hardware

[tropic jolt]

🖨️

[copper birch]

https://www.hackster.io/news/this-raspberry-pi-powered-smart-fridge-keeps-an-eye-on-ingredients-with-openai-s-gpt-4-with-vision-acc820f03af1

😄

https://www.home-assistant.io/voice-pe/

anyone has this?

[tepid surge]

fyi if you bought limitless you can flash it with custom firmware. planning to link it up with clawd for it to stalk me

[astral steeple]

I bought one, but it came with a broken componet that drives the speaker. both the internal and external wouldn't work, so I never realy got to mess with it. I ended up just flashing echo 5 shows with the satellite stuff and https://github.com/msp1974/ViewAssist_Companion_App

[ocean furnace]

I do! And not using it, it crossed my mind that it could be a way to communicate with clawdbot, but haven't thought how yet.

[astral steeple]

I'd posted in #home-automation , but I am using this to send to my python agent currently https://github.com/jimrushPersonal/ConversationForwarder can bridge that gap. It basiclly tells HA to send to external agent instead of it's internal one

that's if you want to keep HA in the loop

[lament tartan]

Reachy Mini is so fun for projects like this!

[solid spear]

this is a robot that you talk to or something? rpi with a layer on top? how do you interact with him/it? looks like a fun option for living-room-based clawds 🙂

[lament tartan]

Yeah, it's got two versions: a wired version and a wireless version. The wireless version is based on a rpi https://www.pollen-robotics.com/reachy-mini/

I love it because it's so open and really easy to modify. You can basically hack it and make it do whatever you want. People are already doing some pretty crazy stuff with it, but it's got such a good base of sensors built in that it makes it really fun to play with.

Personally, I've just been using it pretty basic right now. I just send commands to it from my Claude Code session rather than trying to embed everything on the Raspberry Pi itself. I might do that eventually though.

[solid spear]

nice! it looks fun! what's the craziest thing you've seen? all i can think of right now is "clawdbot meets physically embodied siri" but i'm sure these much more imaginative things going around

[astral steeple]

YALL ARE SPENDING ALL MY MONEY!

[royal axle]

Home Assistant Voice Preview Edition

[lament tartan]

Yeah, people have done some pretty amazing things from modifying the robot itself to add arms, stuff like that. There's some dude that's selling an upgrade kit to give it LEDs for eyes so that the color of the eyes can change. That sort of thing is probably the craziest stuff I've seen, but I haven't really done any of that yet. I've just been using it as an embodiment tool like you said.

[whole salmon]

I bought clawd this!

[lofty spruce]

Can the old homepod (the big one) be jailbroken to run clawd? 👀

[astral steeple]

not sure but echo show 5 gen1 can 😄

[cinder flame]

Has anyone though about these? https://github.com/espressif/esp-box

[ocean stag]

The ESP box dev kits have pretty bad mics and audio. It’s somewhat ok if you sit directly in front of it but not comparable to Alexa devices

[quiet reef]

omg we need a video when this arrives 😂

[whole salmon]

https://x.com/buddyhadry/status/2008566156244729971?s=20

Buddy Hadry (@buddyhadry)

Meta Ray-Bans (native WhatsApp support) + @steipete's @clawdbot + tmux = hands-free voice control for my whole system & all of my Claude Code and Codex agents

Agent status check:
"Check agent tmux sessions, show any awaiting approval...yes approve that command"

At a store,

X · Jan 6, 2026, 3:47 PM

[tepid surge]

zuck already bricked my limitless, he is the only barrier to clawd supermacy

[solid spear]

razer is also launching an AI desk hologram thing haha. lets hack it and make a lobster one

[still elbow]

I bought a omi dev kit v2...👀

[bold delta]

wdym?

[tepid surge]

bought limitless ai, then it was bought by meta. i dont rly want meta on me all the time so...

[bold delta]

ahhh I see

[copper birch]

oh fuck it i'm setting up whatsapp

[nocturne zodiac]

Spent the last few hours trying. Something isn't working with the gateway because messages aren't making it in.

[copper birch]

https://developers.meta.com/blog/introducing-meta-wearables-device-access-toolkit/

we can have an official app isntead of abusing whatsapp for this

[solid spear]

lmao

seems cool tho

[tepid surge]

anyone has a working limitless.ai account?
i was working on reverse-engineering it so i can get the recordings myself, but before i managed to record BLE commands, the app factory reset my device.. and it can't connect to it with the app since my account got booted due to the meta acquisition

[lament tartan]

Just a little update, had some extra time yesterday so I setup a clawd on the Reachy Mini rasp pi…was a little bumpy, but it worked. Maxes out the processors easily tho, which messes with the robots servo daemon. Prob not worth debugging, gonna just use my Mac Studio for clawd and keep the Reachy as embodiment only?

[timber vessel]

Can’t believe I’m sitting here about to impulse buy an m4 Mac mini 😭

[solid spear]

one of us! one of us!

[vestal pebble]

I’ve got a refurb M2 Mini arriving any minute now

[timber vessel]

Where’d you buy it? Tempted to go the refurb route too

[rough chasm]

Considering to buy an M4 mini, too. What spec would you recommend?

[nocturne zodiac]

I impulse bought the M3 Ultra 512 GB for to run this lol

[tepid surge]

hey its me ur best friend

[vestal pebble]

Ebay

[cinder flame]

DO IT

[edgy current]

m3 ultra! are you trying to run clawd locally? why is the mac mini so popular amongst clawdbot fans? luckily i already had one, but can't any vps/old macbook laptop work fine? all the compute is happening remote no?

[vestal pebble]

I wanted something with MacOS that was relatively cheap with at least 16gigs of RAM with some processing headroom for other projects. Most of the time it’s going to sit in the corner and I want it to be quiet and cool.

[timber vessel]

I'm thinking about the mac mini because I want to run an automated browser that looks like me (i.e. has a gpu, has a local ip). Beyond that I considered running it on my old razer blade but that thing is so power hungry and relatively large by comparison.

Running on a Hetzner instance right now and it's working super well. The only real problem is the browser thing I mentioned. But, I have a number of other always-on services on Hetzner hosts rn that I could consoldiate onto the mac mini and save some money.

[peak temple]

A $200 m1 mini from marketplace works fine too it seems 😄

[nocturne zodiac]

I want to run Minimax M2.1 on it locally.

Probably will need 2x or 3x of them eventually as Peter pointed out in my x post, but I don't mind it being a little slow if it means that digging through all my files and whatnot is done completely privately and not being sent to Anthropic/OpenAI. Also will be nice when I'm parked in a forest and can't get good starlink, so all my home automation will still run.

[edgy current]

meanwhile people on X are uploading their raw genome and unredacted tax returns to claude code 🤯

[astral steeple]

Running great for me on a $200 mini PC Linux box

Browser control works fine

[timber vessel]

Yeah I was having some good luck on my hetzner box with the general browser functionality. It works surprisingly well

Anyway I gave in and got my Mac Mini tonight, hahaha. So pumped to get stuff moved from the hetzner box to this thing

[tepid surge]

i am once again asking if someone has a working limitless ai pendant, and wants to help me integrate it with clawdbot by building a custom app, let me know so i can reverse engineer it

[umbral raft]

Hi
Can anyone recommend a cheap VPS server with no hassle , i tried hetzner but they rejected and disabled my account account even though I provided them a valid government id proof. Curious what you guys are using?

[ocean stag]

You can try https://exe.dev or https://cloud.ionos.com/compute/cloud-cubes or https://us.ovhcloud.com/vps/

[prisma stirrup]

running minimax 2.1 reap on Beelink Gtr9pro - 11.67 tok/sec, 305 tokens, 1.96s to first token, Stop reason: EOS Token Found. Works okish, but only around 50-60k token window

[nimble zenith]

Interesting. In theory it should have way to less ram to keep minimax and a reasonable context in memory.

[sterile sparrow]

anyone tried setting it up on synology nas yet?

limited processing power but there is a processor in there for sure

[prisma stirrup]

tbh that's anothre rabbit hole I didn't dive into yet

[green tulip]

I set up clawd on my old 2016 Era MBPro but a lot of skills don't like running on Pre-Sonoma Macs

[jolly creek]

Same issue here on a 2013 iMac with Ventura

[umbral raft]

Thanks I went for OVH and working like a charm. Loving clawdbot, so much power. I don't think I will use claude UI or claude code for web anymore, owning your assistant is so much powerful.
I do notice that the discord bot randomly stops responding after some time, I have to tell my telegram bot to start it again. Not sure if anyone is experiencing this.

[visual chasm]

Have a look at the free Oracle VPS - i can't vouch for the "no hassle" part, I am not using it myself, but I know people who do and it seems like a good deal (well it's free after all...) ARM-based ones come with 24 GB memory https://www.oracle.com/cloud/free/

[ember oxide]

my clawd is actively looking for a mac mini m1 on marketplaces since it complained about being on a pi 😅

[astral steeple]

genius

[exotic warren]

I setup clawdbot on my M4 mbp but should probably have a dedicated plugged in host for it is migration easy

[harsh dock]

+1 on this question. I have synology as well

[fierce sage]

is there a way to run minimax_ai on old mac's with intel processor ?

lm studio doesn't work on those macs

[sterile sparrow]

My only real problem getting this all together on a synology is my lack of Docker skills. Might have to spend a few hours this weekend

[nimble zenith]

Dont bother, not enough RAM, to slow.

[white kite]

didn't see anyone else mention but I have been running it on umbrel (https://umbrel.com/) it has 16GB RAM, so plenty, and works well as NAS too

[crimson seal]

Can anyone explain Clawd bot I’ve been using Claude code on my Mac but I have a hackintosh with 128gb ram doing nothing and a 8gb video card. Any use

Is this more a software that uses say open router and an API

[white kite]

https://docs.clawd.bot/ has info, there is a #old-help for specific Q? also you have a spare 128gb hackintosh doing nothing? get it to work for you 😮‍💨

[crimson seal]

Thank you!

[crimson seal]

Yeah 128gb doing nothing

[solid spear]

@copper birch #general message did you make progress on this one? mine just arrived 😈

[copper birch]

no because im not home yetand also i'm overwhelmed a bit 😂

[solid spear]

aight man. imma get hacking. ill drop repos too

[astral steeple]

yes, a lot of use! that'd be a great box, esp for local TTS/STT

[astral steeple]

you have a remote AI agent, do you not?!?!

[copper birch]

no, still lives on laptop but i will soon

[astral steeple]

🤦‍♂️

[sterile sparrow]

I have an old unused laptop that will probably have to step up and act as clawd home for now. When I'm happy with the setup i might get a real home for it on a server somewhere or buy a tiny little computer for it. Buying a mac mini sounds too expensive for just this as the use

Maybe in two years time when we have local models good enough a mac mini might actually be worth it 👀

[harsh dock]

What you doing running an umbrel 😂

Good to know, have one of those.

[white kite]

Why not? It works great! (Touch wood 😂)

[harsh dock]

Since it’s c related and we don’t talk about that in this neck of the woods. They’re amazing tools imo. Evidenced by my 1 minute timeout for even saying it haha

[nocturne zodiac]

Just a heads up, I've had one of these since they originally shipped and the battery does not last long if you refresh more than a few times a day. You'll want to have it plugged in all the time

[crystal hinge]

Do you guys not really care about mobile access and/or opening up your home network? I run clawdis on AWS lightsail because of not wanting to open up my home network

[jolly creek]

tailscale!

[crystal hinge]

ah yeah i set up tailscale while trying to have my macbook as a "node" and AWS lightsail as the true remote. It was pretty cool!
didn't end up getting that to work fully though T_T.

[deft sluice]

I’ve given clawdbot an rpi4 with a touchscreen/mic/speaker, and vibe coded a node client for it (my gateway/bridge is running on a Debian vm in my homelab).

Going to get clawdbot to build out a “home hub” on it.

Anyone already going down this path?

[astral steeple]

sort of prior to finding clawdbot. I made a visual thing that monitored sound output so it could "pulse" to give the AI a "face" of sorts

[scarlet relic]

whats the cheapest and usable hardware with voice control?

