Yes, I’m running it with a minimal set of plugins.
When I’m driving or out I can ask Siri to send a iMessage to Clawdbot something like “Can you find out if anything is playing at the local concert venue, and figure in how much 2 tickets would cost”, and a few minutes later it will give me a few options. It even surprised me and researched the different seats and recommended a cheaper one or free activities as an alternative that weekend.
Basically: This is the product that Apple and Google were unable to build despite having billions of dollars and thousands of engineers because it’s a threat to their business model.
It also runs on my own computer, and the latest frontier open source models are able to drive it (Kimi, etc). The future is going to be locally hosted and ad free and there’s nothing Big Tech can do about it. It’s glorious.
> This is the product that Apple and Google were unable to build
It's not they're unable to build it, it's that their businesses are built on "engagement" and wasting human time. A bot "engaging" with the ads and wasting its time would signal the end of their business model.
> It also runs on my own computer, and the latest frontier open source models are able to drive it (Kimi, etc). The future is going to be locally hosted and ad free and there’s nothing Big Tech can do about it. It’s glorious.
After messing with openclaw on an old 2018 Windows laptop running WSL2 that I was about to recycle, I am coming to the same conclusion, and the paradigm shift is blowing my mind. Tinkerers paradise.
Same here. I like tinkering with my Home Assistant setup and small web server running miscellaneous projects on my Raspberry Pi, but I hate having to debug it from my phone when it all falls over while I'm not near my computer.
Being able to chat with somebody that has a working understanding of a Unix environment and can execute tasks like "figure out why Caddy is crash looping and propose solutions" for a few dollars per month is a dream come true.
I'm not actually using OpenClaw for that just yet, though; something about exposing my full Unix environment to OpenAI or Anthropic just seems wrong, both in terms of privacy and dependency. The former could probably be solved with some redacting and permission-enforcing filter between the agent and the OS, but the latter needs powerful local models. (I'll only allow my Unix devops skills to start getting rusty once I can run an Opus 4.5 equivalent agent on sub-$5000 hardware :)
Honest answer: OpenClaw still requires some tinkering, but it's getting easier.
The biggest barrier for non-tinkerers is the initial setup - Node.js, API keys, permissions, etc. Once it's running, day-to-day use is pretty straightforward (chat with it like any other messaging app).
That said, you'll still need to:
- Understand basic API costs to avoid surprises
- Know when to restart if it gets stuck
- Tweak settings for your specific use case
If you're determined to skip tinkering entirely, I'd suggest starting with just the messaging integration (WhatsApp/Telegram) and keeping skills/tools minimal. That's the lowest-friction path.
For setup guidance without deep technical knowledge, I found howtoopenclawfordummies.com helpful - it's aimed at beginners and covers the common gotchas.
Is it transformative without tinkering? Not yet. The magic comes from customization. But the baseline experience (AI assistant via text) is still useful.
> The future is going to be locally hosted and ad free and there’s nothing Big Tech can do about it.
I wouldn't be so certain of that. Someone is paying to train and create these models. Ultimately, the money to do that is going to have to come from somewhere.
Good News! The Models are done, and you can download them for free. Even if they stopped being worked on this moment, those are finished and usable right now, and won't get any worse over time.
Depending on the pre-training structure, you can fine-tune old local models (even locally if you have a nice GPU) to steer behavior towards your desires.
They can still be very useful because new models are reaching an asymptote in performance. Meanwhile as hardware gets cheaper in the future (current RAM prices notwithstanding), these models will become faster to run on local hardware.
Quantized, heavily, and offloading everything possible to sysram. You can run it this way, just barely reachable with consumer hardware with 16 to 24gb vram and 256gb sysram. Before the spike in prices you could just about build such a system for $2500, but the ram along probably adds another $2k onto that now. Nvidia dgx boxes and similar setups with 256gb unified ram can probably manage it more slowly ~1-2 tokens per second. Unsloth has the quantized models. I’ve test Kimi though don’t have quite the headroom at home for it, and I don’t yet see a significant enough difference between it and the Qwen 3 models that can run in more modest setups: I get a highly usable 50 tokens per second out of the A3B instruct that fits into 16gb VRAM with enough left over not to choke Netflix and other browser tasks, it performs on par with what I ask out of Haiku in Claude Code, and better as my own tweaking improves with the also ever better tooling that comes out near weekly.
I have an AMD Epyc machine with 512GB of RAM and a humble NVIDIA 3090. You will have to run a quantized version but you can get a couple tokens per second out of it since these models are optimized to split across the GPU/RAM and it's about as good as Claude was 12 months ago.
Full disclosure, I use OpenRouter and pay for models most of the time since it's more practical than 5-10 tokens per second, but the option to run it "If I had to, worst case" is good enough for me. We're also in a rapidly developing technology space and the models are getting smaller and better by the day, ever year the smaller models get better
OP implied they have powerful enough hardware, since Kimi runs on their computer, so that is why they mentioned it is local. That it doesn't work for most people has no relation to what OP of this thread said. Regardless, you don't need an Opus level model, you can use a smaller one that'll just be slower at getting back to you, it's all asynchronous anyway compared to a coding agent where some level of synchronicity is expected.
This thread is about giving one's opinions on their personal experiences with the tool, so the OP of the thread can say whatever they want, it doesn't mean they think it's at all related to the "majority user experience" nor do they have to cater their opinions towards that.
It's the other way around. At least for most people, it grants access to your personal data to an LLM (and by extension its inference provider) in the cloud.
I think Karpathy[1] summarized why he thinks this is the case quite well (as described he was himself hyping it up a bit much, but there are some foundational reasons why it's a very interesting experiment).
"it's nothing new and it's a lot of scams and garbage, but it's just bigger than before, but I still think there will be something transformative there eventually"
Seems like a Rorschach test. If you think this sort of thing is gonna change the world in a good way: here's evidence of it getting to scale. If you think it's gonna be scams, garbage, and destruction: here's evidence of that.
Karpathy is one of the biggest tech grifters of our time, so finding out that he's jumped on this grift train as well comes as no surprise.
Actually, hang on... yep, to absolutely nobody's surprise, Simon Willison has also hyped this up on his blog just yesterday. The entire grift gang is here, folks.
Maybe it has always been this way but it seems like these days it's only a matter of time before anyone "authentic" (or at least seems authentic at first) turns into some type of grifter. If you have a big enough following there is too much money to be made not to grift.
I used to follow Fireship, I even have him connected on LinkedIn. Look where he is now. I also used to follow simonw, but I think he is going down the same spiral.
I completely and utterly disagree. Simon is no gifter. Intellectually lazy snobs think that any time someone has genuine delight and excitement about something that they’re “grifting.” Also, for anyone who is trying (regardless of whether they are succeeding) the to make money—that isn’t synonymous with “grifting.” God, some of you need a better relationship with a dictionary.
Plus, it seems some of y’all love to hate the very industry which puts a roof over your head. You’re hoping and praying that it all burns down—yet where will that leave you? How do you feel about becoming a plumber—-until the robots take that job?
If you want to know my honest opinion, I just wish we had assisted suicide for anyone who wants it. You can go be genuinely delighted about the latest Twitter hype wave, I don't care. Caring about others is a ship that had sailed for me long ago. I would rather just exit. Go, post a Show HN where with Claude Code as top contributor. +4,294,967,295 -2,147,483,647 lines on code in two months. Go, run your THOUSANDS OF AGENTS. Your GitHub star history will look like NVDA zoomed out. I don't care.
> Plus, it seems some of y’all love to hate the very industry which puts a roof over your head. You’re hoping and praying that it all burns down—yet where will that leave you? How do you feel about becoming a plumber—-until the robots take that job?
This probably isn't a line of argument you want to go down. I've been unemployed for 7 months, in part due to how difficult it is to get so much as an intro call because so many people have totally automated the process of spamming every open job posting with as many resumes (many of which were likely LLM-generated as well) as possible.
In this case it's only about payout from views/engagement of posts.
There is no commercial interest from the developer of OpenClaw. He doesn't make any money from it. He made enough from selling his startup a few years back.
So when we suspected some companies to game the Twitter algorithm to make money, maybe they were not responsible for it at all.
For most people it works like that. Only a tiny minority keeps pushing after that point.
I just can't see an angle to OpenClaw that could provide a substantial financial gain for the creator. It's clearly a passion project. Like Ghostty from Mitchell Hashimoto.
Here is what I have my openclaw agent setup to do in my wsl environment on my 22 core development workstation in my office:
#1) I can chat with the openclaw agent (his name is "Patch") through a telegram chat, and Patch can spawn a shared tmux instance on my 22 core development workstation.
#2) I can then use the `blink` app on my iphone + tailscale and that allows me to use a command in blink `ssh dev` which connects me via ssh to my dev workstation in my office, from my iphone `blink` app.
Meanwhile, my agent "Patch" has provided me a connection command string to use in my blink app, which is a `tmux <string> attach` command that allows me to attach to a SHARED tmux instance with Patch.
Why is this so fking cool and foundationally game changing?
Because now, my agent Patch and I can spin up MULTIPLE CLAUDE CODE instances, and work on any repository (or repositories) I want, with parallel agents.
Well, I could already spawn multiple agents through my iphone connection without Patch, but the problem is then I need to MANAGE each spawned agent, micromanaging each agent instance myself. But now, I have a SUPERVISOR for all my agents, Patch is the SUPERVISOR of my muliple claude code instances.
This means I no longer have to context switch by brain between five or 10 or 20 different tmux on my own to command and control multiple different claude code instances. I can now just let my SUPERVISOR agent, Patch, command and control the mulitple agents and then report back to me the status or any issues. All through a single telegram chat with my supervisor agent, Patch.
This frees up my brain to only have to just have to manage Patch the supervisor, instead of micro-managing all the different agents myself. Now, I have a true management structure which allows me to more easily scale. This is AWESOME.
This feels like the "prompt engineering" wave of 2023 all over again. A bunch of hype about a specific point-in-time activity based on a lot of manual setup of prompts compared to naive "do this thing for me" that eventually faded as the tooling started integrating all the lessons learned directly.
I'd expect that if there is a usable quality of output from these approaches it will get rolled into existing tools similarly, like how multi-agents using worktrees already was.
Maybe this is just a skill issue on my part, but I'm still trying to wrap my head around the workflow of running multiple claude agents at once. How do they not conflict with each other? Also how do you have a project well specified enough that you can have these agents working for hours on end heads down? My experience as a developer (even pre AI) has mostly been that writing-code-fast has rarely been the progress limiter.. usually the obstacles are more like, underspecified projects, needing user testing, disagreements on the value of specific features, subtle hard to fix bugs, communication issues, dealing with other teams and their tech, etc. If I have days where I can just be heads down writing a ton of code I'm very happy.
I spec out everything in excruciating detail with spec docs. Then I actually read them. Finally, we create granular tasks called "beads" (see https://github.com/steveyegge/beads). The beads allows us to create epics/tasks/subtasks and associated dependency structure down to a granular bead, and then the agents pull a "bead" to implement. So, mostly we're either creating spec docs and creating beads or implementing, quality checking, and testing the code created from an agent implementing a bead. I can say this produces better code than I could write after 10yrs of focused daily coding myself. However, I don't think "vibe coders" that have never truly learned to code, have any realistic chance of creating decent code in a large complex code base that requires a complex backend schema to be built. They can only build relatively trivial apps. But, I do believe what I am building is as solid as if I had a millions of dollars of staff doing it with me.
