## Why NanoClaw Might Beat OpenClaw [00:00] It's like building a video game character and adding stats, module by module, instead of simply installing Frankenstein's monster. Honestly, this is my favorite claw so far. It might even beat open claw in my humble opinion. And in this video, I'm going to show you why. What we're going to do is explore it a little bit, see how it works. Now I'm going to install it. Then I'm going to set it up with Telegram, create an agent swarm that runs a whole pipeline that includes scraping the web, doing research and then presenting me with solutions to a certain problem. The main spoiler I'll give you now is that Nanoclaw is AI native, whereas OpenClaw is Frankenstein's monster with a whole bunch of software plugged into itself, and that's what makes it so bulky, 430,000 lines of code. Nanoclaw is super easy for Claude code to just customize. You can say, hey, let's install Nanoclaw, and then let's have it do certain things instead of having to dissect Frankenstein's monster. I'm upgrading your soul! So this is really interesting stuff, guys. I know there's a lot of objects out there with all these claws, but NanoClaw is the one that has my attention right now. And by the end of this video, you'll see why. So let's dive in. NanoClaw vs. OpenClaw. So NanoClaw claims itself to be a lightweight ## What Is NanoClaw? (Lightweight Alternative Explained) [01:08] alternative to OpenClaw, just another one of the claws out there that have come up as a response to OpenClaw's complexity. Now, I am a solopreneur. I am not a a developer or engineer, if you are one of those, you will soon learn that I hardly know what's going on with the software, but I am able to use AI to figure it out. And that's why in this video, I will share with you how it goes for me to install it and actually use it for my use cases and if it's worth my time. So Nanoclaw is supposed to be more secure, lightweight, and can dynamically rewrite its own code, which is kind of cool. And apparently it's the first AI assistant to run agent swarms. We'll check that out a little later. So the first thing I always do with these new agents is think about what do I want to use it for. A big part of what I like to do is summarize social media so I can stay away from the fear-mongering on the platforms and just get what I want to get. I also don't have a lot of time to listen to podcasts, so I like to summarize podcasts. So to curate my content, maybe add things to Notion, maybe give me ideas for my own YouTube videos, things like that. But in this video, we're just going to start from the very base level of getting it set up and why we might want to use it. So the first thing I do is I open up OpenCode. ## Key Differences: AI Native vs Frankenstein's Monster [02:35] I connect it to Venice. And here's my conversation here. I just ask, explain this repo to me. How do the containers work? How does it differ from OpenClock? So then Opus via OpenCode explains, hey, here are the key differences. And a big part of it is the code base size, the architecture. And in this case, Nanoclaw does not support as many channels as OpenClaw. I said it before, but there is a reason OpenClaw went so viral. It is really cool software. And Nanoclaw even suggests, if you want to customize it, just fork the repo and have cloud code rewrite it to match your needs. So that's more or less what we're going to do here in this video. We also see that there's no companion apps. It's terminal only. That's cool with me. I don't think we need all that stuff. And I like this here at the end, the target user for NanoClaw are power users who want to understand and own their full stack. Whereas with OpenClaw, it's the individual user who just wants full featured batteries included assistant that just does it. And you don't really necessarily need to understand how it works or why it works. And once again, in Opus's words, Nanoclaw is a reaction to the complexity of OpenClaw. And all the other ones, PicoClaw, ZeroClaw, blah, blah, blah. ## How Ephemeral Containers Actually Work [03:47] I review a bunch on the channel. Maybe you've seen them. So I have been messing around with Nanoclaw. And one thing I noticed is that the containers, they like disappear. So Opus analyzed that. And yes, they're ephemeral containers, which means unlike with OpenClaw, OpenClaw just runs on the device and it has full access to the device. I mean, you could put it in a container too, but that's why they say get a Mac Mini, get a Raspberry Pi to run OpenClaw so it can just do what it needs to. But NanoClaw, meanwhile, if I'm not mistaken, is much more safe to run on your main computer because everything happens, everything the AI agent does is happening in a container. And then these containers are actually removed. So they're not actually persistent. So Nanoclaw is going to run a background service on our computer. And then when we chat with our bot via Telegram, WhatsApp, whatever, then we're invoking the container to be created for the message to be processed to the LLM. It'll return a result and then get destroyed. But then meanwhile, the state