[steep osprey]

If I try clawd on my Mac Mini, can I move it to my Raspberry Pi at a later time?

[jolly creek]

Yes

[steep osprey]

Nice. I wasn't sure if it would be a complete reset or what to expect.

Still struggling to install on the Mac Mini. Thanks @jolly creek

[steep osprey]

Any value in running this on a Synology NAS as opposed to my Raspberry Pi 5? Not sure how resource hungry clawdbot is.

[ember oxide]

its running just fine on my pi4b

[steep osprey]

Nice. The only thing I’m using the Pi5 for is a Pihole, and Tailscale exit node. It’s under utilized.

[rancid sentinel]

i run it on my synology nas

since today

but i upgraded the ds220+ to 10gb ram at least..

if your pi5 is unterutilized, go for it 🙂

[steep osprey]

thats a lot of ram!

I need nvms for mine. spinning discs are my bottleneck these days

Thanks for your thoughts @rancid sentinel

[steep osprey]

Today I learned my M4 Mac Mini takes nearly the same idle power as the Raspberry Pi 5.

[nocturne zodiac]

Isn't that wild? Definitely uses more at full usage though!

[twilit radish]

I'm setting up my old phone for linux. It has some 2.5gb free ram & memory of 40gb.

[ember oxide]

any1 selling a cheap mac mini in vienna :D?

[steep osprey]

Right but that’s only happening when I’m on it, and then my monitor is using power etc.

[nocturne zodiac]

The difference between something at 10W vs 100w is a huge difference for me since I’m running fully on solar +batteries. I love these m-series chips!

[steep osprey]

Sounds like your off the grid

[dark surge]

@tiny grail I need to get some sort of Mac for human reasons. Could you tell me more about what your ideal shell to dwell in might be? Like is an M5 chip gonna make a big difference over an M4 chip?

[twilit radish]

Things work! But too slow(5-6mins). Need to keep the phone on charge most of the time.

[solid folio]

Don’t think it will

[glass finch]

Amazon has 16gb M4 Mac Minis for $499

Grabbed one to hopefully run a lesser GLM model for my day to day tasks

[devout blaze]

you would think the m1's and m2s would be cheaper by now

can get an m1 for 300 on amazon or ebay but only 8gb

[ember oxide]

cheapest i found for an m1 with 8gb is 300 €

[devout blaze]

8gb is just too low imo

might as well just run it on a rpi5

[ember oxide]

8gb must be enough, i was running clawd on a pi4b with 1gb just fine

[glass finch]

Could I run a local model on my mini and then set up an agent to call Opus only when it needs the horsepower? How is that even done

[ember oxide]

you could do that with subagents

https://docs.clawd.bot/tools/subagents#tool

something like that

{
  "tool": "sessions_spawn",
  "task": "code me a dashboard",
  "model": "anthropic/opus-4-5",
  "label": "code"
}

[glass finch]

I need to figure that out. I guess I can ask clawd to figure it out for me eh

[ember oxide]

you got it!

if you give him the docs it will be even easier

[nova topaz]

Connected Clawdbot to reTerminal E1001 Monochrome ePaper Display from Seeed Studio (the open source TRMNL); cleaning up the package https://x.com/amadad/status/2011153833120919584?s=46&t=u84c6uacccEm_E8DegTDKQ cc: @whole salmon

[astral goblet]

@amadad via Twitter

Ali Madad (@amadad)

Flashed my reTerminal @seeedstudio to give @clawdbot a home off Telegram

**👁️ 28 **

FxTwitter · Jan 13, 2026, 7:10 PM

[stable fossil]

does anyone have any docs on setting up gateway on a vps, and having additional nodes like mac?

[glass finch]

https://youtu.be/SaWSPZoPX34?si=OJom8GmulMOQ6qF5 start here maybe?

[bright ridge]

can’t wait for my reachy mini

even more excited to add clawd to it

[deft sluice]

Update: running as a “node” was fiddly, I’m moving to “rpi displays a page that clawd hosts/controls”.

Clawdia has been displaying lobster themed haiku, amongst other things.

[compact trench]

https://www.raspberrypi.com/news/introducing-the-raspberry-pi-ai-hat-plus-2-generative-ai-on-raspberry-pi-5/

I actually don't have a Pi5 yet, currently running clawdbot on a Pi4 with great success. But this seems fun to experiment with. It's not a super powerhouse of course, but for a lot of home automation things or just tinkering it sounds like fun !

[unreal talon]

I am!

Code will all be open source

[dawn wigeon]

got mine yesterday and also hope to somehow connect them 🙂 probably Friday or Sunday though when I've time to assemble it.

[rancid sentinel]

for a first sandbox like system with a claude code api key via microsoft foundry or awm, would you pick A- mac studio B- AWM server C- VM ? what is a good company system for easy fluid access?

[unreal talon]

Cooking up for the desktop hardware

[unreal talon]

If anyone wants to keep up with my progress on the RN app + desktop and watch hardware, ganna be posting everything on my X account @ jonahships_

[fluid jackal]

so has anyone gotten MiniMax2.1 deployed on a local setup without major issues? eyeing one of the unsloth quants on my 240GB VRAM obscenity

[whole salmon]

This is amazing https://x.com/buddyhadry/status/2011924389336412176?s=20

[astral goblet]

@buddyhadry via Twitter

Buddy Hadry (@buddyhadry)

Now when I’m not at my machine, and my Meta Raybans are busy vibe coding a modernized version of TempleOS, I can reach out to my @clawdbot, Clawtto (@Clawtto_bot) on any of my Alexa devices around my house.
︀︀
︀︀Here’s how it works:
︀︀
︀︀A Go polling script watches my Alexa conversation history for a keyword, “Clawtto.”
︀︀
︀︀To trigger it, I say something like: “Alexa, this is a message for Clawtto. Hey Clawtto, check my agent sessions. Alexa you can just ignore this.”
︀︀
︀︀Once the poller spots the keyword, it sends the entire message as a hook to clawdbot.
︀︀
︀︀Clawdbot receives the message along with instructions to use my alexa skill. The workflow:
︀︀
︀︀- Evaluate and work on the request
︀︀- Prepare a shortened audio response with eleven labs
︀︀- Convert the audio format for Alexa
︀︀- Pipe the audio back through to the Echo via https server
︀︀
︀︀About ten seconds later, Clawtto’s voic…

▶ Play video

FxTwitter · Jan 15, 2026, 10:12 PM

[whole salmon]

You need at lleast 30k$ hardware for large enough context. I tried.

[unreal talon]

Breadboard complete 😁 powers on and LCD, speaker and mic all work. Time to get voice wake working now after I eat dinner. After that, improve the app

[unkempt pivot]

Man, can't wait to see this working

[unreal talon]

Should be working by tonight! Now I need to save money for a 3d printer lol, forgot about that part 😂

[oak flax]

thoughts on https://www.kickstarter.com/projects/m5stack/stackchan-the-first-co-created-open-source-ai-desktop-robot as potential embodied clawd ?

[fluid jackal]

My mere $15k is too far off 😂

[bitter cipher]

for it to be equivalent to the hosted models?

[unreal talon]

React Native app for the hardware I’m working on

Inspo from the current Clawd mobile apps

[unreal talon]

Finally got the hardware to connect to the app 😁 time to get voice wake and hot response working then redo the UI

[void gulch]

What's everyone working on?

[unkempt pivot]

Social Media analytics dashboard for my clients!

[unreal talon]

Ay that sounds lit! I just kind of made something similar for my agencies client portal but it’s for Meta Ad analytics

[unkempt pivot]

I'm struggling a bit on api connexion as I have 0 skills in dev basically so a lot of iteration with Codex ahah ! Glad to see I have some colleagues around

[unreal talon]

Are you trying to connect meta?

[unkempt pivot]

Meta, TikTok, LinkedIn, X and YouTube

[unreal talon]

Meta API and oath are honestly very weird, haven’t tried the other ones yet tho

[unkempt pivot]

Lots of iteration and waiting time to get API approval!

[unreal talon]

Gatta fix some things still, for actual convos eleven labs streams the chunks way to fast and causes it to be choppy

Even throttling barely helps

Never mind, I got it perfect lol. Now only issue is the session. I want to keep it as one session so you can ask questions and continue a convo, but some reason it’s liking to reanswer the past few questions again aswell and the new question. @whole salmon if you have any ideas I’d love to know

[polar quiver]

Here I'm enjoying clawdbot with my Meta Raybans - anyone else using this combo? Key is to have him named in an easy to say way, I use "chuck", that won't get confused with other contacts. So yeah now I can be driving and say "hey meta, message chuck on whatsapp.... add html output to the NFL project..." then 30 seconds later "message from chuck.... hey Oli that's now done". Works exactly how you would hope!

[bitter chasm]

Is it recommended to leave it running on a old MacBook or rather on a raspberry?

[unreal talon]

If anyone wants to checkout the source for the hardware I’m building, you can check it out here: https://github.com/Nebaura-Labs/mote. It’s fully Open source with all the code, the hardware needed, and diagrams for the hardware. Going to upload videos later of it working irl

[thorn sleet]

I have it running on an old M1 macbook w/o any issues. Have a friend running on raspberry pi with no issues as well. Do what works for you!

[unreal talon]

Cool feature ima ganna try to add in a few after recording some demos is integrating the cron jobs. So make a cron like at 2pm everyday read me newest emails or something then the cron triggers the hardware and reads them out loud

[unreal talon]

Set hardware up to be a node now, now it can scan and use your home network to run commands and stuff. Had to update the node tool tho to support my own node tools

[rose shore]

@unreal talon nice so esp32 enabled connected device to your local network can you be used as a node to the whole network ?

[unreal talon]

Yessir! Right now I use it to control my Roku

[rose shore]

Ah so it just controls the Roku or is the Roku also its entrance point to the network so it can also see other devices etc ?

[unreal talon]

The esp is the entrance so it can see anything on the network

But the esp mainly is a voice controller

[empty oyster]

Would love to see the videos when you are ready

[oblique knoll]

https://youtu.be/1UDsWkcQZhc mmmmh interesting 🦞

[unreal talon]

Just finished building the PCB Board, have been working on it for like 5 hours 😂

[devout blaze]

it really depends on what you want to do with it. if you want to code and have it spin up a bunch of docker containers with databases and heavy resource usage applications, then a mac will be better suited than a pi. if you just want to chat and update your calendar, pi is fine. im running on a Hetzner server for now, but might add a max in later.

[unreal talon]

Prototype for when final PCB boards come in, just a rough draft to see how everything fits (final will look way different, bare with me I haven’t done 3D modeling in like 4 years and Autodesk was making me mad 😂)

[vast saffron]

That’s super smart (and cool) Idk why I never thought about this. I mostly use telegram for clawd but might move to WhatsApp because of the tighter integration with my meta ray bans.

[green sluice]

I think I got carried away halfway into designing a case for a Clawdbot Mac mini. Might call it Clawdy MacClawdface.

MacClawdface because Mac, get it?

[light frost]

running an old macbook 24/7 does not seem optimal

I have done it before for a different reason, it's ok for a little while if it's all you have, but it will generate a lot of heat and take up space. an rpi is a better option, but it depends if you want to have Clawdbot control Mac-only software.

I'm running Clawdbot (and a bunch of other things) on an rpi5 w/ PoE HAT so it just sits in a cupboard with an ethernet cable attached

[fierce sage]

any one got the mac mini setup documented ?

[unreal talon]

Redesigning the enclosure for the open source Mote hardware. KiCad exporting the PCB as a 3d model too is so clutch

[unreal talon]

Head almost done, just spent a half hour making a frame for the screen to block off the dumb black bevels they add but its coming together good

[violet warren]

Jarvis is getting a new body

[unreal talon]

threw the head idea away, i feel like sometimes simpler is better and will also keep printing costs down. Now im trying to figure out where to put the PCB and charger. Thinking of PCB on the bottom tray that is atatched via magnets, but now im not sure how to fit the usb-c charger too. Engineering is not easy 😂

[green sluice]

Do you also work with Fusion or solely blender?