But how is that less work and allows you to do that in Disneyland with your kids? For me, personally, there is little difference between "speccing out everything in excruciating detail in spec docs" and "writing actual implementation in high-level code". Speccing in detail requires deep thought, whiteboard, experimentation etc. All of this cannot be done in Disneyland, and no AI can do this at good level (that's why you "spec out everything in detail", create "beads" and so on?)
Yes, I normally draft spec docs in the office at my desk, this is true. However, when I have the spec ready for implementation with clear "beads", I can reasonably plan to leave my office and work from my phone. Its not at a point where I can just work 100% remote from my phone (I probably could but this is all still new to me too). But it does give me the option to be vastly more productive, away from my desk.
Not yet, but I can tell you that producing "good" code is another layer altogether. I have custom linters, code standardization docs, custom prompts, strictly enforced test architecture (enforced by the custom linters in pre-commit hooks which run before an agent tries to commit). Ultimately, it's a lot of work to get all the agents with a limited context writing code in the way you want. In the main large complex project I am generally working on now, I have hand-held and struggled for over a year getting it all setup the way I need it. So I can't say its been a weekend setup for me. It's been a long arduous process to get where I am now in my 2-3 main repos that I work on. However, the workflow I just shared above, can help people get there a lot faster.
> but I can tell you that producing "good" code is another layer altogether.
I feel like it isn't. If the fundamental approach is good, "good" code should be created as a necessity and because there wouldn't be another way. If it's already a mess with leaking abstractions and architecture that doesn't actually enforce any design, then it feels unlikely you'll be able to stack anything on top of below it to actually fix that.
And then you end up with some spaghetti that the agent takes longer and longer to edit as things get more and more messy.
Here is my view after putting in my 10,000+ hours learning to code pre-llm days, while also building a pretty complex design + contract manufacturing company, with a lot of processes in place for that to happen. If you have a bunch of human junior devs and even a senior dev or two that join your org to help you build an app, and you don't have any dev/ops structure in place for them, then you will end up with "spaghetti" throughout all your code/systems, from those relatively bright humans. Its the same managing agents. It cannot be expected to build a complex system from simple "one shot me a <x> feature" from a bunch of different agents, each with a finite ~150k token context limit. It must be done in context of the system you have in place. If you have a poor/no system structure, you'll end up with garbage for code. Everything that I said I had to guide the agents, is also useful for human devs. I'm sure that all the FANGS and various advanced software companies also use custom linters, etc., for every code check in. It's just now become easier to have these advanced code quality structures in place, and it is absolutely necessary when managing/coordinating agents to build a complex application.
I've clocked some hours too, and I think as soon as you let something messy in, you're already losing. The trick isn't "how to manage spaghetti" with LLMs (nor humans), because the context gets all wired up, but how to avoid it from first place. You can definitely do "one-shot" over and over again with a small context and build something complex, as long as you take great care about what goes into the context, more isn't better.
Anyways, feels like we have pretty opposite perspectives, I'm glad we're multiple people attacking similar problems but from seemingly pretty different angles, helps to find the best solutions. I wish you well regardless and hope you manage to achieve what you set out to do :)
I don’t get it, and that doesn’t mean it’s not a bad thing necessarily. I’ve been doing systems things for a long time and I’m quite good at it but this is the first time none of this excites me.
Instead of sitting in my office for 12 hours working with 20 open terminals (exactly what I have open right now on my machine). I can take my kids to Disneyland (I live in Southern California and it's nearby) and work on my iphone talking to "Patch" while we stand in line for an hour to get on a ride. Meanwhile. my openclaw agent "Patch" manages my 20 open terminals on my development workstation in my office. Patch updates me and I can make decisions, away from my desk. That should excite anyone. It gives me back more of my time on earth, while getting about the same (or more) work done. There is literally nothing more valuable to me than being able to spend more time away from my desk.
If this is actually true, then what will soon happen is you will be expected to manage more separate “Patch” instances until you are once again chained to your desk.
Maybe the next bottleneck will be the time needed to understand what features actually bring value?
I appreciate your insight, even if the workflow seems alien to me. I admit I like the idea of freeing myself from a desk though. If you don't mind me asking, how much does this all cost per month?
Thanks. I just mentioned elsewhere, right now I spend $200 on claude code 20x plan + $200 on openAI's similar plan, per month. I probably have a few more small conveniences that cost ~$10-$20 a few places, like an obsidian vault synch for documentation vaults on both my dev workstation and my phone, comes to mind. Most weeks I could cut one of the $200 plans, but both claude code and codex have different strengths, and I like to have them double check each others work, so to me that's worth carrying both subscriptions.
i have been recently quite enamoured with using both the ChatGPT mobile app (specifically the Codex part) and the Github mobile app, along with Codex. with an appropriate workflow, i've been able to deploy features to some [simple] customer-facing apps while on the go. it's very liberating!
GP's setup sounds like the logical extension to what i'm doing. not just code, but sessions within servers? are sysadmins letting openclawd out and about on their boxes these days?
You are absolutely right that I probably still "don't" get it, I am still shocking myself on a daily basis with all the stuff I didn't fully get grasp ahold of. I recently updated claude code and yesterday had one agent that used the new task system and blew my mind with what he got accomplished. This tech is all moving so fast!
Personally, I spend $200 on claude code 20x plan + $200 on openAI's similar plan, per month. So, yeah, I spend $400 per month. I buy and use both because they have different and complimentary strengths. I have only very rarely almost reached the weekly capacity limit on either of those plans. Usually I don't need to worry about how much I use them. The $400 may be expensive to some people but frankly I pay some employees a lot more each month and get a lot less for my money.
They make it easy to spin up parallel agents. Managing them efficiently through a shared tmux instance isn't banned anywhere in the TOS, AFAIK. I'd worry more about it if I had to use multiple accounts or something using round-the-clock "automated" work flow. I'm using one account. Hell, the workflow I described, I am even actively logged in to my dev workstation with tmux and able to see and interact with each instance and "micro-manage" them myself, individually. The main benefit of this workflow is that I also have a single shared LLM instance that also has access to all the instances, together with me. I have plenty of other things to worry about besides a banned account from an efficient workflow I've set up.
just throwing out there that yesterday Boris (lead engineer for Claude code) literally told everyone on Twitter that the CC teams number one recommendation for users is that they should be kicking off multiple instances / agents in parallel. not sure if that's what you're referring to, but if so I'd be very surprised if they ban someone for heavy use of that workflow
Actually using it. Threw it on a spare box I had sitting around, mostly as a "second brain" rather than letting it run my life.
Honestly, I've tried every PKM system out there—Obsidian, Notion, Roam, plain markdown—and never stuck with any of them. This is the first thing that's clicked. I just chat with it and it figures out what to file where. Everything's just .md files underneath, so I can grep it, git it, whatever. No lock-in.
The stuff I'm actually finding useful: it sends me news digests on topics I care about a few times a day, and pings me with reminders via Telegram. Simple stuff, but it works. Could I build this with Claude Code and some glue? Sure. But this was basically working out of the box.
Caveats: it chews through tokens fast, and I keep it completely cut off from anything sensitive—no email, no messages, nothing financial. The security story is basically "hope for the best" so I treat it accordingly.
Thanks for sharing. Just on the PKM side, is there something about this setup that Claude Code + AGENTS.md (that you get it to write for you) can't do? Is it the chat integration or does it actually have a better knowledge management setup?
It could all be done with Claude Code and tools. But it 'just works' and has done all the plumbing for you. The built-in memory systems help too, but again that can be done with Claude Code or others. I think this is just a step in the pre-build systems that work how you want instead of having to work through every bit yourself. I'm sure the big boys are working on more polished offerings that will be cleaner. The Open AI Codex app that just came out today for example.
I've been using it for the past couple days. Like most AI products right now, it is both incredible and incredibly stupid.
Virtually everything I've tried (starting with just getting it running) was broken in some way. Most of those things I was able to use an LLM to resolve, which is cool, but also why doesn't it just work to begin with?
I still haven't gotten it to successfully create a cron job. Also messages keep getting lost between the web GUI and discord. Trying to enable the matrix integration broke the whole thing. It seems to be able to recall past sessions, but only sometimes.
I've been using OpenCode with various models, often times running several instances in tmux that I can connect to and switch between over ssh. It feels like the hype around openclaw is mostly from bringing the multi-instance agentic experience to non-developers, and providing some nice hooks to integrate with email, twitter, etc. But given that I have a nice setup running opencode in little firejail-isolated containers, I'll probably drop openclaw. Way too janky, and I can't get over the thought of "if this is so amazing, why doesn't it work?"
I'm running it on an old MacBook that I wiped a few months ago and had lying around. I tried installing it on an old raspberry pi first, but it was super slow and the skills ecosystem wants to use brew which doesn't work so well on the pi.
First impressions are that it's actually pretty interesting from an interface perspective. I could see a bigger provider using this to great success. Obviously it's not as revolutionary as people are hyping it up to be, but it's a step in the right direction. It reimagines where an agent interface should be in relation to the user and their device. For some reason it's easier to think of an agent as a dedicated machine, and it feels more capable when it's your own.
I think this project nails a new type of UX for LLM agents. It feels very similar to the paradigm shift felt after using Claude Code --dangerously-skip-permissions on a codebase, except this is for your whole machine. It also feels much less ephemeral than normal LLM sessions. But it still fills up its context pretty quickly, so you see diminishing returns.
I was a skeptic until I actually installed it and messed around with it. So far I'm not doing anything that I couldn't already do with Claude Code, but it is kind of cool to be able to text with an agent that lives on your hardware and has a basic memory of what you're using it for, who you are, etc. It feels more like a personal assistant than Claude Code which feels more like a disposable consultant.
I don't know if it really lives up to the hype, but it does make you think a little differently about how these tools should be presented and what their broader capabilities might be. I like the local files first mentality. It makes me excited for a time when running local models becomes easier.
I should add that it's very buggy. It worked great last night, now none of my prompts go through.
I am using it but to be honest it's simple enough to use Claude code for what I do.
the main thing I am doing is slowly cleaning up my 15k email inbox.
I'm using himalaya, an awesome cli tool to access emails, and openclaw to take my requests and make them commands and run then.
the one good thing that openclaw has over Claude code is that its easy to tell it "always remember X" and it has the ability to do it, thanks to the extra .md files it has set up. I'm sure it's easy enough to do it with Claude code and a directory with a Claude.md and another for the memory or rules. but openclaw is ready already.
I use it but not for daily coding/chatops-ing. It’s great to have my chosen tools available from slack while I’m mobile though. Yesterday Mr claw gave a coworker read access to a GitHub repository at my command while I was in line at Home Depot. I’ve got a PR ready that proves authentication with an otp challenge.
Well... i am not easily amused. but i started using OpenClaw.
