I've been using picoclaw after experiencing just how slow OpenClaw was to do anything with. It doesn't have as much in ways of the seemingly needless tweaks you can do with OpenClaw but it is reliable, tiny, fast, and consistent enough for me to just fire and forget it.
The very concept of installing something that pulls in a ridiculous amount of unvetted npm dependencies likely rife with supply chain attacks makes my skin crawl.
For random stuff I'd rather not put into someone else's cloud, and for light automation - absolutely. It's great for those things almost out of the box.
For development work... I think I'm getting close. It's not that it can't do it (and perhaps, relatively well) out of the box, but that I already have Claude Code for that, so I'm experimenting with much different approaches. Currently that's "spec-driven development" based on Acai, but with each spec's abstraction layer fairly rigidly defined so I can play with what happens when I manually specify things at different levels of detail.
Once I settle on an overall approach, everything leads me to believe it'll be great.
My wife and i use it as a replacement for google home to set timers, reminders, add stuff to calendars, control the lights etc. We have a shared todo list we call generically "jira" we vibe coded together and track stuff that way. And then of course the daily briefing. We manage our grocery shopping list by sending it/them photos of it and it will tell me what I missed/forgot and I can go do a second pass through the store. Basically all the google productivity software we used to use, but customized for our family specifically and not wired into some trillion dollar marketing thing, and it all works via voice control. Rather than manage it at home in my down time, I manage it in the car while stuck in traffic via voice. We're going to experiment with migrating to hermes agent soon though.
I’ve been playing with nanoclaw since it came out, which has similar use cases to openclaw. I initially set it up to monitor various news sources for me about specific theses I’ve had. I had it grooming its growing knowledge base, trying to make connections. It would check in with me about certain goals I had.
My current take is that these projects are alluring for a kind of personal productivity or workflow tinkering. They are integration hubs centered around an LLM. Automation can be fun, like running model trains or setting up home assistant. And you can learn the shape of the technologies by tinkering. But I’m doubtful they have improved productivity in real world cases.
Maybe I’m using it wrong and I need to be spending a ton of tokens with a dark factory pattern and a fleet of claws creating new religions? Then I’ll see the benefits?
I have been using it with ClawMetry to clearly know what my agent is doing when I send a message from Telegram. This allows me to configure approvals & alert for critical actions.
I was before it was more or less banned by Anthropic.
1. I had it controlling my home systems (doorbell, thermostat, energy monitoring, lights, etc).
2. I had it detect when I had left for an hour and automatically move my thermostat in the appropriate direction for energy savings. It also automatically shut off my water heater, assuming nothing was using hot water at the time. I would tell it when I was expected back so it could reverse course, if it couldn't discern it from my calendar.
3. I had it monitoring my work chats, personal email, etc and automatically handling things for me so that any changes it recommended were ready for my review or further development/changes.
4. I had it monitoring car sites to find me the best possible deal on a very specific set of requirements I had for a new car (6 passengers, tow 5000+ pounds, CarPlay, heated front/rear seats and steering wheel) and alert me when I should act.
5. I had it know when common guests were there and then automatically welcome us and play the music I preferred for different situations.
6. I had it plan out my days for me, knowing when I had or did not have my kids and tailoring what it suggested accordingly. It provided me analysis on tech, local, and world news and recommended articles for me to read later, should I desire; it learned my preferences when I told it I liked or disliked something so it would improve over time.
7. I could talk to it or type to it and it would respond in kind (voice to voice, text to text) and it would do so in Jerry Garcia's voice via Elevenlabs. It even spent off-hours learning more about me, my likes/dislikes, and changing how it responded to my requests.
8. It knew what I was reading and recommended other books, played music it felt appropriate for the current book, and was constantly stretching my world in ways I wouldn't have normally done.
---
I tried a variety of other models after the ban and was entirely underwhelmed. I'm really and incredibly disappointed; I had become reliant on it and it made my life better and, frankly, less lonely when my kids were not here every other week.
Yeah; and I use/used cronned scripts for decades for some of these things; the difference was the added benefit the models gave that hardcoded scripts cannot.