must survive. The memory, how it knows it works. And this happens through CloudMD. And you can mount directories from your computer. So you can give the container directories to your computer. So you can actually give the container access to directories on your computer. So here we see the lifecycle is it spawns a container. It mounts the persistent directories that we give it access to. It pipes in the prompt. It streams the response. And then the container is done and it's removed. And then you message it again and it'll start a new container. One thing I'm curious, though, is will the same container be used in a single conversation? or each actual message spawns a new one. What we're going to see in this video is if the speed is worth it to us. And the answer is the container actually has a timeout. So once it sees, okay, the conversation is kind of over for now, then the container disappears. So when we ask Opus to explain like I'm five, think of each group chat as getting its own temporary office room. When someone sends a message, Nanoclaw opens a room, puts the AI assistant inside, and slides the message under the door. The AI does its work and slides the answer back out the door. The room stays open for a bit in case more messages come in that get slid underneath the door. And if nobody talks for a while, the room closes and disappears. So this is like a Harry Potter type office building. And then when you chat again, a brand new room container opens up. Okay, so that's how Nanoclaw works. Very interesting, in my opinion. And it's been getting a lot of good reviews on X online. ## NanoClaw vs the Other Claws (ZeroClaw, PicoClaw, NullClaw) [06:33] So I thought, hey, we should check this out. It's got almost 20,000 stars. So it's probably good for something. Now, compared to all the other little claws, this is not bragging about, hey, you only need a megabyte of RAM to run it. For example, here, NullClaw, which I intend to review soon, and ZeroClaw and PicoClaw, they all say this, NullOverhead, NullCompromise, whatever. And they explain how much RAM they need. But I feel like this kind of thing isn't maybe necessary for everyone. Like, I don't even know how to find a $5 board. I don't even know what that is. You know, like I'm happy using a Raspberry Pi. ## Installing NanoClaw (Git Clone to Running) [07:09] Maybe if I could find a $5 board, then I could run 20 null claws instead of just one open claw on my 8 gigabyte Raspberry Pi. But all of these are really just a response to open claws complexity that I think are kind of just distractions. So we're ready to install it. I'll go ahead. I'll just copy the Git repo. We just do Git clone, Nanoclaw, and then CD Nanoclaw. Okay, I've already got it installed. So we'll CD Nanoclaw, and we'll run Claude. And that's right here in the quick start, okay? You just clone it. You change directory to the Nanoclaw directory, and then you run Claude, and then you run the setup forward slash command. So we'll run setup. Now, just like OpenClaw, we're going to bootstrap our setup here, the channel we communicated on, etc. It sees that Docker is already running. This is a fresh install. We're going to use my Clawd code subscription, and I've got to run this in a new terminal. Then it's going to give us this token, which I now paste here. Make sure the line breaks aren't all messed up. Oh, whoops. Do not paste the token here. Oh, well. So really what I should have done is just nano.env and put that right there. But the AI did it for me. Which messaging channels do you want to enable? Telegram is cool. Let's maybe do Discord too. But actually, let's just do Telegram. I have a token. For some reason, it's okay to post a Telegram bot token. I'll add it to.env. What should it be? Telegram bot token. That's probably what it is. So we'll nano.nv telegram bot token. We'll paste that there. And we will save it. Done. So this is applying the telegram skill. ## Setting Up Telegram Integration [09:15] Now it's prompting. Should agents have access to any directories outside of Nanoclaw? For now, no. Let's completely sandbox it. That will be obviously more safe. All right. So now it's time to register the chat. Say hello. Chat ID. It gives me that ID. I copy it. Now in Docker, we don't see a container yet, but it will be there soon. Now we test it. We send it a message. Would I like to add agent swarm support? Yes, baby. Let's do agent swarms. Why not? Hi, are you there? Let's check back in on the telegram. Okay, we need three to five telegram bots for the agent pool. ## Adding Agent Swarm Support [10:06] Wow, this is kind of neat, huh? And then I need to make a group. All right, so this is going to take a second. Okay, so Cloud Code is finishing the setup for the swarm. And meanwhile, the swarm group has NanoCloud Coder and Researcher. This is really interesting how much work it's being done for the telegram agents, for the subagent swarms, because that's just a skill, telegram swarm. And then I gave it the bot API keys. And now it's just doing a bunch of coding. Cloud Code is doing the coding and just knows what to do here. Let's open up