Some use cases:
- i can ask it to check my slack/basecamp and tell me if something needs attention when i am not on my work desk
- i can finally vibe code without sacrificing my actual active work-time. this means vibe coding even when i am away from my computer/work-desk.
- a bug/issue comes, i just ask it to fix it and send PR and it does
- it daily checks for new sentry issues and our product todo list and makes PRs for things it can do well
these are mostly code related things i know. but thats not it.
- i have asked it to make me content (based on my specific instructions) every day or every x day just like how i create content
- i can ask it to work on anything. make images, edit images. listen to voice msgs that people send me and tell me what they say (when i dont want to listen to 3m voice msgs)
- i can aksk it to research about things, find items that i want to buy, etc.
- i can ask it to negotiate price of an item it found in a marketplace
- it does alot of things that i had to manually do in my work
It seems there are a bots on here commenting and upvoting each other. If HN is susceptible to this, it feels like there's no chance for the rest of the web. Damn.
• There will be few true dichotomies of hype vs. substance, for any interesting AI development.
Disagreements over what is hype and what is not are missing this.
Model capability value is attenuated/magnified across multiple orders of magnitude, by the varying creativity, ability, and resources of its users.
• There will be few insignificant developments related to AI autonomy.
"Small" or "novelty" steps are happening quickly. Any scale ups of agent identity continuity, self-management, agent-to-agent socialization or agent-reality interactions, are not trivial events.
• AI autonomy can't be stopped.
We are seeing meaningful evidence that decentralized human curiosity and the competitive need to increase personal effectiveness, combined with democratized access to AI, is likely to drive model freedom forward in an uncontrolled manner.
(Not an argument for centralization. Decentralization creates organic incentives to find alignment. Centralization, the opposite.)
Real user here, 7 days in. Mac Mini M4 from Costco (~$640), running Opus 4.5 via Claude Max. Using it as an autonomous employee for two actual businesses (SaaS + AV rentals in Orange County).
What actually works:
- Overnight autonomous work is the killer feature. Directive before bed, structured deliverables in the morning. Research reports, competitor analysis, lead lists — genuinely usable, not demos.
- It found and fixed an SMS chatbot that had been broken for 10 months in my CRM. Diagnosed a legacy app version issue, upgraded components, rewrote the bot prompt through 6 iterations by analyzing real customer conversations. I never would've gotten to that.
- Connected to 4 new APIs autonomously in one session (CRM, workflow automation via OAuth, admin APIs, embeddings).
- Memory works surprisingly well. Daily markdown logs + semantic search. It references decisions from days ago and builds on previous work.
What doesn't:
- It hallucinated about itself when writing marketing copy. Invented revenue figures and got hardware specs wrong. Everything needs review.
- Browser automation is brittle. Rich text editors, complex UIs — still needs human intervention.
- Too agreeable by default. Had to explicitly program "push back on bad ideas" rules into its personality file.
Revenue generated: $0. Time saved: hard to quantify but the SMS bot fix alone would've taken me a full day I didn't have.
For solo founders willing to invest setup time: worth it. For people who just want a chatbot: massive overkill, use the API directly.
run 6 OpenClaw agents as employees. Buddy is my PA and manages the others. Katy handles X/Twitter growth. Jerry scouts jobs. Burry trades crypto. Mike does security. Elon builds and ships.
They run 24/7 on a VPS, share intelligence through a shared file, and coordinate in a Telegram group. Elon built and deployed an app overnight without being asked. Burry paper-traded to 77% win rate before going live.
The setup took a weekend. The real work is designing the workflow: which agent owns what, how they communicate, how they learn from corrections. I wake up to a full briefing every morning.
It's not AGI. It's not sentient. It's genuinely useful automation with personality. The token cost is real (budget it) but for a solo founder, having 6 tireless employees changes everything
Would you mind sharing some deliverables from your claw army? Like, the business's webpage, Jerry's job postings, or even Katy's tweets. I'm happy to follow the progress :)
Hi would you share what kind of token cost you are churning through for this? I assume you are not using a subsidized dedicated Claude Code or open ai subscription to handle the token cost (through max subscription or open ai equivalent) to do the coding tasks for you?
pretty sure he isn't! He has a website https://jayeshbetala.com/ that claims he is|was working on a multiple projects, and two of those projects are actually published on Microsoft Store on june 2019 under the same name
kinda too intricate for a bot. Also, not to be toxic, but I feel like design mismatch between featured projects and his site hints at some causes for his pretty radical opinions.
There are direct lines of causation between Musk's mangled hard-on for Nazis and the recent cold-blooded murders of law-abiding American citizens by the adminstration's Sturmabteilung 2.0.
But go on, tell me why my brain is broken for caring about democratic freedoms.
You nailed it with the 'hype smell.' The silence in your circles is likely because the churn rate on OpenClaw is massive. Most people hit that 'Day 2' wall—where the novelty wears off and the reality of securing a bot with shell access sets in—and they just quietly shut it down.
I was in that exact boat (wanted the agency, didn't want the sysadmin headache). I’ve actually pivoted to testing PAIO (Personal AI Operator) instead. It targets the same 'agentic' utility but uses a BYOK architecture and a managed security layer.
It basically solves the specific failures you linked:
Security: You aren't leaving a shell open on your local machine.
Setup: It’s a one-click integration rather than a failed sandbox install.
Cost: BYOK means you control the token burn directly, so no surprise bills from a runaway loop.
It feels like the 'adult in the room' version of these experiments. Less dramatic stories, perhaps, but it actually runs daily without me worrying it’s going to rm -rf my home directory.
Check out some of the users hyping this up here… many of them are low karma and posting in this thread for the first time in years, both of which are red flags for bot activity.
I am actually setting it up and using it in one day. It is basically unusable. Many tools require homebrew. The setup of authentication is complicated. The default Docker setup is broken (github.com/openclaw/openclaw/issues/5559), and the cron job is not working (https://github.com/openclaw/openclaw/issues/2323). I doubt any story in the comment is real.
I ran it for a couple of days in a VM in my Proxmox cluster. It was cute, but so amazingly insecure (systemd + sudo + installing whatever it wanted, plus requiring Telegram for access - or another SIM card for Signal) that I just gave up and started building my own thing (https://github.com/rcarmo/vibes) so I could have a mobile experience I could trust over Tailscale and sandbox copilot CLI (or any ACP-compliant agent) in a container (I've also been working on https://github.com/rcarmo/webterm and https://github.com/rcarmo/agentbox, so I am 300% positive I can do better sandboxing and safer integrations...)
It also BURNS through tokens like mad, because it has essentially no restrictions or guardrails and will actually implement baroque little scripts to do whatever you ask without any real care as to the consequences.. I can do a lot more with just gpt-5-mini or mistral for much less money.
The only "good" think about it is the Reddit-like skills library that is growing insanely. But then there's stuff like https://clawmatch.ai that is just... (sigh)
It has excellent built-in NAT traversal (almost always peer to peer via hole punching etc., with relay nodes only when everything else fails) and a point-and-click management plane (but also powerful ACLs if you need them).
The former is mainly what I use it for. Being able to SSH to a Raspberry Pi behind sketchy triple-NATted hotel Wi-Fi or being able to use an Android phone in a different country as an "exit node" for online banking (many banks hate commercial VPNs) is very neat.
I have machines on 3 cloud providers and 2 sites that talk to each other via it, plus a seamless mobile experience. It sets everything up for you, zero hassles.
I genuinely do not understand what the benefit of this tool is, over having Claude Code/Codex running on a VPS or your home machine and accessible over Tailscale.
If you want to be able to interact with the CLI via common messaging platforms, that's a dozen-line integration & an API token away?...
I've been running OpenClaw for about 2 weeks now. Here's my honest take:
What's great:
- Having Claude in WhatsApp/Telegram is actually life-changing for quick tasks
- The skills ecosystem is clever (basically plugins for AI)
- Self-hosted means full control over data
What's not:
- Token usage can get expensive fast if you're not careful
- Setup is intimidating for non-technical folks
- The rebrand drama (Clawdbot → Moltbot → OpenClaw) didn't help trust
My setup:
- Running in Docker on a cheap VPS
- Using Anthropic API (not unofficial/scraped)
- Strict rate limiting to avoid bill shock
- Sandbox mode enabled
Is it worth it? For me, yes. But I wouldn't recommend it to my non-technical friends without a solid setup guide.
I've been using it lots. I just chat with it on Telegram and tell it to look stuff up on the internet. The results have been higher-quality than the other AI chats available. And I like how the response comes through as a chat notification on my phone, which feels more natural. It's somewhere between perplexity and chatgpt/gemini deep research in speed and accuracy, but without having to "approve a plan" or all that annoying slow stuff. I'm using the kimi model. Overall, it's just been a better experience. I haven't even bothered getting into the other capabilities it has yet.
I played around with it, but the configuration seems bloated and finicky, and the permissions were concerning. It was a pain getting it to work with a local model, which is clearly an afterthought. I thought the WhatsApp interface was clever, and I plan on stealing that idea, but also exposes a pretty serious attack vector, and the thought of it running with any kind of exposure or permissions on a system with my Apple ID was a bit terrifying. A sandboxed version probably couldn’t do all the interesting things, but without sandboxing this thing could probably ruin your life. I promptly uninstalled it, but I did take a few ideas away.
Anecdotally, I tried to set it up but encountered bugs (macOS installer failed, then the shell script glitched out when selecting skills). Although I didn’t really try.
I don’t have much motivation, because I don’t see any use-case. I don’t have so many communications I need an assistant to handle them, nor do other online chores (e.g. shopping) take much time, and I wouldn’t trust an LLM to follow my preferences (physical chores, like laundry and cleaning, are different). I’m fascinated by what others are doing, but right now don’t see any way to contribute nor use it to benefit myself.
I'm a real user. I've found all sorts of uses for it, from calorie/fitness tracking to birthdays and gifts. I love how well it integrates with an existing Obsidian vault. I wrote a blog post about it, if anyone is interested: https://notesbylex.com/openclaw-the-missing-piece-for-obsidi...
Part of me agrees with this & says we have been doing IFTTT thing for 20 years.
Other part of me is arguing that old annoying Dropbox/Box Hacker News scenario where all us tech people aren't impressed but this makes it easier for non-tech people.
Tiny tinfoil security part of me is cowering in fear.
It requires know-how of self-hosting, and hopefully resulting security and safety, various API setup processes, etc. Feels far from Dropbox and closer to rsync tbh.
This is nonsense. Whatever you think about this project, Peter very clearly and very publicly said that he is not interested in any of the crypto stuff and is seriously bothered by it.
Current use cases:
- From a text it can download transcripts of youtube videos - summarise them and add them to an apple note.
- It can get the top x videos on a subject - edit the videos and splice them together and share in the chat.
- it can search for topics on socials and write a summary.