I suppose combining it might be a way? The logic to handle hard coded things like "after 30 minutes of condition X do Y" I would not let a agent run, but Y can then be agent work like "analyze this text in file/webpage". And then do some other logic and then maybe knvoke a different agent. I am in the process of finding out how best to achieve it. Or does OpenClaw provide exactly this? (I never took it serious so far, so never investigated)
I have been underwhelmed by the available models I was able to run locally compared to what Anthropic was providing. The other models I have access to are for work purposes (Gemini and open weight models running on company GPUs) and I’m not going to run this on those.
But, if I did find a viable model, I wouldn’t see why a script would be better than OC.
I just learned, that you can also invoke claude code from the terminal with a specific prompt for this without API access
claude - p "prompt"
Just for one prompt/job, though. But that might be enough for some of my use cases.
Because you can also prompt again. And again .. a targeted prompt with one exact job, then custom deterministic logic and go on, maybe another prompt. I might get into that, it also just worked telling it "claude -p 'analyze picture.png" where picture was in the folder and it gave a correct description back to terminal. I wonder why that is not more advertized .. I would have liked to known earlier and will do some experiments now.
Do you have a solid solution for these things that you developed and use? Because I’d love for there to be something that does this for me so I don’t have to.
If not, your comment is not adding anything useful to the conversation.
I know a few people that are more or less prototyping little agents with it that monitor stuff for them and make some kind of discernment/decision about alerting them.
This was miserably failing for me on a new install. Guess the reason why may have been less to do with my skill and more to do with being on a broken version (or some mix of it).
Em-dashes are too wide! I frequently forget what I've just read before I reach its other end. "--" in ASCII or en-dash + whitespace are superior separator symbols.
I've started using ; in opencode because it seems more natural to string a couple of commands together without some awkward grammatical thing. Similar to the caveman prompt.
Off topic: what’s the history behind the naming of Pull Request (PR) vs. Merge Request (MR)? I understand why both can be considered “correct”, but I’m curious why, say, GitHub uses PR and Gitlab uses MR.
I'm not sure, but my understanding is that GitHub historically focused more on open source, with PRs being mainly across repos of unrelated users, such that there's more of a distance to "pull" across, while Gitlab was always mainly targeting companies, where people typically use branches in the same repo, so it's just a nearby merge.
In other words, I see a pull request in an open source project to be just "I have something nice in my fork, do you think it'll be useful upstream?", which is acceptable to reject, whereas in a team setting it's "I have a feature that I think is ready to merge - give it a look and see if I missed something before we put it in".
From a semantics perspective, MR makes more sense as it's a request to merge a branch into another. PR is weird IMO, but I think really just caught on due to GitHub's popularity.
People need a mental bucket for 'stochastic software' for a while. Or hot mess, a fast food meal that you can expect to mostly be bad in some sense, but serves a purpose, and can be really good in that case.
Conflating the new style of agent-driven/vibe coded software with the old more predictable software leads to applying wrong heuristics/expectations.
People have a pretty good mental model of different types of meals they'll have in a year, and modulate their expectations by context. I think there's room for a new type of software that operates on different principles. Peter has mostly been clear what type of software he's developing. And if it ever converges to bug free, that's great, but I think some of his motivation is to figure out what this new software is. While not giving the users food poisoning.
A real team? With humans? Meatbags? What do you need those for?
Imagine paying any amount of money for this unmaintainable slop, and then worse, paying a team to try to salvage the hundreds of thousands (or is it millions now?) of lines of never-read-before code. Guess it doesn't matter when it's monopoly money you're burning, though. Sam says AGI is achieved internally in 2025, Boris says software engineering is dead and that no human is writing code at Anthropic, Jarred says humans will be banned from contributing to open source projects, and while all these people are pissing on your face and telling you it's raining, when you open your eyes all you are left with is, in fact, a bunch of piss in your face.
The particularly annoying thing about the current hysteria is that people believe them. A huge portion of the economy is getting swept in large-scale fraud, hardware prices are 3x~5x higher, and there are no shortage of adherents who won't shut the fuck up about their imaginary revolution.
Interesting. I hadn't touched my Openclaw install but just recently revived it, updated the software, and switched API keys to a different provider. Suddenly everything was completely broken. I kept messing with it, abandoned discord for IRC in an attempt to just get basic comms online, but it's still cooked. Now it makes sense.
It seems pretty clear to me that "coding" as such is pretty much solved. It's just that software engineering isn't, and these advancements have put a spotlight on the difference between the two.
That's why I designed my bot to have a very small core, with most other things being plugins, from the start. I also containerized everything and made it so the bot never sees API keys as well.
