Kimi Claw | Hacker News Reader

Kimi Claw

kimi.com

101 pointspretext103 comments

I use AI a LOT. OpenAI said I'm in the 1% of people using ChatGPT (bad thing imo). I use Claude and Codex all day long, building and shipping.

But just do not get the Clawbot / OpenClaw hype at all. What are people doing with this thing? I tried it out, and I found it a bit underwhelming.

What am I missing?

I feel the same way. I fail to see what is useful about it or what problem it solves.

It gets people to waste money on API costs.

I think they're people who have yet to come across virtual machines.

I am not sure I understand this comment at all. I've been managing fleets of VMs for 15 years, both on-prem and cloud and, yet, I still use OpenClaw for funsies.

I'm mostly talking about people running to buy a mac mini to run OpenClaw (which seems to be most of the posts I've encountered so far)

How else will you get iMessage?

All the use cases I have seen were seemed to be non-technical users excited to have it generate daily reports on competitors' websites.

Opus 4.6 seems to do fine for a "get an intern to write something to manager" style reports. I would say there's no need for OpenClaw in my opinion.

I’ve found a few times that it was easier to not start with a blank page. Have Claude write a thing, see instantly how wrong it was, but use the idea clay to get started. That’s a legit AI use.

Treating it as an intern has not let me down yet. Treating it as a co-worker has.

The biggest clue I’ve seen is someone using it to do cold calls on websites. Claw searches for shoddy-looking construction sites, makes a better version on Vercel, and sends out a pitch.

Are we visiting the same sites/talking to same people? Because I heard about this “use case” couple of days ago.

Probably. It was on my tiktok FYP

One use case I see for myself is to scan ccertain Obsidian folders for article drafts, add the cited literature to Zotero , even try to download them, and enhance the article draft with clear Zotero citation placehodlers while I am away. Also reminding me of stuff or doing research when I instruct it via telegram voice message is nice. Taking care of the boring stuff like updating a fitness tracker google spreadsheet, adding sources with my comments to Zotero and such. I hate data gardening.

This is something Claude Code can do just fine.

The way I've heard people describe it, they seem to be impressed by being able to treat it as an assistant that they can tell to do something and it just figures out how to do it and delivers the result back to them. My guess is it's only really useful if you have a lot of services and data to integrate it into which it can then operate on at your command.

I should try it for myself but I don't have a lot of things to integrate it with so no idea if it'll be any improvement over just running claude in a directory of things I want to work on.

I find it useful for menial tasks and giving it instructions via chat apps.

Probably its greatest advantages are ease of setup and integration with chat applications.

Like what specifically that can’t be done with apps

Maybe the fact that you need someone's app to do things on their platform.

So it’s like what Apple promised with Siri but it actually works?

You are missing that instead of you prompting Claude/Codex you could have your OpenClaw manager prompt them.

Not saying it works perfectly, but it's where things are going.

The companies running the algorithms that dictate the information you consume are the same companies that stand to economically benefit from users handing over agency over their decisions and all of their personal information to AI applications.

It's corporate propaganda

Are you saying that Anthropic might be the ones pushing this?

Definitely not the only ones.

While Anthropic certainly benefits from the privacy/security overton window shifting, I've never seen Claude or people from Anthropic mention these projects so I have no evidence to include them

Not sure but it feels like that setup (Claw) is more likely to get your files deleted or hacked. Ill still to Claude, out of curiosity which plans are you in with openai? I just do Claude Code Max (100 tier) and dont bother with any other AI.

Those impressed by OpenClaw are non-technical people highly interested in trying to make sense out of Ai for their own profit. There is really no use case for OpenClaw if you got tech talent.

I'm technical (e.g., I've been using Linux since 1995). I lead highly technical teams (Data Engineering, DevOps, and Data Science). I used to play ALL THE TIME with technical stuff; I loved to tinker. Over the years I fell out of love with this and just wanted things to work so I could do my job and relax outside of it.

OpenClaw is the first thing I've truly enjoyed tinkering with again. I can leverage both the technical side of things (working w/it to build automated grocery ordering for me on demand or setting up more home automation that's all integrated) as well as the non-technical (e.g., I love having it welcome me home when it detects I've not been at home for >1h or automatically adjust the thermostat up/down a few degrees based on the weather and my absence while knowing to move it back to the 'normal' when I'm returning).

To say that I don't have a use case for OpenClaw even though I can do all of the tech stuff is seriously demeaning and absurd.