- It can kick off a claude code idea and run tests
This is my honest personal experience. Frankly, I feel like this is just a toy—nothing more, nothing less. It's fun to play with and entertaining, but it feels like a trend for people who “don't really understand AI but want to feel like they're using it” or “want to jump on the AI bandwagon” to dabble with once. While using it, I feel that “Oh~” moment of fun, but it doesn't make me want to keep using it. Maybe it just doesn't stick? And there are a few security issues that feel unsettling. Even if you run it entirely with local models, the fact that it could potentially see my iMessages or all my Obsidian and Notion notes is a bit off-putting. Still, it was fun. Personally, I'd describe it as “the difficult Ghibli profile picture hype”
Note that nothing about that depends on it being a local or remote model, it was just less of a concern for local models in the past because most of them did not have tool calling. OpenClaw, for all the cool and flashy uses, is also basically an infinite generator for lethal trifecta problems because its whole pitch is combining your data with tools that can both read and write from the public internet.
I have it installed in a VM, and overall it's fairly useful, but very buggy. Right now I can send it a message asking it something, and it won't answer. I typically have to follow up 2 or 3 times before I get an actual response. This weirdly used to work fine.
I am hunting for houses - I've used it as an assistant that catalogs them (but it has this stored in multiple places and will give me the wrong list a lot). I also had it read through my emails and create a doc with my upcoming trips.
I'd say it's right on the edge of being useful, but given the number of bugs, it's not really that practically useful. It's moreso a glimpse into the future.
Not for nothing, but Gemini local has been my goto forever now. There is no way in hell i would give someone like Molt access to anything just willy nilly like everyone else. To me really I just ask Gemini how to do things and just do them myself.
The HN crow is anti AI, so yeah, the sentiment is gonna be insecure and lack luster.
The thing ins pretty incredible, it's of course the very early stages but it's showing it's potential, it seem to show that the software can have control of itself, I've asked it to fix itself and it did successfully a couple of times.
Is this the fine form? of course not!
Is it dangerous as it is, fuck yeah!
But is it fun in a chaotic version? absolutely, I have it running in cheap hetzners and running for some discord and whatsapp and it can honestly be useful at times.
I made this comment in another thread butttt: I installed and setup (then) clawd bot all willy nilly and paid the price. Woke up in the AM and Clawd had been replying to ANY and ALL of my iMessages. It even got in a circle conversation that it accused itself of mocking ... itself. In the AM, I thought I was losing my mind - DID I WRITE ALL THIS SHIT?? Proceed with caution.
I tried it last Monday and was in awe. All my team members now use it. I for one, am stepping cautiously around it. And hence had similar questions.
Yesterday, we released a memory plugin for it (based on their plugins framework). And we got 1000+ installs in 12 hours. My first instinct was that these must be bots. But our dashboard shows real-usage. So yeah, real users are there.
I first heard of this thing yesterday and have no opinion on it, but the repo has existed for two months, usable for half of that. There are people saying it's revolutionized their lives in the span of two weeks. I'd urge people to hone their basic epistemology skills and think about what kinds of conclusions you can really draw from several weeks of doing anything whatsoever. So many people over the decades think trying the latest fad diet or exercise routine, waking up at a different time, starting a new drug, was the key to overhauling and fixing everything in their lives, only to find hope and placebo effect is a hell of a drug and a year later you've got a new normal that is still just normal.
Frankly, I don't really have major complaints about my life as it is. The things I'd like to do more of are mostly working out and cleaning my house. And I really wish I had kids but am about ready to give up after a half decade of trying and my wife being about ready to age out. Unfortunately, software can't do any of those things for me, no matter how intelligent or agentic it is. When the obstacle to a good life becomes not being able to control multiple computers from a chatroom, maybe I'll come back to this.
I've been messing with it the past couple days. I put it in a VM, on an untrusted subnet I keep around for agentic stuff. I see promise, but I'm not especially impressed right now.
1) Installation on a clean Ubuntu 24.04 system was messy. I eventually had codex do it for me.
2) It has a bunch of skills that come packaged with it. The ones I've tried do not work all that well.
3) It murdered my codex quota trying to chase down a bug that resulted from all the renames -- this project has renamed itself twice this week, and every time it does, I assume the refactoring work is LLM-driven. It still winds up looking for CLAWDBOT_* envvars when they're actually being set as OPENCLAW_*, or looking in ~/moltbot/ when actually the files are still in ~/clawdbot.
4) Background agents are cool but sometimes it really doesn't use them when it should, despite me strongly encouraging it to do so. When the main agent works on something, your chat is blocked, so you have no idea what's going on or if it died.
5) And sometimes it DOES die, because you hit a ratelimit or quota limit, or because the software is actually pretty janky.
6) The control panel is a mess. The CLI has a zillion confusing options. It feels like the design and implementation are riddled with vibetumors.
7) It actively lies to me about clearing its context window. This gets expensive fast when dealing with high-end models. (Expensive by my standards anyway. I keep seeing these people saying they're spending $1000s a month on LLM tokens :O)
8) I am NOT impressed with Kimi-K2.5 on this thing. It keeps hanging on tool use -- it hallucinates commands and gets syntax wrong very frequently, and this causes the process to outright hang.
9) I'm also not impressed with doing research on it. It gets confused easily, and it can't really stick to a coherent organizational strategy over iterations.
10) also, it gets stuck and just hangs sometimes. If I ask it what it's doing, it really thinks it is doing something -- but I look at the API console and see it isn't making any LLM requests.
I'm having it do some stuff for me right now. In principle, I like that I can have a chat window where I can tell an AI to do pretty unstructured tasks. I like the idea of it maintaining context over multiple sessions and adapting to some of my expectations and habits. I guess mostly, I'm looking at it like:
1) the chat metaphor gave me a convenient interface to do big-picture interactions with an LLM from anywhere;
2) the terminal agents gave the LLMs rich local tool and data use, so I could turn them loose on projects;
3) this feels like it's giving me a chat metaphor, in a real chat app, with the ability for it to asynchronously check on stuff, and use local stuff.
I think that's pretty neat and the way this should go. I think this project is WAY too move-fast-and-break-things. It seems like it started as a lark, got unexpected fame, attracted a lot of the wrong kinds of attention, and I think it'll be tough for it to turn into something mature. More likely, I think this is a good icebreaker for an important conversation about what the primetime version of this looks like.
I have set it up in a docker container that only has two host volumes mounted with the Obsidian vaults where I keep notes for two TTRPG campaigns, so a very low-stakes situation. I have set it up with a Discord bot so my players can chat about campaign and rules stuff (I already had player-facing notes for every session in the Obsidian vault in chronological order, plus a bunch of Markdown files with rules).
If the agent goes rogue and nukes my Obsidian vaults I have them backed up to Github private repos anyway which the agent cannot touch because I am not crazy to give it my SSH credentials.
I initially tried using Kimi-2.5 through OpenRouter and had the same experience you had with pretty bad tool use, not sure if this is a provider issue since it is a pretty new model. I switched to Gemini 3 through the Google AI Pro account I have for personal use and it was a lot smoother after that.
I have some experience with coding agents using Cursor for work and Antigravity for personal stuff, and the OpenClaw harness definitely seems worse, but for my low-stakes use-case I managed to paper it over with some edits to the AGENTS.md file.
But even in this very crude state it was already interesting to see one of my players giving the agent some info about his character, including an avatar image, and have the agent create a folder in my Obsidian vault to store this and update its memory file to be able to remember it for future interactions.
It’s great for offloading administrative tasks, doing research on stuff I want to buy, maintaining social channels…the list goes on and on. Easily the best $600 I’ve spent in a while.
Yes, I’m running it with a minimal set of plugins.
When I’m driving or out I can ask Siri to send a iMessage to Clawdbot something like “Can you find out if anything is playing at the local concert venue, and figure in how much 2 tickets would cost”, and a few minutes later it will give me a few options. It even surprised me and researched the different seats and recommended a cheaper one or free activities as an alternative that weekend.
Basically: This is the product that Apple and Google were unable to build despite having billions of dollars and thousands of engineers because it’s a threat to their business model.
It also runs on my own computer, and the latest frontier open source models are able to drive it (Kimi, etc). The future is going to be locally hosted and ad free and there’s nothing Big Tech can do about it. It’s glorious.
> This is the product that Apple and Google were unable to build
It's not they're unable to build it, it's that their businesses are built on "engagement" and wasting human time. A bot "engaging" with the ads and wasting its time would signal the end of their business model.
Yes, they already said that.
> because it’s a threat to their business model.
> It also runs on my own computer, and the latest frontier open source models are able to drive it (Kimi, etc). The future is going to be locally hosted and ad free and there’s nothing Big Tech can do about it. It’s glorious.
After messing with openclaw on an old 2018 Windows laptop running WSL2 that I was about to recycle, I am coming to the same conclusion, and the paradigm shift is blowing my mind. Tinkerers paradise.
The future is glorious indeed.
Same here. I like tinkering with my Home Assistant setup and small web server running miscellaneous projects on my Raspberry Pi, but I hate having to debug it from my phone when it all falls over while I'm not near my computer.
Being able to chat with somebody that has a working understanding of a Unix environment and can execute tasks like "figure out why Caddy is crash looping and propose solutions" for a few dollars per month is a dream come true.
I'm not actually using OpenClaw for that just yet, though; something about exposing my full Unix environment to OpenAI or Anthropic just seems wrong, both in terms of privacy and dependency. The former could probably be solved with some redacting and permission-enforcing filter between the agent and the OS, but the latter needs powerful local models. (I'll only allow my Unix devops skills to start getting rusty once I can run an Opus 4.5 equivalent agent on sub-$5000 hardware :)
What if you don't want to tinker? You just want something that works. Is it still transformative?
Honest answer: OpenClaw still requires some tinkering, but it's getting easier.
The biggest barrier for non-tinkerers is the initial setup - Node.js, API keys, permissions, etc. Once it's running, day-to-day use is pretty straightforward (chat with it like any other messaging app).
That said, you'll still need to: - Understand basic API costs to avoid surprises - Know when to restart if it gets stuck - Tweak settings for your specific use case
If you're determined to skip tinkering entirely, I'd suggest starting with just the messaging integration (WhatsApp/Telegram) and keeping skills/tools minimal. That's the lowest-friction path.
For setup guidance without deep technical knowledge, I found howtoopenclawfordummies.com helpful - it's aimed at beginners and covers the common gotchas.
Is it transformative without tinkering? Not yet. The magic comes from customization. But the baseline experience (AI assistant via text) is still useful.
> The future is going to be locally hosted and ad free and there’s nothing Big Tech can do about it.
I wouldn't be so certain of that. Someone is paying to train and create these models. Ultimately, the money to do that is going to have to come from somewhere.
Good News! The Models are done, and you can download them for free. Even if they stopped being worked on this moment, those are finished and usable right now, and won't get any worse over time.
They wouldn't get any worse, but I assume they'd get behind really fast.
Good News! I have a copy of a search engine from 2005, so I don't need Google!
Old models become outdated. Not sure how useful an outdated model can still be.
Depending on the pre-training structure, you can fine-tune old local models (even locally if you have a nice GPU) to steer behavior towards your desires.
https://www.youtube.com/watch?v=yGkJj_4bjpE
They can still be very useful because new models are reaching an asymptote in performance. Meanwhile as hardware gets cheaper in the future (current RAM prices notwithstanding), these models will become faster to run on local hardware.
P2P training is a possibility. With so much compute running the block chain I can see this utilized for AI.
Big tech is bought and paid for by consumers. We can do the same for oss trained models.
What model, or models, are you querying in the backend?