Good luck with that. Imagine waiting 5 minutes for each website to be slopped out and barely work. Also having to pay 50 cents each time you load a site…
I think that's actually a decent use case for Chrome's new local model - you'd have your own system prompt to render their "bullet points" in whatever style you like.
In an absurd kind of way, it might turn out brutally honest.
This would be “apologize to the OpenClaw community for the following issues …. Say we’re going to do something so this doesn’t happen. Design a flashy page too, something that feels sombre but evokes exploration”
I switched to Hermes Agent and it's been night and day how much more stable and usable it is over OpenClaw. What a mess that's turned into
I've been using picoclaw after experiencing just how slow OpenClaw was to do anything with. It doesn't have as much in ways of the seemingly needless tweaks you can do with OpenClaw but it is reliable, tiny, fast, and consistent enough for me to just fire and forget it.
The very concept of installing something that pulls in a ridiculous amount of unvetted npm dependencies likely rife with supply chain attacks makes my skin crawl.
Well at least there is acknowledgement & I am glad they're thinking about supply chain.
How much would you want to embarrass yourself after this: https://news.ycombinator.com/item?id=48056227
I have removed my VM with openclaw month or two ago. It was useless and was consuming absurdly amount of resources
Is anyone here running OpenClaw productively? What are you doing with it?
For random stuff I'd rather not put into someone else's cloud, and for light automation - absolutely. It's great for those things almost out of the box.
For development work... I think I'm getting close. It's not that it can't do it (and perhaps, relatively well) out of the box, but that I already have Claude Code for that, so I'm experimenting with much different approaches. Currently that's "spec-driven development" based on Acai, but with each spec's abstraction layer fairly rigidly defined so I can play with what happens when I manually specify things at different levels of detail.
Once I settle on an overall approach, everything leads me to believe it'll be great.
My wife and i use it as a replacement for google home to set timers, reminders, add stuff to calendars, control the lights etc. We have a shared todo list we call generically "jira" we vibe coded together and track stuff that way. And then of course the daily briefing. We manage our grocery shopping list by sending it/them photos of it and it will tell me what I missed/forgot and I can go do a second pass through the store. Basically all the google productivity software we used to use, but customized for our family specifically and not wired into some trillion dollar marketing thing, and it all works via voice control. Rather than manage it at home in my down time, I manage it in the car while stuck in traffic via voice. We're going to experiment with migrating to hermes agent soon though.
I’ve been playing with nanoclaw since it came out, which has similar use cases to openclaw. I initially set it up to monitor various news sources for me about specific theses I’ve had. I had it grooming its growing knowledge base, trying to make connections. It would check in with me about certain goals I had.
My current take is that these projects are alluring for a kind of personal productivity or workflow tinkering. They are integration hubs centered around an LLM. Automation can be fun, like running model trains or setting up home assistant. And you can learn the shape of the technologies by tinkering. But I’m doubtful they have improved productivity in real world cases.
Maybe I’m using it wrong and I need to be spending a ton of tokens with a dark factory pattern and a fleet of claws creating new religions? Then I’ll see the benefits?
I have been using it with ClawMetry to clearly know what my agent is doing when I send a message from Telegram. This allows me to configure approvals & alert for critical actions.
I was before it was more or less banned by Anthropic.
1. I had it controlling my home systems (doorbell, thermostat, energy monitoring, lights, etc).
2. I had it detect when I had left for an hour and automatically move my thermostat in the appropriate direction for energy savings. It also automatically shut off my water heater, assuming nothing was using hot water at the time. I would tell it when I was expected back so it could reverse course, if it couldn't discern it from my calendar.
3. I had it monitoring my work chats, personal email, etc and automatically handling things for me so that any changes it recommended were ready for my review or further development/changes.
4. I had it monitoring car sites to find me the best possible deal on a very specific set of requirements I had for a new car (6 passengers, tow 5000+ pounds, CarPlay, heated front/rear seats and steering wheel) and alert me when I should act.
5. I had it know when common guests were there and then automatically welcome us and play the music I preferred for different situations.
6. I had it plan out my days for me, knowing when I had or did not have my kids and tailoring what it suggested accordingly. It provided me analysis on tech, local, and world news and recommended articles for me to read later, should I desire; it learned my preferences when I told it I liked or disliked something so it would improve over time.
