OpenClaw has 52+ modules and runs agents with near-unlimited permissions in a single Node process. NanoClaw is ~500 lines of core code, agents run in actual Apple containers with filesystem isolation. Each chat gets its own sandboxed context.
This is not a swiss army knife. It’s built to match my exact needs. Fork it and make it yours.
Edit: I see you, making edits to the readme to make it sound more human-written since I commented ;) https://github.com/gavrielc/nanoclaw/commit/40d41542d2f335a0...
I don't make any attempt to hide it. Nearly every commit message says "Co-Authored-By: Claude Opus 4.5". You correctly pointed out that there were some AI smells in the writing, so I removed them, just like I correct typos, and the writing is now better.
I don't care deeply about this code. It's not a masterpiece. It's functional code that is very useful to me. I'm sharing it because I think it can be useful to other people. Not as production code but as a reference or starting point they can use to build (collaboratively with claude code) functional custom software for themselves.
I spent a weekend giving instructions to coding agents to build this. I put time and effort into the architecture, especially in relation to security. I chose to post while it's still rough because I need to close out my work on it for now - can't keep going down this rabbit hole the whole week :) I hope it will be useful to others.
BTW, I know the readme irked you but if you read it I promise it will make a lot more sense where this project is coming from ;)
I don't mind it if I have good reason to believe the author actually read the docs, but that's hard to know from someone I don't know on the internet. So I actually really appreciate if you are editing the docs to make them sound more human written.
You get paid to get stuff done, period.
And it reminds me of a comment I saw in a thread 2 days ago. One about how RAPIDLY ITERATIVE the environment is now. There area lot of weekend projects being made over the knee of a robot nowadays and then instantly shared. Even OpenClaw is to a great extent, an example of that at its current age. Which comes in contrast to the length of time it used to take to get these small projects off the ground in the past. And also in contrast with how much code gets abandoned before and after "public release.
I'm looking at AI evangelists and I know they're largely correct about AI. I also look at what the heck they built, and either they're selling me something AI related, or have a bunch of defunct one-shot babies or mostly tools so limited in scope they server only themselves with it. We used to have a filter for these things. Salesmen always sold promises, so, no change there, just the buzzwords. But the cloutchasers? Those were way smaller in number. People building the "thing" so the "thing" exists mostly stopped before we ever heard of the "thing", because, turns out, caring about the "thing" does not actually translate to the motivation to getting it done. Or Maintain it.
What we have now is a reverse survivorship bias.
OOP stating they don't care about the state of their code during their public release, means I must assume they're a Cloutchaser. Either they don't care because they know they can do better which means they shared something that isn't their best, so their motivation with the comment is to highlight the idea. They just wanted to be first. Clout. Or they don't exactly concern with if they can as they just don't care about code in general and just want the product, be it good or be it not. They believe in the idea enough they want to ensure it exists, regardless of what's in the pudding. Which means to me, they also don't care to understand what's in the ingredient list. Which means they aren't best to maintain it. And that latter is the kind that, before the LLM slop was a concept in our minds, were precisely ones among the people who would give up half way through Making The "Thing".
See you in 16 weeks OP. I'll eat my shoe then.
As I said in my comment, no shade for writing the code with Claude. I do it too, every day.
I wasn’t “irked” by the readme, and I did read it. But it didn’t give me a sense that you had put in “time and effort” because it felt deeply LLM-authored, and my comment was trying to explore that and how it made me feel. I had little meaningful data on whether you put in that effort because the readme - the only thing I could really judge the project by - sounded vibe coded too. And if I can’t tell if there has been care put into something like the readme how can I tell if there’s been care put into any part of the project? If there has and if that matters - say, I put care into this and that’s why I’m doing a show HN about it - then it should be evident and not hidden behind a wall of LLM-speak! Or at least; that’s what I think. As I said in a sibling comment, maybe I’m already a dinosaur and this entire topic won’t matter in a few years anyway.
Just something that screams "I don't care about my product/readme page, why should you".
To be clear, no issue with using AI to write the actual program/whatever it is. It's just the readme/product page which super turns me off even trying/looking into it.
Before the proof of work of code in a repo by default was a signal of a lot of thought going into something. Now this flood of code in these vibe coded projects is by default cheap and borderline meaningless. Not throwing shade or anything at coding assistants. Just the way it goes
Why not, if they're making people read AI slop without checking it first? They deserve the shit-nudge to fix it.
Don't worry, bro. If enough people are like you, there will be fully automatic workflow to add typos into AI writing.
Assuming the written/generated text is well written/generated, of course.