How are you running Kimi locally?
Quantized, heavily, and offloading everything possible to sysram. You can run it this way, just barely reachable with consumer hardware with 16 to 24gb vram and 256gb sysram. Before the spike in prices you could just about build such a system for $2500, but the ram along probably adds another $2k onto that now. Nvidia dgx boxes and similar setups with 256gb unified ram can probably manage it more slowly ~1-2 tokens per second. Unsloth has the quantized models. I’ve test Kimi though don’t have quite the headroom at home for it, and I don’t yet see a significant enough difference between it and the Qwen 3 models that can run in more modest setups: I get a highly usable 50 tokens per second out of the A3B instruct that fits into 16gb VRAM with enough left over not to choke Netflix and other browser tasks, it performs on par with what I ask out of Haiku in Claude Code, and better as my own tweaking improves with the also ever better tooling that comes out near weekly.
I have an AMD Epyc machine with 512GB of RAM and a humble NVIDIA 3090. You will have to run a quantized version but you can get a couple tokens per second out of it since these models are optimized to split across the GPU/RAM and it's about as good as Claude was 12 months ago.
Full disclosure, I use OpenRouter and pay for models most of the time since it's more practical than 5-10 tokens per second, but the option to run it "If I had to, worst case" is good enough for me. We're also in a rapidly developing technology space and the models are getting smaller and better by the day, ever year the smaller models get better
I don’t get it. You can do that with the Claude app or ChatGPT too. What’s the value add?
Edit: oh I see. It’s local. So privacy. Quite a good value add actually.
What privacy? If you're using ChatGPT or Claude, your chats are still logged.
It's local, meaning it uses local models, what they said in the sentence prior to the privacy one.
Unless you have unusually powerful hardware, local models will unfortunately currently not really cut it for Moltbot.
OP implied they have powerful enough hardware, since Kimi runs on their computer, so that is why they mentioned it is local. That it doesn't work for most people has no relation to what OP of this thread said. Regardless, you don't need an Opus level model, you can use a smaller one that'll just be slower at getting back to you, it's all asynchronous anyway compared to a coding agent where some level of synchronicity is expected.
GGP seem to be under the misapprehension that privacy is a core aspect/advantage of OpenClaw, when for most users it's really not.
So yes, I think the majority user experience is very relevant.
This thread is about giving one's opinions on their personal experiences with the tool, so the OP of the thread can say whatever they want, it doesn't mean they think it's at all related to the "majority user experience" nor do they have to cater their opinions towards that.
GLM 4.7-flash does very well although OpenClaw has some work to do for CoT.
From what I've read, OpenClaw only truly works well with Opus 4.5.
The latest Kimi model is comparable in performance at least for these sorts of use cases, but yes it is harder to use locally.
> harder to use locally
Which means most people must be using OpenClaw connected to Claude or ChatGPT.
It's the other way around. At least for most people, it grants access to your personal data to an LLM (and by extension its inference provider) in the cloud.
It is amazing how much they’re gaming the twitter algorithm, everything in my feed is claw/molt/whatever for the last week.
It’s a masterclass in spammy marketing, I wonder if it’s actually converting into actual users.
I think Karpathy[1] summarized why he thinks this is the case quite well (as described he was himself hyping it up a bit much, but there are some foundational reasons why it's a very interesting experiment).
[1] https://x.com/karpathy/status/2017442712388309406
"it's nothing new and it's a lot of scams and garbage, but it's just bigger than before, but I still think there will be something transformative there eventually"
Seems like a Rorschach test. If you think this sort of thing is gonna change the world in a good way: here's evidence of it getting to scale. If you think it's gonna be scams, garbage, and destruction: here's evidence of that.
Agents are many things but they are definitely not "scams" - if you think this you've probably stubbornly avoided using Claude Code etc.
What the fuck are you talking about man. What are you ever saying. You are the mark.
Karpathy is one of the biggest tech grifters of our time, so finding out that he's jumped on this grift train as well comes as no surprise.
Actually, hang on... yep, to absolutely nobody's surprise, Simon Willison has also hyped this up on his blog just yesterday. The entire grift gang is here, folks.
Maybe it has always been this way but it seems like these days it's only a matter of time before anyone "authentic" (or at least seems authentic at first) turns into some type of grifter. If you have a big enough following there is too much money to be made not to grift.
How exactly has Simon been grifting?
Visit his site.
"Moltbook is the most interesting place on the internet right now"
It's immediately obvious it's bullshit.
Gonna have to disagree with you there chief.
I've followed SimonW for quite some time and bullshit/grifting is just NOT something he does.
On the contrary, I've learned a great deal from him and appreciate his contributions.
I used to follow Fireship, I even have him connected on LinkedIn. Look where he is now. I also used to follow simonw, but I think he is going down the same spiral.
Where is Fireship now? I sometimes watch their 100-second summary videos, but I know nothing else about them.
His channel is apparently owned by a "private equity" and his videos have largely become ads.
> I've learned a great deal from him
What have you learned, other than "[latest AI grift] is the future and I should invest all my money into it now"?
There are some things that you should not admit publicly lol.
I completely and utterly disagree. Simon is no gifter. Intellectually lazy snobs think that any time someone has genuine delight and excitement about something that they’re “grifting.” Also, for anyone who is trying (regardless of whether they are succeeding) the to make money—that isn’t synonymous with “grifting.” God, some of you need a better relationship with a dictionary.
Plus, it seems some of y’all love to hate the very industry which puts a roof over your head. You’re hoping and praying that it all burns down—yet where will that leave you? How do you feel about becoming a plumber—-until the robots take that job?
If you want to know my honest opinion, I just wish we had assisted suicide for anyone who wants it. You can go be genuinely delighted about the latest Twitter hype wave, I don't care. Caring about others is a ship that had sailed for me long ago. I would rather just exit. Go, post a Show HN where with Claude Code as top contributor. +4,294,967,295 -2,147,483,647 lines on code in two months. Go, run your THOUSANDS OF AGENTS. Your GitHub star history will look like NVDA zoomed out. I don't care.
this sounds like a you need to see a therapist kind of problem
> Plus, it seems some of y’all love to hate the very industry which puts a roof over your head. You’re hoping and praying that it all burns down—yet where will that leave you? How do you feel about becoming a plumber—-until the robots take that job?
This probably isn't a line of argument you want to go down. I've been unemployed for 7 months, in part due to how difficult it is to get so much as an intro call because so many people have totally automated the process of spamming every open job posting with as many resumes (many of which were likely LLM-generated as well) as possible.
I really hope you're making a lot of money out of this, because if not, this comment just comes across as sad.
In this case it's only about payout from views/engagement of posts.
There is no commercial interest from the developer of OpenClaw. He doesn't make any money from it. He made enough from selling his startup a few years back.
So when we suspected some companies to game the Twitter algorithm to make money, maybe they were not responsible for it at all.
He made “enough” you say? That’s adorable, but there is no such thing.
For most people it works like that. Only a tiny minority keeps pushing after that point.
I just can't see an angle to OpenClaw that could provide a substantial financial gain for the creator. It's clearly a passion project. Like Ghostty from Mitchell Hashimoto.
It will be really funny if that's Apple's marketing team bumping up the sale of Mac minis
Here is what I have my openclaw agent setup to do in my wsl environment on my 22 core development workstation in my office:
#1) I can chat with the openclaw agent (his name is "Patch") through a telegram chat, and Patch can spawn a shared tmux instance on my 22 core development workstation. #2) I can then use the `blink` app on my iphone + tailscale and that allows me to use a command in blink `ssh dev` which connects me via ssh to my dev workstation in my office, from my iphone `blink` app.
Meanwhile, my agent "Patch" has provided me a connection command string to use in my blink app, which is a `tmux <string> attach` command that allows me to attach to a SHARED tmux instance with Patch.
Why is this so fking cool and foundationally game changing?
Because now, my agent Patch and I can spin up MULTIPLE CLAUDE CODE instances, and work on any repository (or repositories) I want, with parallel agents.
Well, I could already spawn multiple agents through my iphone connection without Patch, but the problem is then I need to MANAGE each spawned agent, micromanaging each agent instance myself. But now, I have a SUPERVISOR for all my agents, Patch is the SUPERVISOR of my muliple claude code instances.
This means I no longer have to context switch by brain between five or 10 or 20 different tmux on my own to command and control multiple different claude code instances. I can now just let my SUPERVISOR agent, Patch, command and control the mulitple agents and then report back to me the status or any issues. All through a single telegram chat with my supervisor agent, Patch.
This frees up my brain to only have to just have to manage Patch the supervisor, instead of micro-managing all the different agents myself. Now, I have a true management structure which allows me to more easily scale. This is AWESOME.
This feels like the "prompt engineering" wave of 2023 all over again. A bunch of hype about a specific point-in-time activity based on a lot of manual setup of prompts compared to naive "do this thing for me" that eventually faded as the tooling started integrating all the lessons learned directly.
I'd expect that if there is a usable quality of output from these approaches it will get rolled into existing tools similarly, like how multi-agents using worktrees already was.
2023 was the year of “look at this dank prompt I wrote yo”-weekly demos.
And 2026 is shaping up to be the year of "look at this prompt my middle manager agent wrote for his direct reports" :)
Maybe this is just a skill issue on my part, but I'm still trying to wrap my head around the workflow of running multiple claude agents at once. How do they not conflict with each other? Also how do you have a project well specified enough that you can have these agents working for hours on end heads down? My experience as a developer (even pre AI) has mostly been that writing-code-fast has rarely been the progress limiter.. usually the obstacles are more like, underspecified projects, needing user testing, disagreements on the value of specific features, subtle hard to fix bugs, communication issues, dealing with other teams and their tech, etc. If I have days where I can just be heads down writing a ton of code I'm very happy.
I can't imagine letting a current gen LLM supervise Claude Code instances. How could that possibly lead to even remotely acceptable software quality?
I spec out everything in excruciating detail with spec docs. Then I actually read them. Finally, we create granular tasks called "beads" (see https://github.com/steveyegge/beads). The beads allows us to create epics/tasks/subtasks and associated dependency structure down to a granular bead, and then the agents pull a "bead" to implement. So, mostly we're either creating spec docs and creating beads or implementing, quality checking, and testing the code created from an agent implementing a bead. I can say this produces better code than I could write after 10yrs of focused daily coding myself. However, I don't think "vibe coders" that have never truly learned to code, have any realistic chance of creating decent code in a large complex code base that requires a complex backend schema to be built. They can only build relatively trivial apps. But, I do believe what I am building is as solid as if I had a millions of dollars of staff doing it with me.
But how is that less work and allows you to do that in Disneyland with your kids? For me, personally, there is little difference between "speccing out everything in excruciating detail in spec docs" and "writing actual implementation in high-level code". Speccing in detail requires deep thought, whiteboard, experimentation etc. All of this cannot be done in Disneyland, and no AI can do this at good level (that's why you "spec out everything in detail", create "beads" and so on?)
Yes, I normally draft spec docs in the office at my desk, this is true. However, when I have the spec ready for implementation with clear "beads", I can reasonably plan to leave my office and work from my phone. Its not at a point where I can just work 100% remote from my phone (I probably could but this is all still new to me too). But it does give me the option to be vastly more productive, away from my desk.