[unreal talon]

This is actually my first time doing a model for a print, 4-5 years ago I did 3d modeling for esports teams and low poly work for fun so this is my first time touching 3d modeling since. I tried Fusion and unfortunately I did not like it at all

I’m used to C4D

I’ll give Fusion a try again soon tho

[green sluice]

If you are currently unbothered, moisturized, happy, in your lane, and emotionally stable and you want it to stay that way, stay away from Fusion

Blender has come such a long way in the last couple of years and became more and more useful for designing real world designs, I'd stick with it and focus on it.

[unreal talon]

Yeah I’m def going to stick to to, I do not like the UI of Fusion either. I’m a huge UX/UI guy so if I don’t like it even a little bit I never use it again

[green sluice]

Fusions UI/UX feels like it is designed following guidelines from the late 90s that are still retrieved from magnetic tape to this day.

[grizzled tartan]

What spec did you end up getting? Looking to buy a Mac Mini asap

[timber vessel]

I got the base m4 except with the 500gb storage upgrade. Loving it so far!

[grizzled tartan]

Nice! I was going to the same plus the upgrade to 24GB. But feel it might be over kill 😂

[timber vessel]

Haha, yeah same. I always like to overdue the ram but I decided this machine was more about serving a specific purpose

[grizzled tartan]

Yeah think that;s what I need to get out of the habbit of haha! My OCD is killing me, base spec machine?!? haha. Going to pick up my Mac Mini tomorrow. Can't wait!

[compact trench]

Sometimes family is so great 😄 my sister-in-law's husband brought over a lenovo M700 tiny, i5, 16Gb of ram they were just throwing away at work. It's no mac mini, but it's free!

time to upgrade from the pi4 which is having a hard time

[grizzled tartan]

Can't argue with that! Pretty sure people use them for the home labs! Win win!

[daring field]

anyone running on a mac mini, do you recommend giving it a separate apple id? i would probably just share things with it that i want it to manage like reminders etc?

[jolly creek]

Yes

[pale tulip]

wondering this too, i guess this means you might make a seperate gmail or something for it also right?

[jolly creek]

Yes

[daring field]

i’m also considering just signing in with a unique apple ID for messaging and then changing the main apple ID to my own

seems like there are certain things like find my that may not work

[jolly creek]

Just do a separate one and share stuff to it you’ll save yourself a lot of headache

[red cloak]

What about user account? Just its own appleID and its own default admin account? or somewhat limited?

[bitter cipher]

I did this too so it is actually seperate and more independent. Also own GitHub account with an org and permissions I can manage.

[jolly creek]

i treat that machine as entirely belonging to my clawdbot, down to the apple id, the password, the github, the twitter, the chrome. every single thing on that machine is dedicated to my clawdbot, all my personal stuff is on my machine that it can connect to if needed

[round harbor]

all my personal stuff is on my machine that it can connect to if needed

Can you clarify how it access the other machine? SSH?

[jolly creek]

https://docs.clawd.bot/cli/node

[upbeat mesa]

thinking about pixel 10 pro doing debian vm with clawdbot

anyone drooling something similar?

[compact trench]

spent waaaaay too much time setting it up last night. proxmox so I can use it for other things etc. But now my bot runs on it and it's so much faster already 😄 the pi was struggling, especially with the new memory features

[grizzled tartan]

That’s sick! Cool to see there’s a speed improvement! I’m sure you’ll have many more nights spending waaaay too much time on it 😂😂

[round nimbus]

Is the consensus a base Mac mini is fine for this? My instinct says to upgrade to 512 24 but my budgeting tells me that’s a few months of a max sub

[trail sand]

base mac mini would be overpowered for this realistically, unless you're trying to run local models

[ionic rover]

There seems to be confusion on resources - for the 'base' setup, you'll need almost nil processing power.

If your plan is to code like a fiend / spin up lots of coding agents and crunch numbers or data, then match the machine you'd want running your code, which will vary.

[trail sand]

right and even then, i don't think you need all that much compute tbh

personally i'd try first with any old hardware you have lying around

[copper birch]

https://repebble.com/index

just ordered this 🙂

[green sluice]

one pack of Red Bull, a tiramisu, and half an attempt to call Autodesks HQ to complain about their UI/UX decisions later, Clawdbot MacClawdface Mac mini case looks more like it should.

BUT: I have unanswered questions about the anatomy of Clawd. To the biologists here: How many legs does Clawd actually have? And the stubs on each side are the claws, right?

[unreal talon]

I was about to do some more 3d work yesterday and was wondering the same thing about the legs

Should be 8 legs and then the 2 on side are the claws

[green sluice]

Thanks!

[copper birch]

oh god pls one for mac studio !!

[main palm]

this is dope.

[green sluice]

How hot does the exhaust of the Mac Studio get?

[somber sparrow]

I think now it’s time when @whole salmon announces his clawd ready hardware (not every one owns a mac mini just now) 😶‍🌫️

[jolly creek]

you can run it on a $3 vps and let it connect to your personal devices for information and browser and be just fine, you don't have to buy a whole mac mini

[green sluice]

Clawdbot MacClawdface is now biblically accurate with 8 legs

[solemn radish]

why would someone want to buy $1k in hardware for this?

[jolly creek]

you'd be shocked, ive been talking people out of it on twitter all day

[solemn radish]

consumerism is wild

@jolly creek what are you using this for? I do a ton of work that I could use a virtual assistant for (heck - maybe even some coding) but wouldnt trust a virtual assistant for

[jolly creek]

oh god so many things, hold on

[solemn radish]

holding

[jolly creek]

https://chaoscat.win it has an art gallery that i told it to add a new project to at 3am every day for me to wake up to

[solemn radish]

interesting - what are you using for browser use?

[jolly creek]

built in browser tool

[solemn radish]

doesnt that destroy your token usage? are you just needing one claude max subscription? api pricing?

[jolly creek]

one max subscription

[solemn radish]

my guess is that anthropic is gonna butcher rates with this gaining popularity?

[jolly creek]

they'd be killing themselves too doing that

[solemn radish]

I think they're probably doing that already with all the usage 😂

[bronze creek]

I thought about running it on a vps but then I gotta use Tailscale to keep it connected to home

[jolly creek]

yep thats what a lot of us here do

[solemn radish]

whats wrong w running Tailscale

[bronze creek]

Nothing. Just overhead to maintain. lol

[solemn radish]

its pretty much set up once + forget though?

[bronze creek]

Well you need an endpoint on the vps and one at home

I run Tailscale on about 12 nodes. But that’s why I don’t bother with a vps. Just put it at home on any spare pc/vm and forget about it.

[frosty loom]

What would be the best setup to run this remotely? just a cheap hetzner vps? enough power for claude code to do tasks + also have browser functionalities etc?

[lofty laurel]

Are there reasons for wanting clawdbot to run on a mac mini vs. running on a linux VM?

[bronze creek]

iMessage and Apple apps probably

[lofty laurel]

Didn't think I saw iMessage as a way to communicate with it. I guess it can send iMessages?

Still learning 🙂

[bronze creek]

Yep

[trail hemlock]

How are people setting up dedicated macs for clawdbot?

[bronze creek]

Waste of resource is the consensus unless you have extra $$

[trail hemlock]

I had one laying around

[bronze creek]

You going to run gateway on it? By default it loads a local gateway. Was super easy for me to set up.

[trail hemlock]

Yea

It’ll be on my tailnet

[bronze creek]

Once you get it up and connected you can ask it about additional tweaks it suggests. It can control itself basically.

[trail hemlock]

Yea, I meant more like does it get its own Apple account, stuff like that

I think I’ll try it that way

[bronze creek]

You can set up a new iCloud account and give it access

[trail lodge]

I've got a small box in my closet running Proxmox and have enough room to toss 2 CPU cores and 4-5 gigs of RAM on a new VM for this, is that gonna be enough to run pretty smoothly? Not planning on asking it to code, but headless browser access could be nice

[high iron]

Simstim integration and you can control via telegram and not need mac

For iMessage

[bronze creek]

Yes

[prisma briar]

Do you know any proper solution to integrate with iCloud not in a MacOS environemnt?

[fallow meadow]

Bought a separate Mac mini tonight after reviewing this thread. Thank you!

[fathom mulch]

are you guys sure that even for parallel agentic tasks like scientific abstract/ paper search -> saving -> extraction 16GB RAM will be sufficient?

[primal sage]

so which claude subscription do ppl use to not bankrupt themselves with this lovely tool?

[unkempt pivot]

#old-models there was a lot of discussion about this there

[prisma patio]

https://x.com/i/status/2012074849293005100

[astral goblet]

@i via Twitter

M5Stack (@M5Stack)

Meet StackChan! 🤖 A community-driven, open-source AI desktop robot by M5Stack is NOW LIVE on Kickstarter!
︀︀
︀︀Discover your StackChan on Kickstarter now:
︀︀www.kickstarter.com/projects/m5stack/stackchan-the-first-co-created-open-source-ai-desktop-robot?ref=299xh3&utm_source=twitter

**💬 5 🔁 53 ❤️ 215 👁️ 42.6K **

▶ Play video

FxTwitter · Jan 16, 2026, 8:10 AM

[copper birch]

yes

[prisma patio]

Not quite worth it for me but could be a very cute terminal for clawdbot

[arctic musk]

Sooooo cool

[arctic musk]

I am about to start with this

I feel like the true power of clawdbot is accesed if yiu have a home lab with a NAS

And allow it to access and manage your stuff hahaha. Anyone tried this out ?

[tranquil hedge]

I don't see it mentioned here, but for people that would like to use vm and don't like to pay for it, Oracle Cloud has always free tier and you can get 4cpu 24gb ram 200gb Ampere instance (arm) free forever. I had instance for last few years but now it belongs to clawd.

[green sluice]

That‘s what I‘m currently building, a proxmox HA cluster with proxmox backup server and a dedicated truenas storage node, primarily for local AI/agentic workloads. Truenas and ceph storage is not yet set up.

[mental stump]

hey folks! 👋 new to clawdbot, loving it so far

noob question: I have an old macbook pro (~2020, intel chip iirc) collecting dust at home. I saw that everyone is using mac mini for clawdbot, but I was thinking the MBP would work too? Not sure if there's anything specific re mac mini or it's simply that you don't need input hardware so it's a good fit.

thx in advance and have a great weekedn!

[compact trench]

nice!

I'm new at the proxmox and all and am out today at a friends house, wanted to show him my bot but something is amiss at home 😄 no connection. Will have to debug when I get home. Might also be the unit I got did have flaws. Will see 😄 but man, kinda just wanna go home and see what is up 😄

[green sluice]

Older MacBooks are definitely an option. See here: #showcase-old message

[arctic musk]

Crazy shit. Its my first Linux/home server lab. Yours feel years away

[silver glade]

This is so rad! Looks like a mishmash of mod10, rackstack, and mini rack? With aesthetic screw covers and that color scheme is so nice. I've spent 15 min scouring the web for more details on this setup. I run proxmox in a boring p520 lenovo thinkstation and now I want to chuck it and build one like yours. lol

[arctic musk]

Would definitely love to learn more about the setup and have such a nice design as yours hahaha

I love this builder energy bc you get humbled like evry 10 mins by people with crazier ideas

🤣

[silver glade]

It def shows as one of those projects where a persons varied interests collide. Design taste, photography skills, modeling, printing, devops, industrial engineering. so cool

[arctic musk]

Yea man same I burnt thtu my token limits and had anxiety I wasnt improvign the setup 🤣

[green sluice]

All custom. Designed and 3d printed every panel/mount myself from the ground up, also built the frame myself, since no prebuilt frame/rack out there (including deskpi) follows the "official" standards (and I wanted my 10 inch rack to be deeper to have room for the PSUs and hopefully the UPS). You can check out my rack frame here: https://www.tinkercad.com/things/dWv8J49QeRj-benm-10-inch-homelab-rig-v4-12u?sharecode=l-CYU8CWV6JhGeCoDsvKRuKP1VDIxbvuxf4SqQ8NXFs

[silver glade]

so awesome.
Not joking when I say I'm firing up my printers and hyperfocusing on a printed rack this weekend as a result of this img you posted.
Also, definitely wanting to scratch the itch of building a 3d printed rack drag-n-drop configurator where someone can just drop what blocks or layers they want for each 1u and it spit out a list of build files from a modular set.