A hundred times this. It's fine until it isn't. And jacking these Claws into shared conversation spaces is quite literally pushing the afterburners to max on simonw's lethal trifecta. A lot of people are going to get burned hard by this. Every blackhat is eyes-on this right now - we're literally giving a drunk robot the keys to everything.
As a former (bespoke) WP hosting provider, I'd counter those usually did. Not sure I ever met a prospective "online" business customer's build that didn't? They'd put their entire business into WP installs with plugins for everything.
Our step one was to turn WP into static site gen and get WP itself behind a firewall and VPN, and even then single tenant only on isolated networks per tenant.
To be fair that data wasn't ALL about everyone's PII — until by ~2008 when the Buddy Press craze was hot. And that was much more difficult to keep safe.
1. what if, ChadGPT style, ads are added to the answers (like OpenAI said it'd do, hence the new "ChadGPT" name)?
2. what if the current prices really are unsustainable and the thing goes 10x?
Are we living some golden age where we can both query LLMs on the cheap and not get ad-infected answers?
I read several comments in different threads made by people saying: "I use AI because search results are too polluted and the Web is unusable"
And I now do the same:
"Gemini, compare me the HP Z640 and HP Z840 workstations, list the features in a table" / "Find me which Xeon CPU they support, list me the date and price of these CPU when they were new and typical price used now".
How long before I get twelve ads along with paid vendors recommendations?
Where does this idea come from? We know how much it costs to run LLMs. It's not like we're waiting to find out. AI companies aren't losing money on API tokens. What could possibly happen to make prices go 10x when they're already running at a profit? Claude Max might be a different story, but AI is going to get cheaper to run. Not randomly 10x for the same models.
> It's hard to say for sure because they don't publish the financials (or if they do, it tends to be obfuscated)
Yeah, exactly. So how the hell the bloggers you read know AI players are losing money? Are they whistleblowers? Or they're pulling numbers out of their asses? Your choice.
Sam Altman has made similar statements, and Chinese companies also often serve their models very cheaply. All of this makes me believe them when they say they are profitable on API usage. Usage on the plans is a bit more unknown.
They spend money on growth and new models. At some point that will slow and then they’ll start to spend less on R&D and training. Competition means some may lose, but models will continue to be served.
Furthermore, companies which are publicly traded show that overall the products are not economical. Meta and MSFT are great examples of this, though they have recently seen opposite sides of investors appraising their results. Notably, OpenAI and MSFT are more closely linked than any other Mag7 companies with an AI startup.
https://www.forbes.com/sites/phoebeliu/2025/11/10/openai-spe...
Heck we were spoiled by “memory is cheap” but here we are today wasting it at every expense as prices keep skyrocketing (ps they ain’t coming back down). If you can’t see the shift to forceful subscriptions via technologies guised as “security” ie. secure boot and the monopolistic distribution (Apple, Google, Amazon) or the OEM, you’re running with blinders. Computings future as it’s heading will be closed ecosystems that are subscription serviced, mobile only. They’ll nickel and dime users for every nuanced freedom of expression they can.
Is it crazy to correlate the price of memory to our ability to localize LLM?
None of these went 10x. Actually the internet went 0.0001~0.001x for me in terms of bits/money. I lived through dial-up era.
What if a thermonuclear war breaks out? What's your backup plan for this scenario?
I genuinely can't tell which is more likely to happen in the next decade. If I have to guess I'll say war.
> Skills over features. Contributors shouldn't add features (e.g. support for Telegram) to the codebase. Instead, they contribute claude code skills like /add-telegram that transform your fork.
I’m interested to see how this model pans out. I can see benefits (don’t carry complexity you don’t need) and costs (how do I audit the generated code?).
But it seems pretty clear that things will move in this direction in ‘26 with all the vibe coding that folks are enjoying.
I do wonder if the end state is more like a very rich library of composable high-order abstractions, with Skills for how to use them - rather than raw skills with instructions for how to lossily reconstruct those things.
https://maordayanofficial.medium.com/the-sovereign-ai-securi...
At least 42,665 instances are publicly exposed on the internet, with 5,194 instances actively verified as vulnerable through systematic scanning.. The narrative that “running AI locally = security and privacy” is significantly undermined when 93% of deployments are critically vulnerable. Users may lose faith in self-hosted alternatives.. Governments and regulators already scrutinizing AI may use this incident to justify restrictions on self-hosted AI agents, citing security externalities.Apple containers have been great especially that each of them maps 1:1 to a dedicated lightweight VM. Except for a bug or two that appeared in the early releases, things seem to be working out well. I believe not a lot of projects are leveraging it.