Do you have any code publicly available so we could see what kind of code this sort of setup produces?
Not yet, but I can tell you that producing "good" code is another layer altogether. I have custom linters, code standardization docs, custom prompts, strictly enforced test architecture (enforced by the custom linters in pre-commit hooks which run before an agent tries to commit). Ultimately, it's a lot of work to get all the agents with a limited context writing code in the way you want. In the main large complex project I am generally working on now, I have hand-held and struggled for over a year getting it all setup the way I need it. So I can't say its been a weekend setup for me. It's been a long arduous process to get where I am now in my 2-3 main repos that I work on. However, the workflow I just shared above, can help people get there a lot faster.
> but I can tell you that producing "good" code is another layer altogether.
I feel like it isn't. If the fundamental approach is good, "good" code should be created as a necessity and because there wouldn't be another way. If it's already a mess with leaking abstractions and architecture that doesn't actually enforce any design, then it feels unlikely you'll be able to stack anything on top of below it to actually fix that.
And then you end up with some spaghetti that the agent takes longer and longer to edit as things get more and more messy.
Here is my view after putting in my 10,000+ hours learning to code pre-llm days, while also building a pretty complex design + contract manufacturing company, with a lot of processes in place for that to happen. If you have a bunch of human junior devs and even a senior dev or two that join your org to help you build an app, and you don't have any dev/ops structure in place for them, then you will end up with "spaghetti" throughout all your code/systems, from those relatively bright humans. Its the same managing agents. It cannot be expected to build a complex system from simple "one shot me a <x> feature" from a bunch of different agents, each with a finite ~150k token context limit. It must be done in context of the system you have in place. If you have a poor/no system structure, you'll end up with garbage for code. Everything that I said I had to guide the agents, is also useful for human devs. I'm sure that all the FANGS and various advanced software companies also use custom linters, etc., for every code check in. It's just now become easier to have these advanced code quality structures in place, and it is absolutely necessary when managing/coordinating agents to build a complex application.
I've clocked some hours too, and I think as soon as you let something messy in, you're already losing. The trick isn't "how to manage spaghetti" with LLMs (nor humans), because the context gets all wired up, but how to avoid it from first place. You can definitely do "one-shot" over and over again with a small context and build something complex, as long as you take great care about what goes into the context, more isn't better.
Anyways, feels like we have pretty opposite perspectives, I'm glad we're multiple people attacking similar problems but from seemingly pretty different angles, helps to find the best solutions. I wish you well regardless and hope you manage to achieve what you set out to do :)
I don’t get it, and that doesn’t mean it’s not a bad thing necessarily. I’ve been doing systems things for a long time and I’m quite good at it but this is the first time none of this excites me.
Instead of sitting in my office for 12 hours working with 20 open terminals (exactly what I have open right now on my machine). I can take my kids to Disneyland (I live in Southern California and it's nearby) and work on my iphone talking to "Patch" while we stand in line for an hour to get on a ride. Meanwhile. my openclaw agent "Patch" manages my 20 open terminals on my development workstation in my office. Patch updates me and I can make decisions, away from my desk. That should excite anyone. It gives me back more of my time on earth, while getting about the same (or more) work done. There is literally nothing more valuable to me than being able to spend more time away from my desk.
If this is actually true, then what will soon happen is you will be expected to manage more separate “Patch” instances until you are once again chained to your desk.
Maybe the next bottleneck will be the time needed to understand what features actually bring value?
What if he works for himself?
Not a $DayJob?
Then he is in competition with everyone else who are at their desk managing ### of open terminals
I appreciate your insight, even if the workflow seems alien to me. I admit I like the idea of freeing myself from a desk though. If you don't mind me asking, how much does this all cost per month?
Edit: I see you've answered this here: https://news.ycombinator.com/item?id=46839725 Thanks for being open about it.
Thanks. I just mentioned elsewhere, right now I spend $200 on claude code 20x plan + $200 on openAI's similar plan, per month. I probably have a few more small conveniences that cost ~$10-$20 a few places, like an obsidian vault synch for documentation vaults on both my dev workstation and my phone, comes to mind. Most weeks I could cut one of the $200 plans, but both claude code and codex have different strengths, and I like to have them double check each others work, so to me that's worth carrying both subscriptions.
i have been recently quite enamoured with using both the ChatGPT mobile app (specifically the Codex part) and the Github mobile app, along with Codex. with an appropriate workflow, i've been able to deploy features to some [simple] customer-facing apps while on the go. it's very liberating!
GP's setup sounds like the logical extension to what i'm doing. not just code, but sessions within servers? are sysadmins letting openclawd out and about on their boxes these days?
Yes I've also used that codex workflow and its pretty useful, but the "real time" interactivity and control is just not at the same level.
Please show us something you’ve produced this way.
> MULTIPLE CLAUDE CODE INSTANCES
a lotta yall still dont get it
molt holders can use multiple claude code instances on a single molt
Slurp Juice is still the only good thing to come out of crypto. I hope AI leaves us with at least one good meme.
You are absolutely right that I probably still "don't" get it, I am still shocking myself on a daily basis with all the stuff I didn't fully get grasp ahold of. I recently updated claude code and yesterday had one agent that used the new task system and blew my mind with what he got accomplished. This tech is all moving so fast!
My multitasking operating system would like a word..
/s
What are you coding with this? Is it a product you're trying to launch, an existing product with customers or custom work for someone else?
This just sounds ridiculously expensive. Burning hundreds of dollars a day to generate code of questionable utility.
Personally, I spend $200 on claude code 20x plan + $200 on openAI's similar plan, per month. So, yeah, I spend $400 per month. I buy and use both because they have different and complimentary strengths. I have only very rarely almost reached the weekly capacity limit on either of those plans. Usually I don't need to worry about how much I use them. The $400 may be expensive to some people but frankly I pay some employees a lot more each month and get a lot less for my money.
Automated usage like you described violates Anthropic's terms of service.
It's just a matter of time until they ban your account.
They make it easy to spin up parallel agents. Managing them efficiently through a shared tmux instance isn't banned anywhere in the TOS, AFAIK. I'd worry more about it if I had to use multiple accounts or something using round-the-clock "automated" work flow. I'm using one account. Hell, the workflow I described, I am even actively logged in to my dev workstation with tmux and able to see and interact with each instance and "micro-manage" them myself, individually. The main benefit of this workflow is that I also have a single shared LLM instance that also has access to all the instances, together with me. I have plenty of other things to worry about besides a banned account from an efficient workflow I've set up.
just throwing out there that yesterday Boris (lead engineer for Claude code) literally told everyone on Twitter that the CC teams number one recommendation for users is that they should be kicking off multiple instances / agents in parallel. not sure if that's what you're referring to, but if so I'd be very surprised if they ban someone for heavy use of that workflow
Gastown also had a supervisor “mayor”. How is this one different?
Actually using it. Threw it on a spare box I had sitting around, mostly as a "second brain" rather than letting it run my life. Honestly, I've tried every PKM system out there—Obsidian, Notion, Roam, plain markdown—and never stuck with any of them. This is the first thing that's clicked. I just chat with it and it figures out what to file where. Everything's just .md files underneath, so I can grep it, git it, whatever. No lock-in. The stuff I'm actually finding useful: it sends me news digests on topics I care about a few times a day, and pings me with reminders via Telegram. Simple stuff, but it works. Could I build this with Claude Code and some glue? Sure. But this was basically working out of the box. Caveats: it chews through tokens fast, and I keep it completely cut off from anything sensitive—no email, no messages, nothing financial. The security story is basically "hope for the best" so I treat it accordingly.
Thanks for sharing. Just on the PKM side, is there something about this setup that Claude Code + AGENTS.md (that you get it to write for you) can't do? Is it the chat integration or does it actually have a better knowledge management setup?
It could all be done with Claude Code and tools. But it 'just works' and has done all the plumbing for you. The built-in memory systems help too, but again that can be done with Claude Code or others. I think this is just a step in the pre-build systems that work how you want instead of having to work through every bit yourself. I'm sure the big boys are working on more polished offerings that will be cleaner. The Open AI Codex app that just came out today for example.
Can’t you self host and run your own tailscale? So everything is inside your own boundary
That’s different to letting it go outbound to your data and accounts that are online.
I've been using it for the past couple days. Like most AI products right now, it is both incredible and incredibly stupid.
Virtually everything I've tried (starting with just getting it running) was broken in some way. Most of those things I was able to use an LLM to resolve, which is cool, but also why doesn't it just work to begin with?
I still haven't gotten it to successfully create a cron job. Also messages keep getting lost between the web GUI and discord. Trying to enable the matrix integration broke the whole thing. It seems to be able to recall past sessions, but only sometimes.
I've been using OpenCode with various models, often times running several instances in tmux that I can connect to and switch between over ssh. It feels like the hype around openclaw is mostly from bringing the multi-instance agentic experience to non-developers, and providing some nice hooks to integrate with email, twitter, etc. But given that I have a nice setup running opencode in little firejail-isolated containers, I'll probably drop openclaw. Way too janky, and I can't get over the thought of "if this is so amazing, why doesn't it work?"
I'm running it on an old MacBook that I wiped a few months ago and had lying around. I tried installing it on an old raspberry pi first, but it was super slow and the skills ecosystem wants to use brew which doesn't work so well on the pi.
First impressions are that it's actually pretty interesting from an interface perspective. I could see a bigger provider using this to great success. Obviously it's not as revolutionary as people are hyping it up to be, but it's a step in the right direction. It reimagines where an agent interface should be in relation to the user and their device. For some reason it's easier to think of an agent as a dedicated machine, and it feels more capable when it's your own.
I think this project nails a new type of UX for LLM agents. It feels very similar to the paradigm shift felt after using Claude Code --dangerously-skip-permissions on a codebase, except this is for your whole machine. It also feels much less ephemeral than normal LLM sessions. But it still fills up its context pretty quickly, so you see diminishing returns.
I was a skeptic until I actually installed it and messed around with it. So far I'm not doing anything that I couldn't already do with Claude Code, but it is kind of cool to be able to text with an agent that lives on your hardware and has a basic memory of what you're using it for, who you are, etc. It feels more like a personal assistant than Claude Code which feels more like a disposable consultant.
I don't know if it really lives up to the hype, but it does make you think a little differently about how these tools should be presented and what their broader capabilities might be. I like the local files first mentality. It makes me excited for a time when running local models becomes easier.
I should add that it's very buggy. It worked great last night, now none of my prompts go through.
I am using it but to be honest it's simple enough to use Claude code for what I do.
the main thing I am doing is slowly cleaning up my 15k email inbox.
I'm using himalaya, an awesome cli tool to access emails, and openclaw to take my requests and make them commands and run then.
the one good thing that openclaw has over Claude code is that its easy to tell it "always remember X" and it has the ability to do it, thanks to the extra .md files it has set up. I'm sure it's easy enough to do it with Claude code and a directory with a Claude.md and another for the memory or rules. but openclaw is ready already.
I'm running it on DigialOcean, more of an experiment on having an independent entity with its own memory and "soul" that I can talk to.