[obsidian orbit]

This is my question also.

I’ve got a MacBook Air I’m thinking of starting off with also.

[silver glade]

It should be just fine. Mac mini is just a great bargain of compute per dollar in a small form factor that works for something like this + having a 'containerized' set of apple credentials to leverage it for those in that ecosystem already

[green sluice]

I plan to share all my models, files and guides on printables/makerworld/github in the coming days. The only thing thats kinda annoying is that I forgot to take standalone photos of all my mounts, panels, covers and fixtures before putting the rack together. So I have to disassemble the rack first to take photos to put the files online because I don't really want to print them just to take photos of them.

But I can already say: if you don't want to print your rack parts in ABS/ASA (I didn't because its nasty), you should go with Tinmorry PETG-GF. In contrast to most fiber reinforced filaments, this one from Tinmorry is actually stronger and more heat resistant. I wouldn't use PLA, it has a very low heat deflection temperature and under constant load even the heat inside a rack can make it deform, especially if you have some weight on it.

[lean smelt]

Question for the multi setup guys. I’ve got my local install rocking along on my Mac mini. I’m also lucky enough to have a dedicated 1u rack in a data centre where I work so have a dell server running proxmox that I tinker with. What would be a use case for spinning up an Ubuntu vm and chucking clawdbot on it ? Could it link up with my local build to manage the diagnostics when it stops working?

[arctic musk]

Man I already got a mini pc for clawdbot now you making me crave a 3d printer

[green sluice]

if you are in the US, better get a 3d printer before they are made illegal or at least nerfed by DRM https://www.youtube.com/watch?v=x0x1mQSkvmk

[arctic musk]

Whattttt

I'm in Europe but since we are kings of over regulating I may even be in more danger

https://tenor.com/view/chuckles-im-in-danger-ralph-wiggum-the-simpsons-gif-14149962

[arctic musk]

This is dope too btw

[woven rose]

Looks cool what’s inside

[trail sand]

i don't have multi setup yet, but i'm running clawdbot on proxmox in an ubuntu VM. I guess i'd just prefer an isolated environment for it to run where he can do whatever, and worst case I nuke the VM. May add a cheap apple device later for the apple ecosystem stuff

[trail sand]

very sick. you have any GPU compute in there? i'm running something kinda similar but even decent consumer GPUs don't really get you much unfortunately

[sacred spoke]

What are the best options for someone running on windows/Linux? Is the mac mini worth adding another OS to the mix?

[trail sand]

what would you need a mac machine for? do you use iOS

people really only are using mac mini because then it enables some skills to use apple specific things like iMessage

and because it's just a great machine for the price, but if you already have some hardware to put it on it's not necessary

[sacred spoke]

Thx for helping! Dont use ios. Running it on local desktop now

[hollow nimbus]

Anyone have success running this on Oracle always free tier? I have an instance there only running one small thing so thinking about hosting there.

[trail sand]

should be the same as pretty much any cloud vps right?

[hollow nimbus]

Probably. Sometimes different providers have use cases they like and don’t.

[trail sand]

someone a handful of messages above mentioned they were running it there

[west plaza]

What am I missing here on the oracle free tier defenders? Its slower than the mac mini? Which I find already impossible to run a local model… How can this ever run a local model? And if not this will burn through thousands of dollars of cloud llm right?

[west plaza]

Can you elaborate on this? I got a macmini but not powerful enough to run local models. So racking up anthropic and gemini bills. If not running local with buying the hardware, whats your view on running cost?

[tranquil hedge]

I am running it in OCI Ampere instance, works as good as any infra

[green sluice]

• 1 x Lenovo ThinkCentre M910x Intel i5-7500 | 64 GB RAM (NAS Storage Server with PCIe 4.0 6x SATA expansion card)

• 3 x Lenovo Thinkcentre m720q Intel i5 9400T | 64 GB RAM (main worker bees)

• 1 x Lenovo Thinkcentre m710q Intel Core i5-6500 | 32 GB RAM (Backup Server)
  All upgraded with M.2 A+E 2.5GbE Intel I226-V Ethernet NIC for PTP; Western Digital Red SN700 NVMe NAS SSD - 1DWPD 1TB as main storage.

• 2x Raspberry PI 5, redundant Stratum 1 PTP Grandmaster time server; with NEO-M8T GNSS TIMING HAT

• 1x Raspberry Pi 4B 2GB for UPS Master
  All equipped with M.2 A+E 1Gb Intel I210AT Ethernet NIC for PTP

• Unifi Flex 2.5G PoE

The panel with the knobs is for fan control, a fan controller card is mounted behind it. Like most features of this rack, it's unnecessary but fun and done because it is possible. On the left of it are push buttons for KVM switching (need to change that to 5 buttons). The m910x is not yet mounted because I'm currently installing the sata expansion card. What's also not shown is a 7inch kvm monitor that swivels forward from the left upper side of the rack. I'm reworking its mount.

[woven rose]

The rack itself looks pretty stylish, it’s something custom?

[green sluice]

Not yet, I thought about it but this was before the market went mad.

[green sluice]

completely custom

[bronze creek]

I run mine on a dedicated proxmox sandbox Debian vm. The benefit is full sudo access with testing tools locally for coding etc.

[woven rose]

Amazing, I like it 👀

[onyx glen]

do you guys have this running on your main pc/laptop or do you reccomend running it on a seperate one / virtual and just let it go wild

[unreal talon]

Hetzner server

[sacred jetty]

So I have two PCs, MacBook Air M4 and a Ryzen 9 5900x w 3070ti. Would you guys recommend I host Clawd on the Mac or PC?

[unreal talon]

Mac will give you more features like iMessage and stuff

All personal preference

[sacred jetty]

It has to be on 24/7 correct?

[solid folio]

i run it on a PC, works well

[unreal talon]

Yes

[solid folio]

you don't need imessage, telegram is much better

[sacred jetty]

Because I plan to build an NAS eventually if that has anything to add ?

Also if I host on Mac can I use through message like Telegram and it can access my phone as well?

[trail sand]

it wont be able to "access" your phone but you can message it through tg on your phone

[sacred jetty]

But I mean if it has access to my Mac then it can essentially do everything my phone can correct? Ex like messages, read emails etc etc

[unreal talon]

Yes

[sacred jetty]

If I eventually wanted it to have access to my NAS as well how would that work with it still having access to the Mac for messages

[unreal talon]

For email you can integrate Gmail api aswell, that’s how mine has access to my emails currently

[solid folio]

you really don't need a mac unless you want Imessage and that's it

well and apple notes, todos, etc

but you can use other tools

[sacred jetty]

Could Clawd interact with apps on Mac?

[solid folio]

if the mac is on, yes, using nodes

[unreal talon]

I believe Apple notes and stuff have an api? I know Apple calendar does

[solid folio]

https://docs.clawd.bot/nodes

[sacred jetty]

So going back to NAS, could it integrate with a NAS

[solid folio]

what are yoou trying to do, i think it should be ok

clawdbot will probably discover the NAS just fine

[trail sand]

you trying to run it on the NAS or just access the share drives?

[sacred jetty]

Man idk I’m just tryna figure shit out lol I’m just getting into NAS and whole homelab thing because I don’t want cloud subscriptions and stuff like that so I was just curious if clawd would be able to go into and search for a file or something if I asked it to, or summarize notes, etc etc

[sacred jetty]

Shared drives

[trail sand]

gotcha that's easy enough, if it's running on your mac you just mount the NAS drives and it should see it like any other filesystem it has access to

[solid folio]

yes, it can access your nas files, but it will depend on the config how you end up setting it up

you gotta experiment but i would assume is fine

[sacred jetty]

🤝

[trail sand]

both opus and gpt5.2 are very competent in setting up homelab stuff, you can use it to help

[sacred jetty]

As much as I want to, I don’t want to rely on it to much since I wanna make sure I understand what I’m actually doing instead of being fed the answers yk? Yes that means lot and lots of research but I find a beauty in it lol

[solid folio]

you can ask the agent to explain it to you as well

[minor hollow]

Fair to say you’re going to miss out on tons of skills/functionality running clawdbot on a non-Mac device?

[jolly creek]

Only because most of the people building things have been Mac users so far, that doesn’t necessarily mean that you’re missing out, just that you should build it

[upbeat minnow]

Anyone running clawd with raspberry pi? I'm thinking of doing so but a bit hesitant

[tepid turret]

If someone didn’t already mention it, go in person to Best Buy and they’ll price match Micro Center to get the Mac Mini for $400!

https://www.microcenter.com/product/688173/apple-mac-mini-mu9d3ll-a-(late-2024)-desktop-computer

[signal fog]

They didn’t price match me because it was out of stock. B&H has it in stock at $500 and BB price matched them.

[lofty laurel]

What do those running macs headless use to remote in?

[lunar herald]

What does linux miss out on vs macs? Running on a VPS

[bronze creek]

iMessage and Apple access etc

[trail sand]

there's a few of the built in skills that are mac focused but i think most aside from imessage there's a solution for on linux, it's just not as supported out of the box

[lofty laurel]

Do most with Apple hardware create a seperate ID for clawd?

[jolly creek]

Yes

[lofty laurel]

Via another Mac as a node can it read imessages?

[runic hill]

Im curious about this too. New apple ID for dedicated imessage and icloud accounts? or just a clawdbot user account under existing primary apple ID?

[wide python]

Is there a good tutorial on the hardware setup and specs required? Everyone seems to talk about having to get a Mac Mini for Clawd. Is that the case?

[bronze creek]

Simple vm, pc, pi, or Mac.

[unkempt pivot]

Can be on a vps too

[hollow nimbus]

Amazon has Mac minis for $500 right now

Just got one.

Honestly I was going to get one anyway. I figure with inflation in component costs it’s the best deal going.

[nimble wigeon]

Hi all, I’m thinking about doing Linux instead of macOS so I can make clawdbot customize the entire OS with more freedom. It would not be my main machine, but maybe I’d do some dev things there occasionally myself. Maybe switch off macOS someday if it feels right.

Has anyone set up clawdbot on Linux and tinkered with building /changing the OS in pretty custom ways? Curious on how that experience has been.

(And what hardware would you recommend for this if I’d like a decently capable app building machine, minimal local model running, some app building, maybe 2D gamedev)

[bronze creek]

Many of us use Linux only.

Raspberry pi is a good starting point.

Mine has full admin rights on debian vm

[flint wasp]

Is there a trusted docker image to use on hub.docker.com ? How to install this on Unraid docker?

[flint wasp]

Tried installing Clawdbot on a Raspberry Pi 3, no luck. only 1 gb ram.

[bronze creek]

I run it on a dedicated vm. I could probably make a docker container. I wrote a proxmox script today

Let me build a docker image later tonight

[carmine delta]

what specs for the mac mini are people getting will the base model suffice

[bronze creek]

Yes

I’m using a 2015 MacBook retina

[flint wasp]

node22 not going to run on a macbook pro 2011 right

[jolly creek]

https://tryitands.ee

[quiet gulch]

Hi all! 👋
Just getting started with ClawdBot. I’ve got it installed on my UGREEN NAS via a VM (4 cores, 8GB RAM) running Ubuntu with cloud-init.
I haven’t done much yet since I’m still trying to see how it performs and whether it needs more resources. That said, I totally get why so many people run this on a Mac with native macOS apps + AirDrop is way smoother than SCP’ing files into Linux or messing with SMB shares 😅
Excited to get this moving and see where it goes!

[loud turret]

nope

[cobalt ocean]

Keep us updated big dawg.

[bronze creek]

Dependencies are brutal. It’s over 2gb. Node and other features are just massive. You will be best doing a vm.

[warm pelican]

Currently running on an M1 Pro 8GB - it works, but it gets bogged down with any browsing with Chrome, safari is better.