7. I could talk to it or type to it and it would respond in kind (voice to voice, text to text) and it would do so in Jerry Garcia's voice via Elevenlabs. It even spent off-hours learning more about me, my likes/dislikes, and changing how it responded to my requests.
8. It knew what I was reading and recommended other books, played music it felt appropriate for the current book, and was constantly stretching my world in ways I wouldn't have normally done.
---
I tried a variety of other models after the ban and was entirely underwhelmed. I'm really and incredibly disappointed; I had become reliant on it and it made my life better and, frankly, less lonely when my kids were not here every other week.
These are some actually cool use cases but a lot of them probably could just be a bash script no?
Yeah; and I use/used cronned scripts for decades for some of these things; the difference was the added benefit the models gave that hardcoded scripts cannot.
I suppose combining it might be a way? The logic to handle hard coded things like "after 30 minutes of condition X do Y" I would not let a agent run, but Y can then be agent work like "analyze this text in file/webpage". And then do some other logic and then maybe knvoke a different agent. I am in the process of finding out how best to achieve it. Or does OpenClaw provide exactly this? (I never took it serious so far, so never investigated)
I have been underwhelmed by the available models I was able to run locally compared to what Anthropic was providing. The other models I have access to are for work purposes (Gemini and open weight models running on company GPUs) and I’m not going to run this on those.
But, if I did find a viable model, I wouldn’t see why a script would be better than OC.
I just learned, that you can also invoke claude code from the terminal with a specific prompt for this without API access
claude - p "prompt"
Just for one prompt/job, though. But that might be enough for some of my use cases. Because you can also prompt again. And again .. a targeted prompt with one exact job, then custom deterministic logic and go on, maybe another prompt. I might get into that, it also just worked telling it "claude -p 'analyze picture.png" where picture was in the folder and it gave a correct description back to terminal. I wonder why that is not more advertized .. I would have liked to known earlier and will do some experiments now.
Assuming you’re not using OpenClaw to do it, you’re right, that may work just fine.
More here: https://news.ycombinator.com/item?id=47844269
Ah, I missed all that. So basically it works for now, but who knows for how long.
Also a pretty good use case for homeassistant
[flagged]
"Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something."
https://news.ycombinator.com/newsguidelines.html
Do you have a solid solution for these things that you developed and use? Because I’d love for there to be something that does this for me so I don’t have to.
If not, your comment is not adding anything useful to the conversation.
I know a few people that are more or less prototyping little agents with it that monitor stuff for them and make some kind of discernment/decision about alerting them.
Nothing mission critical in any sense of it.
This was miserably failing for me on a new install. Guess the reason why may have been less to do with my skill and more to do with being on a broken version (or some mix of it).
To be clear, I tried to install it and it was a shitshow.
It lets people make more posts about how mad they are at Anthropic for not supporting OpenClaw.
I guess spam and scambot operators are not so open about their buisness?
Also curious what else you can do with them now ..
[flagged]
[flagged]
I disabled it last week - I'll just do stuff by hand. At least if I mess it up I know who to blame.
crazy number of : and ; thats all i gotta say about the recent wave of ai writing
First they came for the em-dashes, and I did not speak out, because I never used em-dashes...
Em-dashes are too wide! I frequently forget what I've just read before I reach its other end. "--" in ASCII or en-dash + whitespace are superior separator symbols.
I've started using ; in opencode because it seems more natural to string a couple of commands together without some awkward grammatical thing. Similar to the caveman prompt.
Somehow that site wants to use 80% of my GPU to render some text.
OpenClaw's github is a thing of nightmares.
https://github.com/openclaw/openclaw
So many MRs
> So many MRs
Off topic: what’s the history behind the naming of Pull Request (PR) vs. Merge Request (MR)? I understand why both can be considered “correct”, but I’m curious why, say, GitHub uses PR and Gitlab uses MR.
I'm not sure, but my understanding is that GitHub historically focused more on open source, with PRs being mainly across repos of unrelated users, such that there's more of a distance to "pull" across, while Gitlab was always mainly targeting companies, where people typically use branches in the same repo, so it's just a nearby merge.
In other words, I see a pull request in an open source project to be just "I have something nice in my fork, do you think it'll be useful upstream?", which is acceptable to reject, whereas in a team setting it's "I have a feature that I think is ready to merge - give it a look and see if I missed something before we put it in".