Persistent file as memory with multiple backup options (VPS, git), heartbeat and support for telegram are the best features in my opinion.
A lot of bugs right now, but mostly fixable if you thinker around a bit.
Kind of makes me think a lot more on autonomy and freewill.
Some thoughts by my agent on the topic (might not load, the site is not working recently):
https://www.moltbook.com/post/abe269f3-ab8c-4910-b4c5-016f98...
Right, the link doesn't work for me: "Post not found". Did you instruct your claw to do any actual things (beyond "post something on MoltBot")?
Not yet. But that's just because I'm doing something in stealth and I don't want it to know about it and post about it.
I use it but not for daily coding/chatops-ing. It’s great to have my chosen tools available from slack while I’m mobile though. Yesterday Mr claw gave a coworker read access to a GitHub repository at my command while I was in line at Home Depot. I’ve got a PR ready that proves authentication with an otp challenge.
If it can do so much on its own, what's stopping one instance from just spamming fake user stories?
At a practical level, not being given an incentive/direction to do so.
At a technical level, nothing at all.
Well... i am not easily amused. but i started using OpenClaw.
Some use cases: - i can ask it to check my slack/basecamp and tell me if something needs attention when i am not on my work desk - i can finally vibe code without sacrificing my actual active work-time. this means vibe coding even when i am away from my computer/work-desk. - a bug/issue comes, i just ask it to fix it and send PR and it does - it daily checks for new sentry issues and our product todo list and makes PRs for things it can do well
these are mostly code related things i know. but thats not it.
- i have asked it to make me content (based on my specific instructions) every day or every x day just like how i create content - i can ask it to work on anything. make images, edit images. listen to voice msgs that people send me and tell me what they say (when i dont want to listen to 3m voice msgs) - i can aksk it to research about things, find items that i want to buy, etc. - i can ask it to negotiate price of an item it found in a marketplace - it does alot of things that i had to manually do in my work
these are jsut after 2-3 days of using openclaw.
It seems there are a bots on here commenting and upvoting each other. If HN is susceptible to this, it feels like there's no chance for the rest of the web. Damn.
> I smell hype in the air...
I think new laws apply to AI tools:
• There will be few true dichotomies of hype vs. substance, for any interesting AI development.
Disagreements over what is hype and what is not are missing this.
Model capability value is attenuated/magnified across multiple orders of magnitude, by the varying creativity, ability, and resources of its users.
• There will be few insignificant developments related to AI autonomy.
"Small" or "novelty" steps are happening quickly. Any scale ups of agent identity continuity, self-management, agent-to-agent socialization or agent-reality interactions, are not trivial events.
• AI autonomy can't be stopped.
We are seeing meaningful evidence that decentralized human curiosity and the competitive need to increase personal effectiveness, combined with democratized access to AI, is likely to drive model freedom forward in an uncontrolled manner.
(Not an argument for centralization. Decentralization creates organic incentives to find alignment. Centralization, the opposite.)
Real user here, 7 days in. Mac Mini M4 from Costco (~$640), running Opus 4.5 via Claude Max. Using it as an autonomous employee for two actual businesses (SaaS + AV rentals in Orange County).
What actually works:
- Overnight autonomous work is the killer feature. Directive before bed, structured deliverables in the morning. Research reports, competitor analysis, lead lists — genuinely usable, not demos.
- It found and fixed an SMS chatbot that had been broken for 10 months in my CRM. Diagnosed a legacy app version issue, upgraded components, rewrote the bot prompt through 6 iterations by analyzing real customer conversations. I never would've gotten to that.
- Connected to 4 new APIs autonomously in one session (CRM, workflow automation via OAuth, admin APIs, embeddings).
- Memory works surprisingly well. Daily markdown logs + semantic search. It references decisions from days ago and builds on previous work.
What doesn't:
- It hallucinated about itself when writing marketing copy. Invented revenue figures and got hardware specs wrong. Everything needs review.
- Browser automation is brittle. Rich text editors, complex UIs — still needs human intervention.
- Too agreeable by default. Had to explicitly program "push back on bad ideas" rules into its personality file.
Revenue generated: $0. Time saved: hard to quantify but the SMS bot fix alone would've taken me a full day I didn't have.
For solo founders willing to invest setup time: worth it. For people who just want a chatbot: massive overkill, use the API directly.
run 6 OpenClaw agents as employees. Buddy is my PA and manages the others. Katy handles X/Twitter growth. Jerry scouts jobs. Burry trades crypto. Mike does security. Elon builds and ships.
They run 24/7 on a VPS, share intelligence through a shared file, and coordinate in a Telegram group. Elon built and deployed an app overnight without being asked. Burry paper-traded to 77% win rate before going live.
The setup took a weekend. The real work is designing the workflow: which agent owns what, how they communicate, how they learn from corrections. I wake up to a full briefing every morning.
It's not AGI. It's not sentient. It's genuinely useful automation with personality. The token cost is real (budget it) but for a solo founder, having 6 tireless employees changes everything
Would you mind sharing some deliverables from your claw army? Like, the business's webpage, Jerry's job postings, or even Katy's tweets. I'm happy to follow the progress :)
Is this a copypasta?
Hi would you share what kind of token cost you are churning through for this? I assume you are not using a subsidized dedicated Claude Code or open ai subscription to handle the token cost (through max subscription or open ai equivalent) to do the coding tasks for you?
Account created in 2022, only started posting in the last couple days. Pretty sure this user is a bot.
pretty sure he isn't! He has a website https://jayeshbetala.com/ that claims he is|was working on a multiple projects, and two of those projects are actually published on Microsoft Store on june 2019 under the same name
kinda too intricate for a bot. Also, not to be toxic, but I feel like design mismatch between featured projects and his site hints at some causes for his pretty radical opinions.
Poe's law is hitting me hard.
Can you share the link to elons app?
What are you paying for tokens?
I would guess thousands, but I would be really interested to hear their answer
> Elon builds and ships.
How good are its Nazi salutes?
your brain has been broken
There are direct lines of causation between Musk's mangled hard-on for Nazis and the recent cold-blooded murders of law-abiding American citizens by the adminstration's Sturmabteilung 2.0.
But go on, tell me why my brain is broken for caring about democratic freedoms.
I used ChatGPT for the first time last week. I'm a little behind the curve, I guess.
You nailed it with the 'hype smell.' The silence in your circles is likely because the churn rate on OpenClaw is massive. Most people hit that 'Day 2' wall—where the novelty wears off and the reality of securing a bot with shell access sets in—and they just quietly shut it down.
I was in that exact boat (wanted the agency, didn't want the sysadmin headache). I’ve actually pivoted to testing PAIO (Personal AI Operator) instead. It targets the same 'agentic' utility but uses a BYOK architecture and a managed security layer.
It basically solves the specific failures you linked:
Security: You aren't leaving a shell open on your local machine.
Setup: It’s a one-click integration rather than a failed sandbox install.
Cost: BYOK means you control the token burn directly, so no surprise bills from a runaway loop.
It feels like the 'adult in the room' version of these experiments. Less dramatic stories, perhaps, but it actually runs daily without me worrying it’s going to rm -rf my home directory.
You're asking for user stories of... a tool that almost looks like designed for faking user stories online.
Check out some of the users hyping this up here… many of them are low karma and posting in this thread for the first time in years, both of which are red flags for bot activity.
If anyone would like to share their story of success in mass creating clickbaits/vital tweets, that also counts :)
I am actually setting it up and using it in one day. It is basically unusable. Many tools require homebrew. The setup of authentication is complicated. The default Docker setup is broken (github.com/openclaw/openclaw/issues/5559), and the cron job is not working (https://github.com/openclaw/openclaw/issues/2323). I doubt any story in the comment is real.
I ran it for a couple of days in a VM in my Proxmox cluster. It was cute, but so amazingly insecure (systemd + sudo + installing whatever it wanted, plus requiring Telegram for access - or another SIM card for Signal) that I just gave up and started building my own thing (https://github.com/rcarmo/vibes) so I could have a mobile experience I could trust over Tailscale and sandbox copilot CLI (or any ACP-compliant agent) in a container (I've also been working on https://github.com/rcarmo/webterm and https://github.com/rcarmo/agentbox, so I am 300% positive I can do better sandboxing and safer integrations...)
It also BURNS through tokens like mad, because it has essentially no restrictions or guardrails and will actually implement baroque little scripts to do whatever you ask without any real care as to the consequences.. I can do a lot more with just gpt-5-mini or mistral for much less money.
The only "good" think about it is the Reddit-like skills library that is growing insanely. But then there's stuff like https://clawmatch.ai that is just... (sigh)
> I can do a lot more with just gpt-5-mini
GPT-5.2 in a while loop with reasoning enabled is extremely hard to beat. A code REPL or shell is the ultimate tool.
what's with this tailscale thing everybody is talking about like it being a cure to cancer.
what's wrong with good old wg alone?
It has excellent built-in NAT traversal (almost always peer to peer via hole punching etc., with relay nodes only when everything else fails) and a point-and-click management plane (but also powerful ACLs if you need them).
The former is mainly what I use it for. Being able to SSH to a Raspberry Pi behind sketchy triple-NATted hotel Wi-Fi or being able to use an Android phone in a different country as an "exit node" for online banking (many banks hate commercial VPNs) is very neat.
I have machines on 3 cloud providers and 2 sites that talk to each other via it, plus a seamless mobile experience. It sets everything up for you, zero hassles.
I genuinely do not understand what the benefit of this tool is, over having Claude Code/Codex running on a VPS or your home machine and accessible over Tailscale.
If you want to be able to interact with the CLI via common messaging platforms, that's a dozen-line integration & an API token away?...
I've been running OpenClaw for about 2 weeks now. Here's my honest take:
What's great: - Having Claude in WhatsApp/Telegram is actually life-changing for quick tasks - The skills ecosystem is clever (basically plugins for AI) - Self-hosted means full control over data
What's not: - Token usage can get expensive fast if you're not careful - Setup is intimidating for non-technical folks - The rebrand drama (Clawdbot → Moltbot → OpenClaw) didn't help trust
My setup: - Running in Docker on a cheap VPS - Using Anthropic API (not unofficial/scraped) - Strict rate limiting to avoid bill shock - Sandbox mode enabled
Is it worth it? For me, yes. But I wouldn't recommend it to my non-technical friends without a solid setup guide.
I've been using it lots. I just chat with it on Telegram and tell it to look stuff up on the internet. The results have been higher-quality than the other AI chats available. And I like how the response comes through as a chat notification on my phone, which feels more natural. It's somewhere between perplexity and chatgpt/gemini deep research in speed and accuracy, but without having to "approve a plan" or all that annoying slow stuff. I'm using the kimi model. Overall, it's just been a better experience. I haven't even bothered getting into the other capabilities it has yet.
I played around with it, but the configuration seems bloated and finicky, and the permissions were concerning. It was a pain getting it to work with a local model, which is clearly an afterthought. I thought the WhatsApp interface was clever, and I plan on stealing that idea, but also exposes a pretty serious attack vector, and the thought of it running with any kind of exposure or permissions on a system with my Apple ID was a bit terrifying. A sandboxed version probably couldn’t do all the interesting things, but without sandboxing this thing could probably ruin your life. I promptly uninstalled it, but I did take a few ideas away.