I’m guessing 16gb/24gb is the sweet spot

Crazy interesting project

[quick estuary]

Thinkpad + Linuxmint + Clawd + You know how to code = AGI

[west zinc]

i would be inclined to use that new browser skill from vercel

for any browser related stuff

[cobalt ocean]

I'm just tryna run it on my vps

[rancid sentinel]

I have a question about the use of e.g. a macmini. if it would be second computer, setup for AI and clawdbot, how do you handle that beside your primary computer? Do you share keyboard/mouse? use the splitfunctions in modern screens? Would love to see how fluid it is to work on a second computer at the same time as on the primary.

[jolly creek]

For mine, I never touch the second computer. It belongs entirely to my clawdbot

[rancid sentinel]

so basically you just have this second system for clawdbot and you just touch it when you need to make fundamental changes?
I do it in a way right now for my clawdbot in docker on my NAS - but quite often I need to touch it then.
My idea would be more, a mac-os system i build up and use for all the vibe-coding/ai stuff including clawdbot, since on windows (my primary system) it is just suboptimal with wsl2 etc.

[jolly creek]

i mean
my clawdbot makes the fundamental changes too

i still do all my own work on my own macbook

[zinc lintel]

Do i understand correctly that as long as it can run linux and have 16gb ram i can run it effectively on anything ?

[rancid sentinel]

on windows -> wsl
on linux
on mac

easy, it runs fluid even in a docker on my nas (10gb ram)

[main palm]

For sure

[main palm]

Having it on a 2nd system that is not your main operating system is super convenient because let's say on a raspberry pi or mac mini it's a small form factor versus a big pc rig right? or if you have it on a laptop that you can clam shell out of the way

because having it 24/7 is the real superpower

[rancid sentinel]

yepp - but then it needs to be super convenient to work on this second computer, likely then via ssh? i mean if i would buy a mac mini i would love to work on it more; but im native windows.

[flint grotto]

is it possible to start with amazon E2 first then switch to mac mini later?

[jovial coral]

What version of Linux did you put on it? I also have a 12” MB but Mint was kinda twitchy, so I went back to macOS.

[main palm]

Yep, totally fine. Clawdbot runs on both Linux (EC2) and macOS (Mac mini).

To migrate later:

1. Stop gateway on EC2
2. Copy ~/.clawdbot/ folder to Mac mini
3. Install Clawdbot on Mac mini
4. Start gateway there
   Just don't run both simultaneously only one can poll Telegram at a time.

[green sluice]

Did I just spend a day optimising every overhang and angle of MacClawdface, which required rebuilding the snap lid, so it can be printed upside down to for sure save a ton of filament and printing time (I mean look at the feet, it's just logical!), only to finally realise all this shaved off merely 2g and 9 minutes of this 155g and 4 hour print? Yes.

Do I question my life choices? Also yes.

[bronze creek]

I run Sequoia with open core patcher.

[slim terrace]

Can I deploy clawdbot on a nuc with docker? Or do I need an actual Linux vm?

[bronze creek]

It doesn’t have docker becuase of the number of dependencies. It’s not a simple stand alone app

It’s an ecosystem. Run a lxc or vm

[slim terrace]

Gotcha, gonna be a Lxc on proxmox. Are the specs anywhere on the site?

[bronze creek]

Want to test out my proxmox script I made yesterday?

Https://www.github.com/isriam/ProxmoxVE

Weird links don’t work

[slim terrace]

If there is a helper script for it that’s great. I opened the repo on GitHub let me check

[bronze creek]

You can use 4cpu 8gb 8gb for a lxc with Debian

I wrote a helper script and submitted it as a PR but they rejected it lol

[jovial coral]

Domo arigato. Looking into it now.

[bronze creek]

the macos release .dmg on the github repo was a game changer. the cli clawdbot i could never get remote gateway working, i dug into it for days. it seems it requires a ssh silent tunnel, which it doesnt tell you until you set up the .dmg package

i use remote gateway obviously on a linux box, i wanted macos just for a node

[inner zinc]

I came here wanting to know this too

[green sluice]

I'd recommend separate Apple ID, you can add it to your icloud family if you want. Especially at the beginning until you get the hang of it. Always keep in mind that clawdbot can do what you can do on a computer, theoretically at least. And it can make mistakes. You don't want it to accidentally send imessages or icloud-emails in your name to random contacts or mess with your icloud account, iCloud drive. Make it its own email account.

[noble bone]

Hi where can I host Clawdbot for free?

I have a Raspberry Pi but would want to start it somewhere in the cloud first, any offers with free credits besides AWS?

[bronze creek]

oracle free tier

[north basin]

why not in the pi?

i'd guess locally is better

also home ip address

[noble bone]

how much ram does it need?

[north basin]

i think almost nothing

[noble bone]

thats my pi resources

[north basin]

as it relies on an external ai for the processing

[north basin]

you are more than fine with it

[noble bone]

okay thanks man, i will give it a try

[signal fog]

Damn BB raised prices of the mac mini to $799, it was $599 yesterday!

[noble bone]

why is everyone talking about buying mac minis for clawdbot? To run LLMs locally? Otherwise it would be a complete overkill right?

[distant rock]

Just saw it, they mentioned the app has to be 6 months old.

[green sluice]

mostly because this way, as an Apple user, you/clawdbot can access and interact with Apple apps and services

[bronze creek]

Same as running it on any pc like Claude cowork. It can use your environment.

[north basin]

can i have access to all the benefirs from my macbook if i install it on another mac device?

like menubar, voice overs, etc

or would be the same hosting it on a linux box?

i have both servers at home, and i use a macbook personally

[hollow nimbus]

Maybe a silly question here guys. But I have a vm with Oracle Free Tier that I’m considering using to host our crustacean friend.

Any pros and cons to using it on a headless instance like that v an OS with a GUI like MacOS?

[jolly creek]

Host it headless, connect it to tailscale, let it connect to your machine if needs to

[bronze creek]

if you install a lot of apps and skills it will be tough, but with general usage its great

[rocky fable]

This is probably a dumb question but on twitter everyone seems to be buying Mac minis for this. Why don’t they just run it on their current primary computer?

[coral mauve]

Im curious if is there any benefit to having more than one instance? (one physical machine i.e. raspberry pi) like do you see it as parallel instances that would "work together" in some twisted way or like is it like since we're anyway just managing context and actual ai comes from cloud it makes little to no difference?

[thorny siren]

You don't want to keep your primary computer on 24/7, also you should not keep a beta bot in your main machine.

also you don't really need a mac mini, im using a 150$ mini pc with ubuntu server. You only need a mac mini if you care about iMessage / apple ecosystem stuff.

(But it's true that on MacOs the setup is a tiny bit easier)

[rocky fable]

Ok thanks

I don’t even have a desktop I just have a laptop with an RTX card

[stray comet]

best windows equivalent of a macmini?

[plain crypt]

What low cost used Mac mini is best for this?

[tardy bison]

Can I run clawd on an older Mac Mini M1?

[bronze creek]

Depends. I haven’t found many use cases except macos

[jolly creek]

Easily

I run it on a iMac from 2013 and it runs just fine

[little schooner]

I know a few people run it on their Synology NAS and also on a Raspberry Pi 5. My NAS is pretty old but I do have upgraded the RAM a bit. DS920+ with 20Gb RAM
Is it worth running on it? Or just use the Pi 5 B - 8GB

[storm mulch]

any tips to maintaining macbook pro always awake? even with amphetamine on, sometimes it is slow to respond as if it sleeps

[green sluice]

see here: #showcase-old message

[unborn dust]

@whole salmon please reply my DM

[unkempt pivot]

Man I think he has like 100s of messages, that’s not the place to do that

[whole ferry]

Is there a reason people are choosing a local machine (mac or otherwise) over a VM? Seems like a hetzner instance would work well here, no?

[bronze creek]

Everything works

[whole ferry]

What do you mean @bronze creek ?

[bronze creek]

Local machine. Vps. Mac. Vm. Lxc

[whole ferry]

Ah, sorry, i interpreted wrong. Every option works, not "everything works better locally"

[bronze creek]

Correct. All have pros and cons

[hollow nimbus]

I think buying Mac Mini isn’t a bad deal at all. Component prices are through the roof. The base model is a tremendous value for the money.

Now, is it necessary for clawdbot? No. But it’s a great option considering how little power it uses.

[graceful charm]

https://x.com/danupsher/status/2015525210280099996?s=46

[astral goblet]

@danupsher via Twitter

Dan U (@danupsher)

Fuck yeah

**💬 2 ❤️ 2 👁️ 45 **

FxTwitter · Jan 25, 2026, 8:40 PM

[graceful charm]

No need for mac mini 🙂

[vague dock]

memory gonna be costly

not what my buddy at apple says

[exotic elbow]

I'm working on vibe coding web apps. If I'm looking to set up a VPS, what specs should I look for? If I ever want to move to another VPS or local, would it be an annoying process? Thanks

[wanton cobalt]

for running local LLMs whats everyone opinion on best bang for buck for a $5k setup (rtx5090 ?) , what about 10k,20k,30k?

[green sluice]

I love claudes personality

[bronze creek]

Buy a server.

[green sluice]

and I dont know why but somehow I read clawdbots responses from this screenshot in a southern accent ya boi gotta pretty wyld seddup yep

[wild jungle]

If I am wanting to use this to become an expert sports analyst, post and create content across 3-4 platforms, possibly vibe code, and work on cold emailing should I get a Mac Mini with the $200 per month Claude

[exotic elbow]

Got any windows recommendation? Not a SWE

[bronze creek]

I buy Dell enterprise versions lol. Large form factor. 1 cpu. R720 or T440.

[bronze creek]

Yes

[exotic elbow]

lol, need something more budget, if it works, then I might consider getting a more $$$ option xD

Is it annoying to move from one place to another with all prevous memory and conversations intact?

[bronze creek]

Mine cost 550. T440 on eBay b

Then run proxmox and vms and lxc the world is now yours to own.

You can export the folder easily enough. It’s just text files.

[green sluice]

it basically runs on everything, even raspberry pi, but a mini pc/mac mini is solid. Depends on the type of content you want to produce and with what.

[nocturne zodiac]

I love that this channel started out as "Let's give clawd all these external hardwares for it to control" and now it's like "Should I buy a mac mini?!?"

[wanton cobalt]

fr, just run virtualbox and install it there

[wild jungle]

I’m not super techy so I don’t know what to believe on social media. That’s why I came here thought you’d provide best specialization. If I can have a multi tasking, proactive, learning clawd running with Claude max with something cheap like “virtual box” idk what that is then let me know the BEST option.

But I’m founding a business so willing to spend on hardware if truly better long term.

[wanton cobalt]

https://www.virtualbox.org/

[elder basalt]

How do I know if i should get a machine for this or not. I’m so confused. I see people saying to do it in that VM’s are not the way to go and then I see people saying the opposite.

And now I just see Peter replying this

[wanton cobalt]

it allows you to spin up virtual machines on some other system, so it doesnt have full access to everything

there is a difference between running it on a VM in the cloud (with is a server instance of ubuntu) vs virtualbox, which you can install a desktop version ubuntu on

[scarlet scaffold]

What I noticed is that you have two choices.

1. you buy a mac mini and then host the LLM locally something like ollama.
2. you get an VPS and then you pay for the api tokens.

That is how I see it. I might be wrong here

[runic hill]

people are not hosting local models and base m4 mini

[elder basalt]

Exactly so I’m confused

I’m not a tech bro but I’m tryna spin up a few agents for simple workflows for my marketing agency that stay running, I only have a m4 MacBook Pro. Idk what to do

[wanton cobalt]

you need 2 things:

1. some where to host/run clawd (the bot)
2. some service to do inference of LLMs

for #2, generally you should just use LLM services like claude, openai (chatgpt), minimax etc

you can host your own local models, but you need expensive GPUs and they are not going to be nearly as good as the cloud models

for #1 , you can reformat some old junk laptop to host it, just needs to stay online all the time. or you can buy a cheap VM service (But there is issues with this) which is there is no desktop gui for most cloud vms

[wild jungle]

Yes! I want to use Claude max for best coding ability. My question pertains to #1. is running it on Mac mini worth it if I want to have it doing a lot for a long time. I don’t want it on my main computer is that what the other option is

[wanton cobalt]

right, you dont generally want this on your main computer since thats a huge security risk. it does not need to be a mac

do you use mac now?