A general code execution sandbox for AI code or otherwise that used Apple containers is https://github.com/instavm/coderunner It can be hooked to Claude code and others.
Is this materially different than giving all files on your system 777 permissions?
Yes, because I can't read or modify your files over the internet just because you chmod'ed them to 777. But with Clawdbot, I can!
It's more (exactly?) like pulling a .sh file hosted on someone else's website and running it as root, except the contents of the file are generated by a LLM, no one reads them, and the owner of the website can change them without your knowledge.
Lesson - never trust a sophomore who can’t even trust themselves (to get overly excited and throw caution to the wind).
Clawdbot is a 100 sophomores knocking on your door asking for the keys.
Thankfully the official Agent SDK Quickstart guide says that you can: https://platform.claude.com/docs/en/agent-sdk/quickstart
In particular, this bit:
"After installing Claude Code onto your machine, run claude in your terminal and follow the prompts to authenticate. The SDK will use this authentication automatically."
> Unless previously approved, Anthropic does not allow third party developers to offer claude.ai login or rate limits for their products, including agents built on the Claude Agent SDK. Please use the API key authentication methods described in this document instead.
Which I have interpreted means that you can’t use your Claude code subscription with the agent SDK, only API tokens.
I really wish Anthropic would make it clear (and allow us to use our subscriptions with other tools).
> Third-party harnesses using Claude subscriptions create problems for users and are prohibited by our Terms of Service.
This project uses the Agents SDK so it should be kosher in regards to terms of service. I couldn't figure out how to get the SDK running inside the containers to properly use the authenticated session from the host machine so I went with a hacky way of injecting the oauth token into the container environment. It still should be above board for TOS but it's the one security flaw that I know about (malicious person in a WhatsApp group with you can prompt inject the agent to share the oauth key).
If anyone can help out with getting the authenticated session to work properly with the agents running in containers it would be much appreciated.
$70 or whatever to check if there's milk... just use your Claude Max subscription.
How wouldn't they know? Claude Code is proprietary they can put whatever telemetry they want in there.
> how are we violating... anything? I'm working within my usage limits...
It's well known that Claude code is heavily discounted compared to market API rates. The best interpretation of this is that it's a kind of marketing for their API. If you are not using Claude code for what it's intended for, then it's violating at least the spirit of that deal.
And apparently it's violating the terms of service. Is it fair and above board for them to ban people? idk, it feels pretty blatantly like control for the sake of control, or control for the sake of lock-in, or those analytics/telemetry contain something awfully juicy, because they're already getting the entire prompt. It's their service to run as they wish, but it's not a pro-customer move and I think it's priming people to jump ship if another model takes the lead.
I think most people fail to estimate the real threat that malicious prompts can cause because it is not that common, its like when credit cards were launched, cc fraud and the various ways it could be perpetrated followed not soon after. The real threats aren’t visible yet but rest assured there are actors working to take advantage and many unfortunate examples will be seen before general awareness and precaution will prevail….
My gut reaction says that I don't like it, but it is such an interesting idea to think about.
Openclaw is very useful, but like you I share the sentiment of it being terrifying, even before you introduce the social network aspect.
My Mac mini is currently literally switched off for this very reason.
I assume this is to keep the footprint minimal on a Mac Mini without the overhead of the Docker VM, but does this limit the agent's ability to run standard Linux tooling? Or are you relying on the AI to just figure out the BSD/macOS equivalents of standard commands?
If you've got an exploit for docker / linux containers, please share it with the class.
What I'm saying is that in practice, containers and VMs have both been quite secure.
Also, you can configure docker to run microvms too https://github.com/firecracker-microvm/firecracker-container...
Minor nitpick, it looks like about 2500 lines of typescript (I am on a mobile device, so my LOC estimate may be off). Also, Apple container looks really interesting.
Quick Start
git clone https://github.com/anthropics/nanoclaw.git
Or is this just so hastily thrown together that the Quick Start is a hallucination?
That's not a facetious question, given this project's declared raison d'etre is security and the subtle implication that OpenClaw is an insecure unreviewed pile of slop.
If it somehow wasn't abundantly clear: this is a vibe coded weekend project by a single developer (me).
It's rough around the edges but it fits my needs (talking with claude code that's mounted on my obsidian vault and easily scheduling cron jobs through whatsapp). And I feel a lot better running this than a +350k LOC project that I can't even begin to wrap my head around how it works.
This is not supposed to be something other people run as is, but hopefully a solid starting point for creating your own custom setup.
> This is the anti-[OpenClaw](https://github.com/anthropics/openclaw).
It's certainly helpful for some things, but at the same time - I would rather improved CLI tools get created that can be used by humans and llm tools alike.