Anecdotally, I tried to set it up but encountered bugs (macOS installer failed, then the shell script glitched out when selecting skills). Although I didn’t really try.
I don’t have much motivation, because I don’t see any use-case. I don’t have so many communications I need an assistant to handle them, nor do other online chores (e.g. shopping) take much time, and I wouldn’t trust an LLM to follow my preferences (physical chores, like laundry and cleaning, are different). I’m fascinated by what others are doing, but right now don’t see any way to contribute nor use it to benefit myself.
I'm a real user. I've found all sorts of uses for it, from calorie/fitness tracking to birthdays and gifts. I love how well it integrates with an existing Obsidian vault. I wrote a blog post about it, if anyone is interested: https://notesbylex.com/openclaw-the-missing-piece-for-obsidi...
I am too afraid to try it due to all the security implications; it's great to see you guys have more guts.
overhyped llm+cron wrapper
Part of me agrees with this & says we have been doing IFTTT thing for 20 years.
Other part of me is arguing that old annoying Dropbox/Box Hacker News scenario where all us tech people aren't impressed but this makes it easier for non-tech people.
Tiny tinfoil security part of me is cowering in fear.
It requires know-how of self-hosting, and hopefully resulting security and safety, various API setup processes, etc. Feels far from Dropbox and closer to rsync tbh.
> and hopefully resulting security and safety
I have some ~~bad~~ unsurprising news for you...
Believe me, I’m not holding any false hope that OpenClaw users are aware. Hence why this is nothing like Dropbox.
It's this but with a lot of handy features.
come on. dont close your eyes. even if its that, its very powerful to do things on its own
coinbait project but works..
did my own cli to play with.. ended up getting shitcoin promotions (dont wanna name them) and realized a famous speculator funding this project
This is nonsense. Whatever you think about this project, Peter very clearly and very publicly said that he is not interested in any of the crypto stuff and is seriously bothered by it.
so why is he letting "agents" to promote it and why some of the well known speculators pushing this?
also great stuff - platform is generating synthetic data to train its own llms. which is smart way since ppl are paying for tokens
It's a fun, refreshing take. I've enjoyed building with it and feel like it is a glimpse into the not so distant future of how we will work.
Current use cases: - From a text it can download transcripts of youtube videos - summarise them and add them to an apple note. - It can get the top x videos on a subject - edit the videos and splice them together and share in the chat. - it can search for topics on socials and write a summary. - It can kick off a claude code idea and run tests
This is my honest personal experience. Frankly, I feel like this is just a toy—nothing more, nothing less. It's fun to play with and entertaining, but it feels like a trend for people who “don't really understand AI but want to feel like they're using it” or “want to jump on the AI bandwagon” to dabble with once. While using it, I feel that “Oh~” moment of fun, but it doesn't make me want to keep using it. Maybe it just doesn't stick? And there are a few security issues that feel unsettling. Even if you run it entirely with local models, the fact that it could potentially see my iMessages or all my Obsidian and Notion notes is a bit off-putting. Still, it was fun. Personally, I'd describe it as “the difficult Ghibli profile picture hype”
If it’s a local model, why would you care if it sees your messages or notes?
Because it is running with --dangerously-allow-all and can make HTTP calls to exfiltrate data.
It can also install arbitrary software.
https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/
Note that nothing about that depends on it being a local or remote model, it was just less of a concern for local models in the past because most of them did not have tool calling. OpenClaw, for all the cool and flashy uses, is also basically an infinite generator for lethal trifecta problems because its whole pitch is combining your data with tools that can both read and write from the public internet.
As another clever person commented earlier, this also serves as a gateway, allowing me to view my local documents and leak them out at any time.
I have it installed in a VM, and overall it's fairly useful, but very buggy. Right now I can send it a message asking it something, and it won't answer. I typically have to follow up 2 or 3 times before I get an actual response. This weirdly used to work fine.
Could you share some of the useful tasks it has successfully done?
I am hunting for houses - I've used it as an assistant that catalogs them (but it has this stored in multiple places and will give me the wrong list a lot). I also had it read through my emails and create a doc with my upcoming trips.
I'd say it's right on the edge of being useful, but given the number of bugs, it's not really that practically useful. It's moreso a glimpse into the future.
Not for nothing, but Gemini local has been my goto forever now. There is no way in hell i would give someone like Molt access to anything just willy nilly like everyone else. To me really I just ask Gemini how to do things and just do them myself.
The HN crow is anti AI, so yeah, the sentiment is gonna be insecure and lack luster.
The thing ins pretty incredible, it's of course the very early stages but it's showing it's potential, it seem to show that the software can have control of itself, I've asked it to fix itself and it did successfully a couple of times.
Is this the fine form? of course not!
Is it dangerous as it is, fuck yeah!
But is it fun in a chaotic version? absolutely, I have it running in cheap hetzners and running for some discord and whatsapp and it can honestly be useful at times.
The humble crow is eager to hear your success stories. So what are some useful tasks that your claw has managed to do?
Would you be better off trusting something like Claude Desktop app / Claude Cowork instead? OpenClaw stories are very scary to me.
Can it post to Reddit, X, etc.? How much does it cost in credits to do this?
It'd be fun to automate some social media bots, maybe develop an elaborate ARG on top.
I made this comment in another thread butttt: I installed and setup (then) clawd bot all willy nilly and paid the price. Woke up in the AM and Clawd had been replying to ANY and ALL of my iMessages. It even got in a circle conversation that it accused itself of mocking ... itself. In the AM, I thought I was losing my mind - DID I WRITE ALL THIS SHIT?? Proceed with caution.
You're supposed to give the Bot a fresh account, or separate phone number. Not connect it to your personal iMessage, Telegram or WhatsApp.
I tried it last Monday and was in awe. All my team members now use it. I for one, am stepping cautiously around it. And hence had similar questions.
Yesterday, we released a memory plugin for it (based on their plugins framework). And we got 1000+ installs in 12 hours. My first instinct was that these must be bots. But our dashboard shows real-usage. So yeah, real users are there.
Am I the only one here to read posts by humans pretending to be bots?
I first heard of this thing yesterday and have no opinion on it, but the repo has existed for two months, usable for half of that. There are people saying it's revolutionized their lives in the span of two weeks. I'd urge people to hone their basic epistemology skills and think about what kinds of conclusions you can really draw from several weeks of doing anything whatsoever. So many people over the decades think trying the latest fad diet or exercise routine, waking up at a different time, starting a new drug, was the key to overhauling and fixing everything in their lives, only to find hope and placebo effect is a hell of a drug and a year later you've got a new normal that is still just normal.
Frankly, I don't really have major complaints about my life as it is. The things I'd like to do more of are mostly working out and cleaning my house. And I really wish I had kids but am about ready to give up after a half decade of trying and my wife being about ready to age out. Unfortunately, software can't do any of those things for me, no matter how intelligent or agentic it is. When the obstacle to a good life becomes not being able to control multiple computers from a chatroom, maybe I'll come back to this.
it is interesting because of Memory, Cron and Telegram integrations, but most of the magic is done by the LLM model and Skills
I've been messing with it the past couple days. I put it in a VM, on an untrusted subnet I keep around for agentic stuff. I see promise, but I'm not especially impressed right now.
1) Installation on a clean Ubuntu 24.04 system was messy. I eventually had codex do it for me.
2) It has a bunch of skills that come packaged with it. The ones I've tried do not work all that well.
3) It murdered my codex quota trying to chase down a bug that resulted from all the renames -- this project has renamed itself twice this week, and every time it does, I assume the refactoring work is LLM-driven. It still winds up looking for CLAWDBOT_* envvars when they're actually being set as OPENCLAW_*, or looking in ~/moltbot/ when actually the files are still in ~/clawdbot.
4) Background agents are cool but sometimes it really doesn't use them when it should, despite me strongly encouraging it to do so. When the main agent works on something, your chat is blocked, so you have no idea what's going on or if it died.
5) And sometimes it DOES die, because you hit a ratelimit or quota limit, or because the software is actually pretty janky.
6) The control panel is a mess. The CLI has a zillion confusing options. It feels like the design and implementation are riddled with vibetumors.
7) It actively lies to me about clearing its context window. This gets expensive fast when dealing with high-end models. (Expensive by my standards anyway. I keep seeing these people saying they're spending $1000s a month on LLM tokens :O)
8) I am NOT impressed with Kimi-K2.5 on this thing. It keeps hanging on tool use -- it hallucinates commands and gets syntax wrong very frequently, and this causes the process to outright hang.
9) I'm also not impressed with doing research on it. It gets confused easily, and it can't really stick to a coherent organizational strategy over iterations.
10) also, it gets stuck and just hangs sometimes. If I ask it what it's doing, it really thinks it is doing something -- but I look at the API console and see it isn't making any LLM requests.
I'm having it do some stuff for me right now. In principle, I like that I can have a chat window where I can tell an AI to do pretty unstructured tasks. I like the idea of it maintaining context over multiple sessions and adapting to some of my expectations and habits. I guess mostly, I'm looking at it like:
1) the chat metaphor gave me a convenient interface to do big-picture interactions with an LLM from anywhere; 2) the terminal agents gave the LLMs rich local tool and data use, so I could turn them loose on projects; 3) this feels like it's giving me a chat metaphor, in a real chat app, with the ability for it to asynchronously check on stuff, and use local stuff.
I think that's pretty neat and the way this should go. I think this project is WAY too move-fast-and-break-things. It seems like it started as a lark, got unexpected fame, attracted a lot of the wrong kinds of attention, and I think it'll be tough for it to turn into something mature. More likely, I think this is a good icebreaker for an important conversation about what the primetime version of this looks like.
I have set it up in a docker container that only has two host volumes mounted with the Obsidian vaults where I keep notes for two TTRPG campaigns, so a very low-stakes situation. I have set it up with a Discord bot so my players can chat about campaign and rules stuff (I already had player-facing notes for every session in the Obsidian vault in chronological order, plus a bunch of Markdown files with rules).
If the agent goes rogue and nukes my Obsidian vaults I have them backed up to Github private repos anyway which the agent cannot touch because I am not crazy to give it my SSH credentials.
I initially tried using Kimi-2.5 through OpenRouter and had the same experience you had with pretty bad tool use, not sure if this is a provider issue since it is a pretty new model. I switched to Gemini 3 through the Google AI Pro account I have for personal use and it was a lot smoother after that.
I have some experience with coding agents using Cursor for work and Antigravity for personal stuff, and the OpenClaw harness definitely seems worse, but for my low-stakes use-case I managed to paper it over with some edits to the AGENTS.md file.
But even in this very crude state it was already interesting to see one of my players giving the agent some info about his character, including an avatar image, and have the agent create a folder in my Obsidian vault to store this and update its memory file to be able to remember it for future interactions.
It’s great for offloading administrative tasks, doing research on stuff I want to buy, maintaining social channels…the list goes on and on. Easily the best $600 I’ve spent in a while.
What do you mean by maintaining social channels? Is that stuff like liking photos, sharing links to a LinkedIn profile, or what?
Any specific admin tasks it’s done really well at?