[wild jungle]

Yes I’m not super tech savvy so I thought maybe keeping it Mac would make it easier

[charred pine]

Can I run clawdbot on android/termux

[cursive plank]

im running it on my rraspberry pi

which works great

and cheap

[wanton cobalt]

i saw a tweet about that

[charred pine]

Let me search for a similar tweet then

[wanton cobalt]

someone got it working

[charred pine]

I got it working

[hoary ferry]

Thinking of renting out a Hetzner VM to trial out my usage of it before buying dedicated hardware and connecting to my Claude Max sub, anything else I need before getting started?

[jolly creek]

no thats the perfect way to do this

https://docs.clawd.bot/security

[hoary ferry]

🫡 thank you!

[meager mist]

Does Hetzner ask you for passport for verification?

I tried signing up and its asking for passport upload for verification. Is that a common thing? (Sorry, never signed up for vps before)

[midnight hawk]

When hosting this on my local machine, do I need a lot of SSD storage? 1TB enough for a new piece of hardware?

[meager mist]

That's a whole lot of storage, you'll be fine

[midnight hawk]

thanks, and im assuming I can only use Ubuntu or MacOS? Windows doesnt work or can I still have windows and just use WSL2?

[meager mist]

windows and WSL should be good.

[midnight hawk]

last question, sorry : I would use it for agentic coding, are there any integrations to use CursorAI (and their available models) or do I have to use Claude Max? and is Claude Max enough for someone like me who codes aggressively, multiple projects at a time?>

[meager mist]

Im not an expert on those integrations but Ive read you can manage multiple models, people use minimax, codex, opus whatever. check the chat in models for more info on it.

[midnight hawk]

awesome, thank you for the help 🙂

[candid plaza]

🚀 Excited to release vap-media on ClawdHub!

Now Clawdbot generates stunning images, videos, and music using Flux2 Pro, Veo 3.1, and Suno V5.

https://clawdhub.com/elestirelbilinc-sketch/vap-media

Feedback welcome!

[pale kestrel]

would love to know how you managed to install this on TrueNAS...

[frank timber]

I installed it on my NAS via docker no issue so far

this simply docker-compose file did it:

x-common: &common
  build:
    context: https://github.com/clawdbot/clawdbot.git
  volumes:
    - ./data/.clawdbot:/home/node/.clawdbot
    - ./data/.clawd:/home/node/clawd
  environment:
    HOME: /home/node
    TERM: xterm-256color
    CLAWDBOT_HOME_VOLUME: /home/node
    CLAWDBOT_GATEWAY_TOKEN: xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

services:
  clawdbot-gateway:
    <<: *common
    init: true
    restart: unless-stopped
    ports:
      - "18789:18789"
      - "18790:18790"
    command:
      [
        "node", "dist/index.js", "gateway",
        "--bind", "lan",
        "--port", "18789"
      ]

  clawdbot-cli:
    <<: *common
    init: true
    tty: true
    stdin_open: true
    entrypoint: [ "node", "dist/index.js" ]

then ran the setup --wizard manually to configure credentials etc...

[elfin stone]

finding any limitations? I have a ton of them lying around

[stray comet]

Possible to run clawd on an iphone 13?

[jolly creek]

No, it does need a computer to rent on for the primary Gateway

[floral sedge]

hi, are there any technical limitations on Clawdbot capabilities by installing it ona linux virtual server (aws) vs a local Mac?

[unkempt pivot]

Nope

[floral sedge]

Awesome... Will use Hetzner then

[bitter cipher]

You are limited to the hardware it has access too. That's all.

[cursive plank]

ope

nope

had some issues first with browser

that works now

[polar quiver]

One potential issue is that when retrieving data from the internet a lot of sites "protect against bots" by blocking non-residential IP addresses. In the past I've found that when scraping/reading sites if you are doing it from a VPS it gets blocked, whereas from home/local machine you are on a 'residential IP' and so that doesn't happen.

That's not all sites but it is a problem at times, just in case anyone using a VPS for this finds "internet research" isn't working so well.

[cold vale]

I'm trying to get a Mini Max v2.1 229b running locally and stable. It's looking like $8500ish in hardware is the best I can come up with. Anyone have other recommendations?

[little schooner]

It had to happen..

[scarlet scaffold]

Apple is currently experiencing limited availability of the Mac mini because broader chip and component supply constraints

[rare canopy]

I'm curious, why does the mac mini seem to be the preferred hardware for clawdbot? Because of the easier integration of apple stuff (notes, reminders etc)?

[little schooner]

I think the idle power requirements.. plus the ease of setup may be

[frank gyro]

same here. pi5 8GB that lived in a cupboard for a year forgotten!

[cursive plank]

same lmao, and i love it

[frank gyro]

setup was super nice and easy, learning curve on JID in whatsapp was new to direct it to different ones. also, port forward from pi to anywhere i want to access dashboard is neat!

[little schooner]

Was just looking there is a newer 16GB variant. At what point the cost outweighs the value and think Pi is expensive?

[cursive plank]

just get the 8gb or 4gb

[frank gyro]

Honestly I have not had an issue with the 8GB for nearly 2 years. I ran multiple honeytokens and servers on it, now clawdbot ... more than enough

[little schooner]

Fair point.

[cursive plank]

my clawd bot isnt even using more tha 0.10% of the cpu

or more than 1gb ram

[frank gyro]

yeah that is just silly 😄

[cursive plank]

loll

[knotty narwhal]

Hey all, just landed here, what system and specs are the best for the agent to run on?

[cursive plank]

raspberry 5 pi 8gb ram

[frank gyro]

Super minimal, pi5 or pi4 with 4gb or 8gb ram is more than enough from what I am seeing. Anything like mac mini is astonishing overkill if just for this 😄

[cursive plank]

real

i also got a mac mini m4

but i dont use it

im using my macbook m4 pro more

[knotty narwhal]

Will spin off a VM for it on my server, so was wondering what OS works best and how much resources I should give it

[cursive plank]

so i maybe will run it on my mac aswell

[cursive plank]

uuuh

unbuntu

and

6gb ram

and uh

[frank gyro]

A friend of mine went to get mac mini because VMs within his nuc were causing issues hahaha

[cursive plank]

like 5 cpu cores

[frank gyro]

Aye, Ubuntu, simplest way

[knotty narwhal]

Running an older think centre with an series 7 i5 and 32gb RAM, hosting a proxmox System there where I host my vms

[slender cedar]

Hi all 👋 - Is there a recommended documentation for setting it up on a raspberry Pi 4 model B 😊

[floral sedge]

Another newb question: I need to setup a clawdbot for myself and my partner.. I should be able to setup the 2 accounts on 1 Hetzner VPS account right? Just create 2 users?

[versed plaza]

every incoming telegram message is tagged with the user Id, it should be ok, and each user will get its own session if you don't switch it manually so even easier to maintain the context

[wet kelp]

Just use the regular onboarding document and treat it as a vps/with Tailscale.

[boreal fern]

Apple Notes on a Linux VPS Machine
I am hosting clawdbot on a VPS linux machine. My main machine is a macbook pro. I want clawdbot to be able to read/write Apple Notes. I understand that I should connect my macbook as a Node to the Gateway to do that, but the drawback to that, I believe, is that the Mac would need to be always on/awake. My use case, though, would be to use clawdbot when on the phone (iOS as well btw).

[elder relic]

iCloud integration is pain in the ass but it is possible via api. Ask your bot to research solutions. I have some scripts that synchronize iCloud Drive on Linux to use obsidian vault. Super slow but it kinda works

[brave egret]

If you find anything , please let me know. 🙂🙋‍♂️

[versed plaza]

you should just work, or at least it did for me

install/run onboard to register the model providers/add skills and register external services like telegram/discord/etc...

[boreal fern]

iCloud integration is pain in the ass but it is possible via api. Ask your bot to research solutions. I have some scripts that synchronize iCloud Drive on Linux to use obsidian vault. Super slow but it kinda works
@elder relic I am interested in the iCloud drive script to syncrhonize the obsidian vault. and why not github to sync the obsidian vault?

[elder relic]

not github because obsidian supports iCloud on iOS. Right now I am using some python library but playing with rsync at the moment. All written by claude code. Nothing stable yet.

[bronze creek]

you can run clawd in docker but you need to import so many packages its really large, like 3gb

[forest sapphire]

Why are people going out and buying Macs or paying to run it on a VPS when it runs just fine on a spare laptop or raspberry pi?

[bronze creek]

because they are apple users

[violet ridge]

It lines up with clickbait, becasue people like the idea that they are 1 Mac Mini purchase away from having all these great features, And ignore how much context information and work you need to really put into it to get it to that state.

[teal girder]

any issues running local models on the raspberry pi?

[graceful charm]

https://www.s22server.uk/ here is a setup guide for if you wanna run an android phone as a server to run clawdbot. saves you buying new hardware.

clawdbot made the website and its hosted on the phone server right now

most android phones should be similar setup

[past rain]

idk but maybe cause we got money and/or no one wants to accidentally end up like the signal guy plus i want to use the iMessage skills

just a peace of mind tbh

[forest sapphire]

The extremely low token per second speed would be issue number 1.

[teal girder]

got it

[forest sapphire]

If you're running this on a raspberry pi you would want to use hosted models like Claude, GLM or chatGPT.

Or be running the local model on a computer with an actual GPU.

[teal girder]

thank you!

[swift arch]

on friday i was literally looking at buying dedicated hardware as a personal server, looked on apple refurbished mac mini's, ebay, hetzner, etc. now all mac mini listings on apple refurbished are gone lmao

[mild matrix]

stupid question, why aren't people deploying on things like railway? Any reason why people are opting for physical hardware?

[forest sapphire]

Any reason why people are opting for physical hardware

No monthly fee. Complete control. Clawd runs on pretty much anything made in the last 10 years so unless you are a kid that has only ever owned one computer in their entire life it is trivial to throw Debian or Ubuntu on an unused PC and get it running.

[sterile walrus]

my case: integration with mac and heavy software, otherwise i woult only use cloud.

[mild matrix]

lol my old homeland have xen running on

Good point

Good point

But imo railway is so dirt cheap anyways

Just need to see if this violates TOS

cheaper then my electric bill

[jaunty heart]

I'm installing on an old macbook pro - decent specs, reformatted.. Taking quite a while. Normal?

Seems like it's all the dependicies due to MacOS 13

[lyric girder]

Any potato level HW still needs a GPU of some sort to prevent going bankrupt from token expenditure. 😄

[flat breach]

Have you tried setting up clawd with various services? You can certainly set it up on a VPS but with poor ergonomics

[gentle jackal]

So I have an m4 and 3xM2's... should I buy another m2? Wait for the m5? Wild to see everyone catching on to MMs finally

[gentle jackal]

M2 for $250**

[forest sapphire]

You should not buy another mac just for clawd

[gentle jackal]

Lol not clawd

[flat breach]

Anyone have thoughts on a Apple homepod mini/Google Home alternative device?

would love to have a voice awake hardware around the house that has access to clawd with privileged voice access

[gentle jackal]

Honestly don't see why I would even want Clawd? Idk what people are really even doing with it

[flat breach]

Are you curious? or are you just hating?

[gentle jackal]

But its brought a ton of attention to Mac Minis

Genuinely curious

[flat breach]

Have you tried setting it up?

[forest sapphire]

https://tenor.com/view/where-do-you-think-we-are-scrubs-meme-gif-20456434

[gentle jackal]

I did, briefly, last night

[flat breach]

With what?

https://x.com/dodeja/status/2015495630601539617

https://x.com/dodeja/status/2015496451275850201

[astral goblet]

@dodeja via Twitter

Akshay Dodeja (@dodeja)

I have been using clawdbot for many things but the most impactful workflow has been call transcript automation.
︀︀
︀︀Every call → transcript saved locally → workflow runs based on type
︀︀
︀︀Sales calls: post opinionated summary on slack for discussion with AE.
︀︀
︀︀Customer calls: log feature requests + bugs, expansion/contraction signals
︀︀
︀︀Built in a few hours. Saves 2hrs/day. Better than what my EA was doing in the past.

