Transcription
Welcome back.
So, at this point, if you follow this channel, you have seen Open Claw, Pico Claw, Nanobot, Iron Claw, and now this new Zero Claw. Honestly, at this rate, I'm going to need a separate playlist just to keep track of all the claw variants. Someone in the Rust community clearly looked at the AI agent space and thought, "I can do this better, smaller, and faster." and Zero Claw is that attempt. It's a Rust-based AI agent framework. 3.4 MB binary starts in under half a second. Supports 22+ providers including any OpenAI compatible endpoint like Olama has, Telegram and other channels built-in, and requires nothing except your own API key or a local model.
And that is what we are going to do in this video. We are going to install Zero Claw locally, and we are going to integrate it with local Olama-based models. If you don't know what Olama is, or the tools which I mentioned around Open Claw, Pico Claw, Iron Claw, and all that stuff, and if you want to check them out from various angles, I have covered them in various videos, as you can see here, from day one. So, let's get started.
I am going to use this Ubuntu system. I have one GPU card, Nvidia RTX 6000 with 48 GB of VRAM. And the reason why I am going to use that is because I have this Olama installed already, and I have GLM 4.7 flash model, which requires GPU VRAM, of course. You can use any model as long as it supports tool use, which means you can access external functionality. So, Olama is there, and now we need to install our Zero Claw, or whatever the name was. I'm actually, you know, starting to forget the names.
And if you're looking to rent a GPU or CPU or any VM of any sort, you can check the link of Mass Compute in the video description, and I will also give you a 50% discount coupon you can use. I'm also renting it from them. So, why not? Also, if you would follow me on X, I would highly, highly appreciate it for any updates without any hype.
So, first prerequisite in order to install Zero Claw is that you should have Rust installed. So, I'm just going to check if Rust is there or not. As you can see that Rust is not installed. Installing Rust is fairly straightforward. All you need to do is to run these two commands. It is going to download the script. It is going to just, you know, ask you a few general questions. I'm just installing it in a standard way. The Rust is quite lightweight. It is quite performant. So, let's wait for it to get, you know, installed. And if you already have Rust installed and if you got the version, just run this command: `rustup update`, and it is going to update your existing Rust installation. And then you can configure your current shell in Linux by running this command. And you can see that the Rust is now installed.
Next up, let's `git clone` the repo. And I will drop the link to it in the video's description. You can see it's a very small repo. And now we can build it with the `cargo` command. It is going to take a few seconds. And the build phase is complete. Now you can install it by running this `cargo install --path .` and that is going to be fairly quick too. Everything is installed, and now we are good to go.
You would just run that onboarding script for Zero Claw. Initially, just give API key dummy and provider `openrouter` or whatever you like, `openai` or whatever, it really doesn't matter because we will be configuring it with Olama. And within a few seconds, you can see that it has onboarded our Zero Claw. The, you know, title looks good. These are some of the workspace layouts, and I will be doing another video where I will be going through all of these workspaces because they are similar across all of these variants, including Open Claws. So, just stay tuned. We will also do a deep dive into the architecture. And this is where, you know, everything is provided. I'm just going to use everything shortly.
So, now what I'm going to show you is what exactly the default config file looks like, and what we can do in order to just use it for Olama. So, I'm opening it in my VS Code, as you can see here. So, this is what the default autogenerated file looks like from the Zero Claw onboard command, and it contains all the default settings like provider set to `openrouter`, `clawed` as the default model, memory, security gateway, and channel configuration, essentially the full skeleton of how Zero Claw runs. And if you notice, the file extension is `.toml`. It is a simple configuration file format which I think stands for Tom's Obvious, Minimal Language, that uses key-value pairs and sections and square brackets, just like, you know, we do in Rust. And it uses key-value pairs and, you know, sections. I mean, it is designed to be human-readable and easy to edit.
Okay, so what I'm going to do, I'm going to replace this whole file with the Olama-based one. And don't worry, I'm going to give you the link to this file in the first pinned command. You can access it in my GitHub repo if you feel like it. Just give it a star to the repo. Up to you. So, all there is to it is that I'm just giving it my local model, which I already showed running with Olama, and no need to put any API key. These two things are important: Olama and default provider, which is Olama, and default model. If you have a different model, of course, just change it over there. Make sure to use the same thing with the tag in your Olama. Make sure you use the full name, otherwise it is not going to work. So, that's it. Let's save it, and let's go back to the terminal and see how Zero Claw works.
And now you can run your Zero Claw by just, you know, with the agent, like we do in Open Claw or any other variant. I'm just asking it, "Are you running on Olama?" And then it is telling me that, you know, it says, "I'm not running on Olama." It says, "I'm running on Linux host with GLM flash model." Okay, but of course, you can check the status, just like we do it in other ones. There you go. So, this is the status where it says that provider is this Olama, and model, and a lot of other things around it. You can also run `doctor` with it, and this is the doctor. You would need to just start that demon, and then it is going to give you the exact answer. And now it is running. You can just go to another window, and then this is the gateway where it is running, and you can run the doctor or whatever you like. We can even access it from any other tool which can access this endpoint at port 8080, just like this. And the rest of it is the same where you can configure your channels, Telegram, and all that. If you want to do that, just go to my channel, watch any of these videos, and I already have shown how to do that similar process.
By the way, at the end, I just want to give you a very quick overview of its architecture. This is the architecture diagram which they have shared. I'll just make it a bit bigger to see, to show you what exactly is happening here. So, if you quickly look at it, every message you send flows from left to right through this diagram. It starts at the channel, as you can see here, which could be Telegram, Discord, Slack, or whatever. And then it passes through a security layer that handles gateway pairing, rate limiting, file system sandboxing, and encrypted secrets before it ever reaches to the agent. The agent sits in the center and coordinates everything. It talks to your chosen AI provider through a pluggable provider trait, stores and recalls context through the memory system, which uses a full hybrid search engine built on SQLite with vector and keyword search combined. And then it hands off tool execution to a sandbox that enforces command allow list and past jelling. On top of that, you have the heartbeat and cron system for scheduled tasks, a tunnel layer for exposing the gateway safely, and a setup wizard that walks you through all of it. Under 60 seconds, everything. Provider, channel, memory, tool, tunnel is a swappable Rust trait. That's the whole point.
So, look, pretty good tool. Still, there are a few rough edges, I would say, but very, very new tool. I think we just have to make sure that we use it in a secure way. The architecture is genuinely solid, I would say, and the performance numbers are real. Keep an eye on this one, I would say. And we already have done it, you know, now how to integrate it with Olama. That's it.
Again, if you think that this helped you, I would request you to please become a member of the channel, or at least subscribe, and please follow me on X. If you're looking for AI updates, please hype the video and share it among your friends. Thank you for all the support.