alt.hn

2/28/2026 at 12:47:39 AM

Zclaw – The 888 KiB Assistant

 https://zclaw.dev

by kristianpaul

3/2/2026 at 5:11:41 PM

“888 KiB Assistant” but the assistant itself is a multi terabyte rental-only model stored in some mysterious data center.

by tehsauce

3/2/2026 at 11:53:11 PM

The whole point is that this fits on an ESP32, which has wifi. We're not quite at the point where it makes sense to run the whole thing locally - if you do try it, it will need a fan, and be loud etc.

For my part, I installed Nanoclaw on my Arch derived OS (I love Arch!), and it worked fine until the next day some update decided to revert the power management settings, and now my glorious assistant is dead.

There's something to be said for a barebones OS. No bullshit, no updates.

Also, playing with hardware watchdog timers and GPIOs and DACs can be so much fun.

by seertaak

3/2/2026 at 5:22:18 PM

I'm getting "serverless" flashbacks.

by amelius

3/3/2026 at 12:39:33 PM

modelless

by pgt

3/3/2026 at 12:52:42 PM

This thread reminds me how Javas heavy GUI written in Java itself was called "lightweight" when in fact it did not feel lightweight at all on the hardware of the time.

by stuaxo

3/2/2026 at 6:42:56 PM

My model is at home... just 16Gb still a lot but just FYI

by kristianpaul

3/2/2026 at 6:28:40 PM

It seems to support connecting to your own LLM on the same LAN

by Rebelgecko

3/2/2026 at 8:14:33 PM

The point is the agent is still the LLM. No LLM, no agent.

by croes

3/3/2026 at 3:58:06 AM

LLMs are not agents. LLMs are language models that simply respond to a text prompt with a textual response. Agents are middleware that take input from the user and then use LLMs to drive tools.

by otterley

3/3/2026 at 6:13:47 AM

They are just a to-do list. The real work is done by the LLMs

by croes

3/3/2026 at 6:26:40 AM

An LLM has no motive power, like a script without an a cast, or a program without a computer to execute it.

by otterley

3/2/2026 at 7:58:05 PM

I tried connecting OpenClaw to ollama with a V100 running qwen3.5:35b but it was really, really, really slow (despite ollama itself feeling fairly fast).

These "claw" agents really multiply the tokens used by an obscenely huge factor for the same request.

by dheera

3/2/2026 at 10:49:23 PM

i recently decided to get into this ocean boiling game too, the 32GB V100 seems like a pretty good VRAM/$. if i may ask, do you make any special accommodations for cooling? i've never dealt with a passively cooled card before and i'm curious whether my workstation fans (HP Z840) will be sufficient. i'm going to try 2 cards at first but i think i might be able to squeeze a third in there

by jcgrillo

3/2/2026 at 11:21:56 PM

No. I have a Titan V CEO edition, which is basically a 32GB V100 but has full active fan cooling which I'm finding works just fine.

by dheera

3/2/2026 at 11:42:37 PM

Oh very cool. Some folks are printing shrouds for dual 40mm fans so I'll probably try that if the stock case fans don't do it

by jcgrillo

3/2/2026 at 7:07:03 PM

For people who don't get this: it's a Home Assistant type thing. You don't do inference on it, you send it a message on Telegram and it does things with physical things through GPIO. You could use a $140 Raspberry Pi with 8GB RAM and host a local model on it plugged into 30W AC power... or you could use a $10 ESP32 which can run for weeks on a tiny battery, and your existing Wifi connection with a cheap cloud model (cloud models are as cheap as $0.02/1MTokens). This makes it easier to ramp up on new ESP32 projects. You can just tell it to do things / give you info, rather than having to write code.

by 0xbadcafebee

3/3/2026 at 2:00:56 AM

I wouldn't trust LLMs with anything, even low stakes things like handling my notes. I fail to see how people willingly give keys to the kingdom, why, asides from just FOMO/trend chasing.

by Bridged7756

3/2/2026 at 9:46:34 PM

The model can be taken care of in cloud but hardware also depends upon what we want to do right? If we want to run some lightweight python scripts etc, we cant use ESP32 right?

by newswasboring

3/3/2026 at 2:25:36 AM

The point is not to write python scripts at all. You just tell it to do something and it does it, no programming. It comes with tools the AI can use. https://zclaw.dev/use-cases.html

(you can write MicroPython for an ESP32, but that's not what this project is for)

by 0xbadcafebee

3/2/2026 at 10:45:21 PM

dope so i can hook it up to my appliances, hvac, etc? maybe it'll do rad things with my smart doorlocks, remote car starter. can it send my self-driving waymobile to pick up my family at the airport too? such wow

by jcgrillo

3/2/2026 at 5:42:00 PM

Me: "GPIO 5 can be active for a maximum of 100ms, then it needs to cool down for at least 1s. Otherwise the MOSFET is fried."