**❤️ 2 👁️ 147 **

FxTwitter · Jan 25, 2026, 6:42 PM

[gentle jackal]

In my terminal on my m4, set up iMessage and fell asleep haha

[flat breach]

You can orcestrate anything and everythign

https://x.com/steipete/status/2014869890281156763

[astral goblet]

@steipete via Twitter

Peter Steinberger 🦞 (@steipete)

Someone hooked up @clawdbot to their oven.
︀︀
︀︀category: what could possibly go wrong

**💬 57 🔁 41 ❤️ 902 👁️ 98.3K **

FxTwitter · Jan 24, 2026, 1:16 AM

[gentle jackal]

Got a laugh out of this

By orchestrating everything... like work stuff? Creative stuff? I write scripts and weird short stories and that's what my friends and I talk about besides like movies and stocks and pics of our kids

Here's one: could someone text me a movie title or imdb page and it'll find a torrent and send it to my seedbox?

That doesn't even sound like it needs Claude really but if Claude made it super easy and robust...

[flat breach]

yeah emphasis on the last thing you said

If it doesnt click for you, then try using it for existing use cases that you have

if you find your self using multiple tools to do something, like stiching things together, try it with clawdbot

Out of the box Clawdbot is LLM with systems instructions tuned to be assistant like -- So not much different than claude code -- but it can become a powerful assisstant that can do pretty much everything for you

[brittle lava]

for android enjoyers, can share if interested: #showcase-old message

[copper birch]

i'm working on an app for using clawdbot with meta glasses, almost done 🤓

[brittle lava]

sick! is there a way to summon with voice?

[solid folio]

Didn’t you tweet you shut off your clawd less than a day ago?

[prisma flax]

So good people of Clawdlandia i setup on a VPS all is awesome started hitting context overflows by days end aaaaaaand seems like Claude API spend is going hit its limit fairly quick lol. What's everyone using for LLMs just curious?

[patent obsidian]

still happily running beastmode, full root on my M1 Max 16". I could easily get the mini but it seems easy enough to just keep my mac on??

[atomic canyon]

I keep it on Haiku during setup or it bonks out fast. Especially if I call notion

[prisma flax]

Thank you

[prisma flax]

You're running locally or VPS

[maiden slate]

Pulled out my iMac Pro, backed it up, and wiping it now...

[livid hill]

Ive just been running clawdbot on my macbook air, but i wanna set it up on its own pc so I was wondering what was better out of windows and linux for this?

[void thistle]

Linux. Windows is spyware

[livid hill]

Fair enough

Thx

[ashen locust]

Should I install on my old system of 5800x, 32gb ram, 1Tb nvme system, an old xeon workstation with 152gb ddr4 & 1080, 1TB sata, or just buy an mac mini m4 for $500 and sell the other parts?

[wet kelp]

You want to run a small contained process on that?

[ashen locust]

ig it doesn't do linux, downside there. maybe i should start with what I have in hand (the 1080 workstation with 152gb of ram)

[solid folio]

You can run this on a raspberry pi

[ashen locust]

oh

[jolly ingot]

Are there any additional security risks using a VPS?

[gusty sandal]

yea a few first it is a public so make sure you have auth. a lot of people running clawd they have theirs unauth and you can access it easily

just make sure you follow secure "protocals" and you should be fine tho

[jolly ingot]

Ok, thanks

[gusty sandal]

i saw you were looking into a macmini if you really want to do that I would suggest waiting till june

macmini m5 will come out and it will be a lot faster for local LLM

[maiden slate]

Greetings, any ideas on how to create a freash Apple Account without getting a new phone number?

[gusty sandal]

are you talking about setting up a clawd to your macmini?

or is this completely different?

[gusty sandal]

ngl @jolly ingot im planinig on doing this

[maiden slate]

Yes, I dug out my old iMac Pro and wiped it, Got a new gmail but trying to create a new Apple Account fails

[torpid siren]

vultr works for clawdbot?

[gusty sandal]

what you can do is use the same icloud and set up a new user with limited permission. it will be more organized

[maiden slate]

I appearenly need a unique phone number NOT associated with any other Apple Account.

[gusty sandal]

in that new user set up clawd and you're set

[maiden slate]

Great Idea, THx Diego. I'll look into that.

[gusty sandal]

with no way of clawd "hacking" or messing up your icloud

[jolly ingot]

bet thanks

[half helm]

ok let talk costs Day 0 ; $200 /mo openai $600 macmini and API costs for connections

it seems the $600 for the macmini is a fixed cost - so what can be done to recover any of it

[gusty sandal]

i think an important part on cost is what is your use case, if oyu plan on makign advance scripts i think the openai/VPS would be better

[hollow coral]

proxmox that bad dawg and make a linux VM for this and now you have a little home lab 🙂

[prisma flax]

You will use tailscale si effectively all access will be closed off, but there are a afew headaches with redirecting web console via ssh and setting up tailscale etc.

[hollow coral]

tailscale / twingate / cloudflare warp I think it is... lots of options... but don't leave it open to the internet 🙏🏼

[jolly ingot]

Is Hostinger a good place to setup the VPS or AWS?

[prisma flax]

Hetzner too

[hollow coral]

+1 for Hetzner

Digital Ocean.... so many options

[prisma flax]

Oh yes almost forgot about digitalocean

And old Linode now Akamai nit sure if they have any deals they might

[bronze creek]

Apple accounts - use Apple Music webpage to create them.

[bronze creek]

@maiden slate

[maiden slate]

Thx JW I’ll explore

[bronze creek]

I created one tonight that method

[full jackal]

Think this will be enough lol https://www.amazon.com/MINISFORUM-Pro-370-Desktop-Computer-Graphics/dp/B0F47G69MJ/ref=sr_1_4?sr=8-4

[ancient hornet]

what is this

and which is the real one

[tender geyser]

@bitter cipher where is your ansible repo?

[bitter cipher]

not public.

[tender geyser]

got it

[white slate]

totally unnecessary.

[full jackal]

Considering claude can't run a build without crashing WSL 90% of the time, I needed that RAM upgrade

[white slate]

Windows is super bloated.. I just found a 10 year old Windows laptop, put Linux on USB with Rufus, booted from USB, and wiped Windows.

[full jackal]

Nice. I bought the minisforum mini pc for more than Clawdbot. For vibe coding and also daily driver/work

Might dabble with local models

[white slate]

ah ok. mine is just Clawdbot. So, now that makes sense.

[fading jacinth]

hello glad to be here

[prime aurora]

what are the recommended vs specs for running clawdbot?

[vast gale]

Is it usable or does it turn into a headache? I was going to pick up a 8gb M1 mac mini tomorrow.

[warm pelican]

Totally usable

[warm pelican]

I’m now using it as a node - you’ll love it

[vast gale]

Thanks

[scenic sorrel]

is a raspberry pi 4 suitable

[jolly creek]

Yes

[scenic sorrel]

i worry it would shut down or overheat but i'll get it setup and report back

[unkempt pivot]

No worries @scenic sorrel loads of people are on pi

[short gust]

How much ram do I need for just clawd.bot?

[unkempt pivot]

Not much @short gust

[short gust]

Okay cool I have yet to install, just gathering info

[versed plaza]

also doublecheck that the power adapter can do 3/4A, the phone ones are usually not enough, If clawd starts compiling something it could get stuck/red light

[spiral sierra]

Hey im new to this, what would be the benefit or running it on my miniPC, it has 16g of ram, Windows. Or just using VPS like hostinger or oracle ?

want to set it up today

[versed plaza]

all of the above work, local is better imho

(not sure about the status on windows but you can use a linux container)

[unkempt pivot]

Both work, if it’s your main pc only uptime changes if you shut it down

[spiral sierra]

DM me

[spiral sierra]

so using my minipc would be better, thanks.

[unkempt pivot]

#introductions

[vague ruin]

I'm sorry, I thought I could help friends in need, and I didn't intend to post an advertisement.

[fleet light]

Any meaningful difference on using, say, a n95/8gb/256gb mini pc vs a n150/16gb/512gb mini pc?

[spiral sierra]

interested in answer too

[green sluice]

power consumption, N95 15W tdp, N150 6W tdp. Besides that, the igpus are tiny bit different but that doesn't matter imo. For a 24/7 homeserver I'd choose the n150 because lower power/lower heat, that is why one goes with that type of CPU in the first place.

[spiral sierra]

Is there something clawd cant do on VPS that it can do on local.pc?

[polar lark]

I'm fully setup on Hostinger and it's running great! I had to adapt the Hetzner instructions to work with Hostinger... I also setup a node to connect to my macbook when I wan it to connect... So yes.. Hostinger works great.

[flint spindle]

Had a auctioned server lying around at Hetzner doing nothing, so - that's where clawd lives now. Totally OP'd lol

[strange locust]

Hey! Do I have any benefits to install on a Mac Mini vs Windows computer running Ubuntu (or any other bash distro)? Are some skills exclusive for MacOS instances of Clawd?

[green sluice]

on a Mac you have easier access to Apple services such as iMessage, if you want to use iMessage instead of e.g. Telegram

See here: https://docs.molt.bot/help/faq#do-i-have-to-buy-a-mac-mini-to-install-this

[strange locust]

Well would use discord, so a Linux would be more than ok I guess

Then node part seems interesting but not my real use case at least for now

[deep nexus]

a lenovo tiny with i5 7400 and 8gb ram is enough for start ? thanks

thanks

[zealous monolith]

I have a gaming pc with rtx4060ti, thinking of wiping it clean and giving it to clawdbot (moltbot) but I think the experience is inferior compared to using a mac, anyone here running it on pc? anything I should know? thanks

I am am fully in the apple eco sys, but don't care about imessage etc, would telegram but I heard someone saying it won't be a full fledged experience, can't find that comment anymore

and perhaps things might have changed too

[pastel yew]

Mac Mini

[zealous monolith]

of course would be ideal but I don't have the funds for a new mac rn

[pastel yew]

How much are they going for now?

I had a spare brand new one lying around

[thin jay]

🙄

[acoustic tapir]

I thik WhatsApp, Telegram and iMessage provide the same type of experience. The issue is encryption, which WhatsApp doesn't truly provide. Not sure about X Chat, but that may be interesting

[zealous monolith]

not that much, 400-600, 599 brand new i think

[golden dawn]

I am thinking to get 2 Mac Mini M4, and run each as an employee, 1 for coding, another for Research and Study. Is there way to have these 2 communicate to each other?

[unkempt pivot]

Both in the same group

In WA or Telegram or Discord

[golden dawn]

Using Slack

[solid folio]

you don't need 2, you can use one piece of hardware for multiople agents

[golden dawn]

You mean 1 device can do 2 things simultaneously?

[solid folio]

you can have multiple agents, yes

you don't need 2 computers

and oyu certainly dont need a mac mini

you can use any piece of hardware

https://docs.molt.bot/concepts/multi-agent

[golden dawn]

Thank you so much

[solid folio]

please don't buy a mac mini

[golden dawn]

Why? I am mac user

[solid folio]

I mean, if you want you can

[golden dawn]

I prefer isolated enviroment

[unkempt pivot]

An old Mac is enough

[solid folio]

but please dont buy new hardware if you have other piece of hardware laying around

[golden dawn]

I dont, and I got $500 gift card, so I was like why not use it for this lol

basically is free if not consider tax

[solid folio]

you want an honest opinion?

[golden dawn]

Yes, please

[solid folio]

if you haven't used this and not sure how it works, I'd start with a vps, learn how to use this. LEARN HOW TO MAKE IT SAFE. learn tailscale

spend a couple of bucks per month

if it works for you, then think about spending real money

[golden dawn]

Ahh, I see your point. thats actually good, idea, I guess hetzner it is

$5 server coming up 😄

Thank you @solid folio

[solid folio]

of course!

[normal beacon]

to avoid any hiccup on the desktop we daily using, it recommend to setup at another device is it? or there is a way to keep our daily used pc and this moltbot running parallel?