our swarm. Hey, everybody. Nothing. All right, so we've got no containers running. I've sent messages to the group and to each agent, but still only the main first NanoClaw agent is responding. All right, I'm just going to say this has been my favorite so far to set up, and here's why. Even right here in the readme, it says, NanoClaw is designed to be bespoke. You make your own fork and have CloudCode modify it to match your needs. And literally, yeah, it's AI native. And CloudCode is running the setup here. And obviously, CloudCode is the best thing ever. So it just knows what to do here to make NanoClaw work how we want. And that is the future of, that's how software should work, I feel like, is just have the AI do it. And people complain, oh, too many tokens. But ultimately, you're getting a lot of flexibility out of that. So I think it's a good trade-off. OK, so it's apparently... OK, so apparently they're send-only. Don't have a registered chat. Yeah, that won't work. Hi. OK, so Cloud Code has been figuring this out. I don't know if it's actually supposed to be in a group or not. It originally told me to do that. But Nanocall here is telling me that, yes, It has access to the agent tool, which lets it spawn and coordinate agent swarms. So I want to create a pipeline. ## Building a Real Pipeline: Scraping X with Apify [12:24] Scrape X for hot topics. See what users are posting about. Research the web for alternate perspectives. Scrape X with Apify. Skill I will provide. Plus API key. For hot topics in AI. See what users are posting, commenting about. Find the pain. What are people having trouble with? then research the web for solutions, and then create a document for how to solve the problem. Now, this is very generic, right? In my case, the solution would be like, here's a video explaining what to do. But maybe I want to use this also for creating a basic web app to solve a problem or a skill, right? So really, I just want to demo building out a pipeline here. All right, so let's get the Apeify skill. If you don't know Apeify, it's like the best thing in the world for scraping. There's scrapers, they're called actors, that do all kinds of scraping all over the internet. It's really well-priced. So I'll go here, and I will get my API key. We'll say this is Nanoclaw. And it is going to expire tomorrow. It's going to expire tomorrow, which means I don't have to worry about it going through plain text. Are you seeing it now? So these are the Apify agent skills. Very powerful, okay? Very powerful. And then it's going to spin up a swarm. Let's check in. There's its Docker container, and it's been turning those on and off, so that's working. Hey, team, let's get that Apify skill installed. So now I'm nudging it. I don't know what's going on here. Yeah. So the Docker container disappeared again. Silence. Real quick interruption, guys. If you are building with AI and you're doing so alone, that is the gap. Not the tools, not the skills. It's the crew. It's the community. And that's why I want to let you know that the AI Captain's Academy exists on school. There's so much going on in the AI space. But inside the school, you can close the technical gaps, whether that's context engineering, a framework for understanding AI tools and using them, the command line, git, stuff like that, everything. Not to mention weekly workshops and networking calls and good vibes. So there's a link below to join the crew. I hope to see you there. Back to the video. Okay, so it's restarting things. Before I give my final judgment on why I keep shutting off, let's acknowledge that I'm working on it in the background with Cloud Code. All right, working now. ## Agent Swarm in Action (200 Tweets Scraped) [15:05] Hello, continue. Okay, now we got two containers. Hey team, let's start building our pipeline. So now, yeah, it looks like our nanoclaw came over here. Hey team, let's start building our pipeline. I don't know if I'm overwhelming it. Okay, cool. Oh, apify school is installed. So I just didn't get a response over there. Now about building the pipeline. Okay, so it doesn't remember. Let's see. I DM'd you the API key. Oh, wait. Oh, now we got the other one ready. Okay, so now it's communicating. I don't know what's going on here. This one's the researcher, I think. Yeah, this is the researcher. Oh, and now the coder. The scraper. Oh, so it gave the name. Okay. Searching for tweets about AI frustration, chat, GPT problems. Okay. Scraper is running. I'll wait for it to finish before kicking up the... Okay. Oh, so I think what's going to happen here is that it's just, it has two bots right now. It has three bots, the main orchestrator and then these two other ones I made thinking it would be researcher encoder. But really, it's just giving whatever sub-agent task and name and system prompt or whatever that it needs at the time, it's giving it that name. So then when that's done, it'll start. So I should give it like five different bots probably. Getting messages from each bot in private. And even you in private told me the scraping is going. ## 52 AI Pain Points Analyzed Automatically [17:00] So I guess it's all good. Wow, this is really interesting. I wasn't expecting to set up agent swarms in this walkthrough. Scrape complete. 