Zclaw: "GPIO 5 is active now, however the server is not responding so I'm awaiting further instructions."

by amelius

3/2/2026 at 5:12:27 PM

I fail to understand why 888 KiB matters if it's just a wrapper around a cloud api.

by gas9S9zw3P9c

3/2/2026 at 5:39:04 PM

Have you seen OpenClaw's codebase? 680.000 LOC.

I care how big it is.

by ramon156

3/2/2026 at 6:30:21 PM

A lot of the *claws emphasize binary size and lines of code. I think for better or worse people treat codebase size as a proxy for "how much of the project is unsupervised, unmaintainable, buggy AI slop?"

by Rebelgecko

3/2/2026 at 5:39:54 PM

Because of resource-constrained environments, the primary deployment target seem to be microcontrollers. You can get ESP32 boards for pretty cheap.

by mihaelm

3/2/2026 at 5:37:28 PM

Because it means you can run it on an ESP32 which is a low power microprocessor package.

by renewiltord

3/2/2026 at 9:06:58 PM

8 is lucky number in China

by boznz

3/2/2026 at 7:13:05 PM

888KiB is quite large, but I see they're including the whole rest of the firmware in that number, fair enough. Their actual application code weighs only 35,742 bytes, compiled.

by Retr0id

3/2/2026 at 6:53:43 PM

Still lightyears from a one-bit AI assistant. Send 1 to save the humankind, 0 to exterminate. And hurry up because it's in undefined state right now!

by wartywhoa23

3/2/2026 at 7:03:34 PM

[dead]

by cwoodyard

3/2/2026 at 5:45:21 PM

There are many concerns and areas for improvement with open claw and other similar projects (continuous loop script with broad OS access that manages your agents and interfaces with a standard messaging app)

However, file size I have never seen on that list. I would rather offer for something that is even bigger in file size so it afford certain functionality like better security tighter permissions however it would do that.

by hidelooktropic

3/2/2026 at 5:55:24 PM

File size is a legit property to keep in mind if your goal is to create an agent that runs on ESP32 boards. They don't expect you to run Zclaw on Mac Mini.

by mihaelm

3/2/2026 at 5:59:40 PM

What's the use case for running this on a tiny board? Isn't the whole point that it can use your computer for you?

by allthetime

3/2/2026 at 6:41:09 PM

For something like OpenClaw yes, but not for Zclaw. I think the naming is more about riding the current wave of Claw-related interest rather than positioning it as competition or replacement for other clawies.

Zclaw is about running an agent in your embedded system.

by mihaelm

3/2/2026 at 6:39:29 PM

The examples seem to suggest it would be chatting with your home automation in natural language.

by markstos

3/2/2026 at 7:20:06 PM

Before you know it your smart thermostat will be blogging. The joke is on everyone who thought IoT couldn't get any worse. Just imagine the new landscape of security vulnerabilities this opens up.

by fc417fc802

3/2/2026 at 7:23:30 PM

My "smart" gas stove can be turned on over the internet (if I allow it to connect)—perfect appliance to put an LLM in charge of.

by markstos

3/2/2026 at 6:48:12 PM

I made a secure one:

https://github.com/skorokithakis/stavrobot

Everything runs in containers (I run it on a server along with everything else), plugins have a permission system so eg the AI can read emails but not delete or send, etc.

I really like it, I run it as my main agent and it has been extremely helpful.

by stavros

3/2/2026 at 8:29:56 PM

Part of the usefulness is based on the same thing that makes it so dangerous.

If it can only read but not act, it’s safer but less useful.

by croes

3/2/2026 at 8:31:45 PM

I can't restrict OpenClaw if I don't need the extra capabilities. I can restrict this.

by stavros

3/3/2026 at 3:04:13 PM

You restrict OpenClaw by not providing it certain credentials.

by croes

3/3/2026 at 3:07:13 PM

Again, with my design you can give it fine-grained access to parts of services, which OpenClaw itself cannot do. This is just a fact.

by stavros

3/2/2026 at 5:48:45 PM

Clicked on this expecting to see a crontab file.

by jesse_dot_id

3/2/2026 at 5:08:06 PM

The domain crashed and burned or something, hopefully this link is correct:

https://github.com/tnm/zclaw

by giancarlostoro

3/2/2026 at 4:51:59 PM

There is the same divide starting to form that NFTs had back in the day. Tech bros instantly like if something has claw in the name, the rest of us will dismiss anything with that naming and philosophy as toxic slop culture. will be interesting to see how far this one will go.

by jFriedensreich

3/2/2026 at 5:32:55 PM

1. why

by karolcodes

3/3/2026 at 6:59:40 AM

Exactly, So many *-Claw releases, with only distinction is either language or memory footprint.

by human_hack3r

3/2/2026 at 6:30:14 PM

2. fun

by dpe82

3/3/2026 at 1:45:18 PM

Crypto

by downboots

3/2/2026 at 5:59:01 PM

2. hype

by rvz