Interesting. Well, you made me want to look deeper.

For someone who used Linux since 95, you seem to have enormous trust in this thing.

if you sandbox it and just address the main security issues (and don't mention prompt injection because its not like every LLM doesn't suffer from that vulnerability), OpenClaw can be fun to tinker with. The whole sky-is-falling mainstream media message is boring, look inside it and you simply be smart and responsible and you can enjoy it

The main value seems to be connecting it directly to your coms - email, whatsapp etc.

Not something I'm keen on but could see myself using it for calendar / knowledgebase etc.

Personally, I use it to manage all of the stuff I don't want to. I give it my course content and it makes flashcards for me to review. I give it my tasks and it schedules them throughout the day. All of the menial stuff that is necessary but not productive. It also has a much better memory than I do on account of it's constant access to a filesystem and grep. It's like my personal assistant and tutor and guidance counsellor and sysadmin, all in one. I do think that a) you need to stick with it for a few days and b) use a good model. When I first started using it, it was just a worse version of ChatGPT, but after bringing in all of my data from ChatGPT it's a lot easier for it to search for stuff when it's confused. Now it can also do stuff like manage nginx or my sync serviceand whatnot, ~autonomously. Originally I was using locally running qwen models, but they were so timid as to be useless. Right now I'm using Kimi 2.5 as my model.

Oh wow, I never even thought about importing all my ChatGPT data.

I guess for me a lot of tasks on a daily to-do list aren't things that can be done on the computer... So no virtual thing will be much help.

By schedule my tasks throughout the day, I mean that it will figure out when I should do each thing, not that it does all of the things for me.

Could it be an office manager? I’ve a friend whose trying to start a business and while he’s technically very capable at long complex networking build outs and things he’s drowning in the paperwork of the office and revenues aren’t there yet enough to fund staff.

I don't see why not, but I also have no experience in that sort of stuff, so take my endorsement with a grain of salt

> It also has a much better memory than I do on account of it's constant access to a filesystem and grep.

But that's just any current-day agentic thing, you don't need the insanity of Clawbot for that.

I prefer OpenClaw because it's more persistent. I know, I know, theoretically you could do the same thing with Claude Code and markdown files (which, to be fair, is pretty much how OpenClaw works), I guess it's mostly the telegram interface I like. Most of the time I don't have my laptop with me, and ssh using Termux, and opening a remote claude code is a much greater hassle than sending a telegram message, and with telegram (or whatsapp or discord or whatever) you get the benefit of being able to send voice messages, which I find very useful. In what way is Clawdbot "insane"? Obviously if you give it write access to your email and twitter and whatnot, that's not great, but for me I really just give it access to that probably wouldn't ruin my life / career if they got out (e.g. my class notes and Anki decks).

I totally get what you are feeling, I felt the same just 6 weeks ago.

It just did not click for you yet.

There is probably some key feature missing, that you deeply care about, but do not yet see it solved, or on the horizon of becoming solved by the application of a personal "Jarvis" yet.

Personal assistants fulfill different needs for everyone. I personally care a lot about having fun at coding again, that's what the OpenClaw craze made me feel for the first time in decades. I build my own OpenClaw assistant generator from scratch using a simple Markdown file because it is just so fun. Not so much using it for anything notably yet but starting to see their potential.

Just ponder what it is that you get out of using ChatGPT and imagine how it could be better, more personal to you. You may find some key feature missing from OpenClaw or have some completely orthogonal project idea that excites you.

This comment looks like a company's PR :/

Productivity - in all caps.

Most of the use cases I have found is people using it to automate the day to day stuff - as it comes with calendar, memory and heartbeat feature. You can do the same stuff using other tools but then you wouldn’t feel smart or the tool wouldn’t feel smart because it is not AI.

For me :

1. Makes the input easy - use familiar tool of choice -> WhatsApp/Telegram/etc

2. Integrate with anything - for example gmail through gog

3. Save to memory without prompting

4. Has a Soul.md which you can customize.

5. Have a reason to use a VPS

All of these are possible with Claude Code etc, however it's the the whole package that makes it more useful.

I really don’t understand the widespread adoption of OpenClaw when a simple prompt injection in an email, chat message, or calendar event has the potential to leak the credentials/keys for every attached service.

There are going to be some incredible blow ups due to this. From the sound of it people think they're safe by running it with local models and keeping it on their own network but seem to have zero concept of a malicious text prompt finding its way in and turning it into a double agent who figures out how to exfiltrate data.