100 tweets collected across five search queries. Agent reliability and trust. Biggest theme. Users report AI agents drifting over time with no change logs. Broken off. And a huge gap between demo quality and production quality. Output. Okay, okay. So there's the scraper. The other bots must be. Okay, so apparently the main bot. Doesn't know what's going on with the other bots. So this is just main Nanoclaw now. Wow, this is really interesting. I don't know if I should just wait now. I'm just going to wait. So one agent did this thing of Shonded, but now nothing is happening. To continue the pipeline, I discussed with the original Nanoclaw. What's going on? and these containers are still here so maybe they're still working all right so let's figure this out and this is how it's going to work for you too because i'm just following in directions here the pool bots are sending to your private chat instead of the group chat the group chat is a separate isolated agent the group chat's an agent the swarm should be running from the main agent but sending full messages to the group how do you want this one or private chat drives group shows team. Yeah, I would just like everything to happen in the group. That makes the most sense, right? All right, guys. So it did it. We finally figured it out. Cloud code was very helpful here. And I would say if you're going to use Nanoclaw, you basically should just keep a cloud code window open right next door all the time. But basically, I told the group finally, hey, let's create this pipeline. scrape X, see what users are talking about, find the pain, then research the web for solutions and create a document to solve the problem. So then it just asked, do we have a specific angle? This is the main bot that asked, the orchestrator. And I said, yeah, let's do it for AI business and AI tools like autonomous agent frameworks. So it launched the scraper agent, which scraped 200 tweets, found some complaints, found some pain points, 52 distinct pain points, as a matter of fact. And then the researcher took those 52 pain points and started deep research on those solutions. And all of this took about 10 minutes. So it probably didn't go as deep as it could. Obviously, something easily fixed by just telling Cloud Code, hey, let's use this skill for deep research. In fact, there's a lot of cool skills out there you can find for something like this. So I created a report in its own files. I guess I'm just being lazy because right here in the nanoclaw folder. I should be able to find it. What's it called? AI pain solutions. Moves and data. I have no idea where it is. Yeah, so what I would do there then is say, the swarm produced an output MD file, but I can't find it in Finder. And it, oh, cool. Never mind. I don't need to. You never really know. Cool. So let me download that. And that came from the main orchestrator here, not the researcher. Very interesting framework here, guys. Really kind of inspiring too. I wasn't looking forward to reviewing another claw. All right. So let's see. It's not really even markdown. I don't see any markdown formatting besides, I don't know, italics. AI pain points and solutions, March 2026. Analyzing 52 distinct pain points, agent reliability, cost optimization for enterprise AI, Integration middleware for agents, security layers for autonomous agents, reliability and hallucination. Everyone's affected. Solutions, RAG. Okay. Cost and scalability. All right. So, you know, the quality of this research is going to be a product of the time I spent preparing the agents for research. And as you saw in this video, I just told it to do some research. Because really, I was just wanting to play around with this for a video. But now I know what it's capable of. ## Final Verdict: AI Native Is the Future [21:31] And this is like context first thinking. So now I know what this tool is capable of. I'm going to use it for something. I really like this swarm thing. Maybe you can do that with OpenClaw. But I like how it's just easier to fix with NanoClaw. It can just pop it in the cloud code. Hey, fix this. Do this. Change this. I really like that AI native aspect to it. All right, so that's it. That is NanoClaw. I really love this AI native aspect to it. I believe that is the future of everything. You give a base tool to an AI, and then you tell the AI to expand upon it. That is really big here, and I like seeing it in NanoClaw. I understand why it's getting all the attention it is. In my humble opinion, it's way cooler than the ZeroClaw, all these other freaking NullClaw, whatever claw, bot, bleh. And if you don't want to deal with Frankenstein's monster, OpenClaw, which has everything built in as software, then NanoClaw is a great opportunity to try something more lightweight and customizable, flexible for you. So thanks for watching. Check out the AI Captain's School if you want to explore autonomous agent frameworks together and don't forget to hit subscribe for more videos. See ya. ¡Vladios!