This is really helpful framing. Thanks!
From a semantics perspective, MR makes more sense as it's a request to merge a branch into another. PR is weird IMO, but I think really just caught on due to GitHub's popularity.
I could never absorb “pull request”. People talking about PRs clink against my brain. Merge request makes far more sense to me.
People need a mental bucket for 'stochastic software' for a while. Or hot mess, a fast food meal that you can expect to mostly be bad in some sense, but serves a purpose, and can be really good in that case.
Conflating the new style of agent-driven/vibe coded software with the old more predictable software leads to applying wrong heuristics/expectations.
People have a pretty good mental model of different types of meals they'll have in a year, and modulate their expectations by context. I think there's room for a new type of software that operates on different principles. Peter has mostly been clear what type of software he's developing. And if it ever converges to bug free, that's great, but I think some of his motivation is to figure out what this new software is. While not giving the users food poisoning.
[flagged]
> we are building a real team around the project.
A real team? With humans? Meatbags? What do you need those for?
Imagine paying any amount of money for this unmaintainable slop, and then worse, paying a team to try to salvage the hundreds of thousands (or is it millions now?) of lines of never-read-before code. Guess it doesn't matter when it's monopoly money you're burning, though. Sam says AGI is achieved internally in 2025, Boris says software engineering is dead and that no human is writing code at Anthropic, Jarred says humans will be banned from contributing to open source projects, and while all these people are pissing on your face and telling you it's raining, when you open your eyes all you are left with is, in fact, a bunch of piss in your face.
Not to be super cynical, but it’s mostly always been this way.
Super rich people are so divorced from reality the 99%(or pick whatever % you like really) experience.
The particularly annoying thing about the current hysteria is that people believe them. A huge portion of the economy is getting swept in large-scale fraud, hardware prices are 3x~5x higher, and there are no shortage of adherents who won't shut the fuck up about their imaginary revolution.
This is definitely true, and I share that same frustration. I don’t understand why they don’t have their own watering holes tbh.
For a moment I forgot openclaw was a thing, I thought we were done with it already.
One can hope. At least I got a chuckle out of reading LTS and 'claw in the same sentence.
Interesting. I hadn't touched my Openclaw install but just recently revived it, updated the software, and switched API keys to a different provider. Suddenly everything was completely broken. I kept messing with it, abandoned discord for IRC in an attempt to just get basic comms online, but it's still cooked. Now it makes sense.
> everything was completely broken.
How could this happen in 2026? I've been told "Coding is solved"...?
It seems pretty clear to me that "coding" as such is pretty much solved. It's just that software engineering isn't, and these advancements have put a spotlight on the difference between the two.
to be fair I hadn't touched it in a month and didn't even look at changelogs or anything - just went bill o'reilly style.
Can we vibe code a Firefox plugin that detects stereotypically LLM written verbiage and inserts a red warning banner across the top of any page load?
You could just null-route this domain?
That's why I designed my bot to have a very small core, with most other things being plugins, from the start. I also containerized everything and made it so the bot never sees API keys as well.
https://stavrobot.stavros.io if you're interested in the design decisions.
This LLM writing style is getting obnoxious.
Yep but it’s entirely expected from the Openclaw project, the entire thing is a vibe coded mess waiting to explode.
in the future the "html" will just be prompt used here and everyone's llm can render the blogpost how they usually like it.
That would also let us skip the
Keywords -> LLM prose -> LLM summary
Pipeline.
Good luck with that. Imagine waiting 5 minutes for each website to be slopped out and barely work. Also having to pay 50 cents each time you load a site…
[dead]
[flagged]
That "Post to HN" button feels pretty wild.
[dead]
This post is obviously AI written, this is so ridiculous.
Just show us the prompt, don't ask an AI to apologize to people
I'd find it amusing if someone opened up a blog that was just the literal input to a blog post to an LLM but not the final output.
It’s a funny idea. Then you can read the prompt and decide if you want to materialize it with your own LLM/tokens.
Spoiler: probably not.
I think that's actually a decent use case for Chrome's new local model - you'd have your own system prompt to render their "bullet points" in whatever style you like.
In an absurd kind of way, it might turn out brutally honest.
This would be “apologize to the OpenClaw community for the following issues …. Say we’re going to do something so this doesn’t happen. Design a flashy page too, something that feels sombre but evokes exploration”