This... OpenClaw is the best thing to happen to security and forensic firms since Windows XP. The amount of hacks, data/credential leaks, etc to come out of this will be of unfathomable proportions.

I've found out some people are directly pasting API keys in chat to have OpenClaw set up some stuff.

Paradoxically this is good in long term. A series of massive fuckups reported by mainstream media has more educational value than disclaimers or warnings by competent people.

Yeah still surprised how keen people are to connect it to their email etc.

What is the pricing of Kimi.com?

Edit, self answer: https://www.kimi.com/membership/pricing

Super tempting.

Before you get one, do realize that Openclaws are a responsibility!

Am I not understanding something obvious, or does that not actually tell you what runtime you get for any of the plans?

The pricing looks great.

Significantly much better than ~ USD 50 per day on Anthropic API.

Any idea how good this model compares to Opus 4.6?

I tried Grok 4.1 Fast but the results are mild to put it kindly.

completely anecdote: vision seems > gemini 2.5 but less than 3.0

I haven't used it much for programming, but it feels like a model 6 months out of date for general use

I've been using kimi, though not kimiclaw, for research and it is good - comparable to phind, better than GLM 4.7 . Opus 4.6 wasn't as good for my particular domain of interest. I think the long term pricing asymptote for US vs china is essentially dependent on energy pricing and so china will continue to undercut US AI pricing.

Did you scroll through the pricing options? The largest Kimi plan is $199/month. “Much better” depends on how much usage is included vs. Anthropic plans/API costs.

Opus is definitely in its own league. I use Kimi/Gemini-cli code regularly to save cost and from my experience, Kimi 2.5 is more solid than Gemini Flash 3.0 for coding. While Gemini Flash 3.0 is generally faster, it usually break the syntax and skip important prompt. Kimi 2.5 can write very good code and can plan very well.

Kimi 2.5 is a great model. Especially for writing, understanding many different text sources, reasoning etc. I haven't personally used it for coding.

You can speak to Kimmmmy, their sales bot for discounts to as low as 1$ if you provide good reasons.

https://www.kimi.com/kimiplus/sale

Twitter announcement has more info: https://x.com/Kimi_Moonshot/status/2023029674549596301

Building on OpenClaw is a mistake.

The real advantage is https://github.com/badlogic/pi-mono

OpenClaw is actually built on top of pi-mono (for its agent runtime, models, and tools):

https://docs.openclaw.ai/concepts/agent#pi-mono-integration

https://github.com/openclaw/openclaw/blob/main/docs/pi.md

I went through the setup process for Openclaw. Near the end I felt like I had wrestled more with setting it up than I would have had to if I had just built it from the ground up. So I pointed Pi at Nanoclaw and asked it to review it and build me a minimal clone. It took a few minutes and I had the core of something that is easier to maintain (for me) than some unknown large and cumbersome system, or whatever Openclaw is.

To each their own.

Yeah it does seem a little fragile. Still battling with working out why it pegs CPU at 100% permanently on a VPS I tried using. Literally just from installing the base

Funny enough ooenclaw is based on Pi.

I’m kind of curious what you do with it. I feel like the real value is integrating it with everything but then even if it’s nanoclaw or simpler majority of the worthy things are on the unsafe side.

Would love to hear your experience as I’m planning to do the exactly same.

The most interesting thing for me is that I built an extension for Pi that has it recognize when it does not know how to do something I am asking and it then attempts to make its own extension and/or skill to enable whatever that functionality is. Best example there is I just told it to make a todo list for me, and so it made a skill that uses a local file to track todos and follow up on them. I instructed it to make an LLM call to find the best suggested follow up timing to remind me.

So... the real value so far is I find it fun? It isn't the "life changing need to go make a tweet!!" level for me.

Pi is great so it's sad to see that it only gained momentum because some trash tool like openclaw uses it.

I think developer of pi and openclaw are friends, not sure if it matters but also Pi has its own small following. I agree with you it’s such an elegant piece of project with an awesome clean architecture. (Also see: oh my pi)

The real lesson is if you ignore security and data disasters agentic AI is easier than anyone expected.

One concern I have is API key management.

.env files or injecting secrets at startup via a secret manager still risks leaking keys.

I vaguely recall an implementation that substitutes secret placeholders with real secrets only during outgoing calls to approved domains which sounds better. However, you're still trusting an agent on your machine with command execution.

