stars, look at the comments/issues/prs on the repo. It feels safe to assume that most of them are from bots and not organic humans who went out to star a cool project.
"My React website can't star React"
"in what sense is this software not a virus?"
"GitHub stars are great for measuring the number of GitHub stars a project has"
etc.
All gold.