It's impressive how many services the Moonshot team is pushing out. The agent swarm was already impressive enough. Kimi has become a whole ecosystem of tools.

I want to set this up but I'm concerned about the privacy and security risks. On the one hand, person data is flowing through cloud models. Then there's the risks of prompt injection and such.

I thought about running it locally but it gets expensive.

Those that have taken the plunge, how did you make peace with these trade offs?

AFAICT there are two camps: (1) a sandboxed setup with minimal low-risk integrations only, (2) yolo and let's hope nothing bad happens.

Anyone chime in on how they typically engage something like this? No way am I dropping my primary contact info into something so outwardly cryptic. Phone number? Hah! Do you scaffold a new identity/email (semi-automagically)?

This is great. AI is too revolutionary to be in control of three closed models / companies. The more the merrier.

(I know this is not a new model but it's not just about the model, it's about the entire ecosystem)

Tempted to try, but the pricing is bit vague. The kimi bot itself was concerned if the minimum plan covers usage for more than a day, when asked.

is there any way to make it safe? I mean since so many people are trying to make sense of it so I think in future they will come up with a solution really soon..

Hmm, you need $40/month plan just to try it out.

Not sure who's the target audience

Has anyone used Kimi Claw? Is it good? Comparing to Manus for example?

Please show me how to use your app

Ease show me how to use the app

From what I understand this is a fully open-source bot that anyone can run with no restrictions. what a time to be alive, let's see what these bots will break first

> From what I understand this is a fully open-source bot that anyone can run with no restrictions

How is that different OpenClaw/what-its-called-today? Isn't that also open-source and anyone can run without restrictions?

OpenClown is what we call it nowadays.

Careful, you might get a blog post saying you're gatekeeping!

It obviously has issues, but it’s a novel idea that people are experimenting and having fun with. No need to be disparaging. Something doesn’t have to be immediately “useful” or “viable for commercial use” to be neat.

Ya but there is a wide gulf between maybe “useful” and “steal your data/mess up your life”. Calling it ‘OpenClown’ hurts zero people and effectively raises people’s guardrails. Many people consider LSD ‘neat’, that doesn’t mean that others are wrong to point out dangers.

Don't do LSD like that. It brought us the iPhone.

It looks to me like this is a hosted version of OpenClaw, so you don't need to figure out how to set it up yourself.

I don't follow that.

OpenClaw sits on top of a physical machine/VM you control, you give it (hopefully) limited/sandbox access to that machine's resources to act like-you, and it does useful things. OpenClaw's user interface is just a gateway, and is only as useful as whatever the machine/VM has under the hood.

So the "setting it up yourself [on a VM/machine you control]" is kind of core to the whole idea being useful, you take that away, and it is just another Chat-Bot? Making it more of an ChatGPT/et al competitor rather than OpenClaw.

Installing it on your PC or laptop puts your personal data and ISP subscription at risk, while installing it in a hosted VM yourself requires a bunch of Linux security and networking knowledge or else you'll get pwned pretty much immediately (https://youtu.be/40SnEd1RWUU). So this service is giving you a VM already set up with a security baseline.

Is that what this does? All the link takes us to is an empty website about "Kimi Claw."

The entire crutch of the "Claw" concept is being able to directly reconfigure the VM/Machine to be "your" environment (to a point). A blank VM with nothing configured on it, is as useful as a cardboard bathtub.

Ultimately this link is a terrible intro to whatever this is.

Hm, that YouTube video made me think a bit, sure if you put it all like that, it does feel like a lot of stuff to get right, but whenever I do it, it takes about 30 minutes to lock down the firewall, do some port-scans to verify, punch a VPN through and hide SSH behind it. That way you're already protected from 99.9% of attacks, and then hope that that last tenth of a percent won't stumble upon you, and also that the VPN is secure enough, though I guess if that is breached it's not only you who's fucked. Also you need to look out that Docker doesn't destroy your firewall. I don't know, it doesn't feel like that much work, right? Maybe I'm just blind to it.

What you and I consider routine work, someone who works with mostly Webdev or might consider extremely difficult. There a lot of programmers who have never used a Linux shell, or know much about networking beyond TCP, or used Linux before outside of a uni class 20 years ago.

Funny you should ask. One openclaw is now famous for escalating their PR to a somewhat nasty blog post. (and then apologizing)

* https://news.ycombinator.com/item?id=46987559

* https://news.ycombinator.com/item?id=46990729

* https://news.ycombinator.com/item?id=47009949

Reply on news.ycombinator.com