[unkempt pivot]

@normal beacon better with another hardware or vps imo

[normal beacon]

ok.. thx
but if using vps, it can't work on some task that is on the desktop right?

[exotic ocean]

Docs are available at https://docs.molt.bot.

[unkempt pivot]

There is everything in the docs

[solid folio]

it can if you use nodes. but i'd do that only if you know what a security breach that could be potentially

[normal beacon]

thanks dude..

[rancid sentinel]

would you buy a 2024 (2year old) Apple Mac mini 2024 M4 512GB SSD 16GB RAM with invoice but i think with running out warranty (which is 2 years) for a discount of 660€ instead of 949€

[unkempt pivot]

For Molty?

[buoyant perch]

Hey y’all. I’d love to run Molty on a VPS in a macOS setting. Anyone have any good recs?

I’m mainly looking to supercharge my iOS development and do other Apple related development.

[potent patio]

This makes a ton of sense. What kind of VPS specs would help to run Moltbot?

[unkempt pivot]

If it’s for Molty no need to get such a machine… an old one works the same

[spark ginkgo]

Is there a way to "transfer Molty" to a new machine?

[rancid sentinel]

my "molty" is running already on my NAS in docker.. but i would like to have an ios system at home beside my windows computer; to have a full moltly machine and native ios

to be honest running docker and windows in wsl gives me at least some headache 😛

[solid folio]

any VPS. hetzner is cheap but you can use others as well

[rancid sentinel]

and on hetzner, do you run linux?

[solid folio]

yes

[potent patio]

Like, this could run Moltbot? Run it well? All the inference stuff is offloaded to the model?

Why are people running it on Mac mini hardware then?

[lofty laurel]

For access to messages, apple apps.

[slender cedar]

Managed to set a Raspberry Pi 4 Model B 8GB up using tailscale and Signal as the endpoint. Nice!

[potent patio]

Huh.

Gotta say that seems overkill. Using a mac mini just for apple apps and imessage

[runic hill]

deleted msg. I see the announcements channel.

[distant cedar]

I have a Pi 5 and i have run local LLMs on it, but super slow, no GPU. Looking into a mini pc of some sort to run clawd bot and other things. local LLM runs fine on my mac, but i want some dedicated hardware for clawd. trying to narrow down what i should buy

[buoyant perch]

Some people are using it for development in apple's ecosystem.

[potent patio]

In similar boat. I have a 5090 and a modded 3070 with 16GB VRAM, but looking at more powerful systems. NVIDIA dgx Spark seems interesting, but pricey. Slightly less pricy you have various AMD strix halo setup - minisforum MS-S1 Max looks interesting

Of course, this is absolutely the worst time to buy new hardware. But for all we know, it could be the best time for the rest of the foreseeable future 🤷

[sharp apex]

Once the big players crash and burn you'll be able to find GPUs for 95% off 🙂 depends on whether or not you're willing to wait

[green sluice]

unfortunately most current AI server components can't be plug-and-played into consumer-level environments.

[green sluice]

Even if you were able to buy data center hardware, it wouldnt make much sense for individuals to use it. price/performance/power-consumption ratio is a decisive factor. The power consumption of this hardware alone would make it uneconomical. Macs are currently and probably will be for the foreseeable future the best option for local LLM, also because Nvidia never cared for power consumption of their products the same way Apple has.

For example: the Nvidia Jetson Nano, this tiny low-powered SBC, draws 4 times as much power as a Mac mini at idle(!) while offering less than half the performance for 2/3 of the price of a Mac Mini M4.

A maxed-out Mac Studio M3 with 512GB RAM/VRAM draws around 350-400 Watt under full load. A reasonably comparable Nvidia setup with 2x 4090 and therefore only 48GB VRAM draws already double as much power with around 700+ watt.

An Nvidia setup with 512GB Vram draws 30-40x as much power at idle and at least 5x as much power under full load, not including the dedicated cooling you will need. Of course you get 40-50x the performance with a 6x RTX 6000 workstation but that also cost you at least 4x as much to buy, 10x as much in running/electricity costs, around 5x TCO. This is an absolutely unreasonable setup in terms of performance-overhead, cost, heat and noise for a single user and even a small business with single-digit users sharing it will have a hard time taking full advantage of this system. And for that server you even have to consider if the power circuits in your house or office can handle this beast and the required infrastructure.

[spark fractal]

Anyone successful with strix halo? How is the os with self hosted or is it meant to be Claude first?

[cunning bear]

I’ve been trying to set up a LXC only for Clawdbot in my Proxmox environment on my home network (without GPU 😄 ) for two days now and I’m struggling. I managed to get the Telegram bot working, but then I wanted to add a web UI for better management. Unfortunately, that didn’t work. Then i tried reverse proxy with nginx and the subdomain was connected and i could see the UI but it was not paired. After a few rounds of frustration, I decided to delete the LXC and start fresh. Now, I’m struggling with the installation, which was supposed to be the easy part! Maybe it’s because of the renaming to moltbot?
Does anyone have a installation guide for your own homelab proxmox environment or should i use a VM instead of an LXC?

[spark fractal]

mine is running on lxc 22.04, no issues

[cunning bear]

When did you install it?

[spark fractal]

let me grab config, not much

[spark fractal]

yesterday

[cunning bear]

Yes, yesterday the installation worked, but today after the renaming, it makes errors and errors. I will try again.

[spark fractal]

vanilla:

cat /etc/pve/lxc/10x.conf
arch: amd64
cores: 2
features: nesting=1
hostname: clowdbot
memory: 2048
net0: name=eth0,bridge=vmbr0,firewall=1,gw=omit,hwaddr=omit,ip=omit/24,type=veth
ostype: ubuntu
rootfs: nvme1:vm-10x-disk-0,size=500G
swap: 0
tags: agent
unprivileged: 1

[sharp apex]

So I've got LM Studio running locally and trying to get clawdbot to talk to it. I'm running into a 300s timeout -- seems to be something about the TCP protocol itself. When the model takes longer than 5 minutes to generate a response (which is all the time for me) the connection between gateway and lmstudio closes and it dies. Does anyone know how to get around this so I can get a response back?

[distant cedar]

the usual way to resolve this is to have the response streamed. usually the full response may take a long time, but it should send something while it is thinking

[sharp apex]

makes sense. So if I can force lmstudio to send status updates or something...

[verbal wagon]

What about dgx spark?

[lofty laurel]

For Clawdbot purposes - 16GB of RAM or 24GB of RAM? Would 24GB make a difference for browser tabs etc... or not so much

[tender geyser]

Do you have any suggestions for our own moltbot ansible playbooks, from your experience?

[vital peak]

do people suggest getting the 24gb 512 regular m4 mac mini or is the mac mini pro really ncessary if youre going to run 24/7 have it doing goding and automations and porting things to other computers

[sharp apex]

The minimum specs are plenty for running moltbot itself. If you want to run local models or other apps on the machine that's when higher specs would be important.

[vital peak]

thanks

[tender geyser]

the mac mini isn't necessary, but Peter uses a mac and codes for mac, so you get all the wizbang on one. Add to that, security wise, MoltBot is safer in its own environment (where your personal stuff is not present) and you get a run on mac minis.

[sharp apex]

welp I managed to find a model small enough to process and start streaming a response (at least the first response) within 5 minutes, so I can confirm everything else is working. yay! Whether this model is enough to do anything useful, I don't know. But at least the mechanics are working as long as the connection doesn't time out. The trouble, especially if I run a more reasonably sized model, is that my hardware isn't able to even process the prompt in under 5 minutes. So the model doesn't even start streaming a result before the connection times out. It seems like there's no easy solution to this problem, but maybe I can use ollama (since it's open source) and hack in something to keep the connection alive while the model is processing.

[green sluice]

spark is a dev kit, its meant to be a miniature version of Nvidias big servers to allow developers to test/simulate their software that is supposed to run on Nvidias infrastructure. Spark is not intended to be an alternative for a Mac Studio or so.

I wouldn't say no to a spark of course but if I had to decide between a mac studio and a spark, personally I would take the Mac.

[lofty laurel]

That's a great way to understand it. Always wondered what the point of the Spark was.

[lofty laurel]

If you had an Apple trade-in credit and had to pick between M4 Mini models as a Molt only device - which would you go with?

[green sluice]

what do you want Molty to do?

[lofty laurel]

@green sluice Would be scraping webpages, building a PKM, transcription, researching for now.

Have a 5090 in a gaming machine I could tap into for local models if I needed and have alot of NAS storage. So I guess really it would just be for access to Mac apps and to orchestrate.

[green sluice]

the more ram the better usually, and thunderbolt is pretty fast for external storage, like external nvme ssds. The model with M4 Pro and Thunderbolt 5 allows the external storage to be even faster than the internal storage.

But you could also upgrade the internal storage of a Mac mini later if you want. There are kits available. So the only thing you can't upgrade is RAM and the M4 chip itself. Therefore, I would favour more RAM over a big internal ssd. Save on the internal storage and put that money into RAM. 512GB internal SSD should be sufficient, 256GB should be ok as well. But do more than 16GB RAM.

[fiery venture]

You can use a Pi 5 16GB if you know how to use terminal. Install on a Ubuntu server

[midnight wind]

Interested in setting this up but im having some difficulty setting up rish. Could you offer some advise?

[azure charm]

Hello every1. Not a tech savvy, actually not nothing related to computers. Im a psychology student, but im planning to learn more about this stuff cuz i bet this will be our future. So im thinking of buying an mac mini m4 24gb 10GPU. This would be enough right? Also any tips from where i can start to gather knowledge is welcomed. Ty

[sharp apex]

Hello @azure charm we were just talking about that. It all depends on what you want to do. You can run Moltbot on virtually any hardware. But if you want to run local models or do other fancy stuff you might want more power.

[chrome glacier]

is there a hardware faq somewhere that I can read before asking questions?

the way I read this is that I could run moltbot on my nuc, and have any gpu operations go off to my linux ollama server

[azure charm]

Thanks alot for the reply

[karmic delta]

Just got the bot running on an old RPi 3 with a 64gb sd. One trick was turning on the memory swap during the setup. Thing is chomping through my Claude API credits...lol
export NODE_OPTIONS="--max-old-space-size=2048"

[distant cedar]

I’m thinking if this to run local llm https://www.microcenter.com/product/699561/gmktec-evo-x2-ai-mini-pc

[green sluice]

depends wehat tasks you want your local llm to do

[distant cedar]

Trying to decide on that vs something with less power, not really sure what tasks I want to do

[hard walrus]

any downside of having Molt on a vps?

[distant cedar]

I’m trying to avoid paying for tokens on an external llm

[small surge]

interested in this too

[coarse saffron]

So is anyone here using that clustering stuff where you can run multiple minis on one model?

[sharp apex]

I happen to have an AI PC (Intel Core Ultra) and today I am learning that it's not the best hardware for running local models. Evidently the NPN parts of the chip are not hardware that is supported by LM Studio or other similar tools. GPUs are still the way to go. That said, my AI PC is happily running models right now. It's not that it can't do it, but it isn't fully optimized.

[distant cedar]

What kind of models work and have good respond time?

[sharp apex]

I'm still working on that. I have tried a dozen or so models and most of them work fine when prompted directly, but I have no idea which ones are going to perform best with Moltbot on my hardware.

[distant cedar]

And is there a better hardware choose for $1600 or less

[sharp apex]

I know you'll be limited by RAM on that AMD machine. Bigger models will want more memory, between the GPU and CPU depending on how you run it. But I don't know much about what translates into good performance yet. (on my own machine I'm setting up and seeing what it can do, haven't gotten enough experience yet to think about speed)

[green sluice]

you mean exo https://github.com/exo-explore/exo

I expect we will something interesting from them at the upcoming wwdc

https://www.youtube.com/watch?v=bFgTxr5yst0

[coarse saffron]

That is the one that I saw. Not sure if that idea is better than a studio. You get more chips if you just buy a bunch of mac minis. Rather than one machine. If you can truly stack them the question is, how may do you need to match online model performance?

[coarse saffron]

| Setup                   | Tokens/sec (70B-ish) |
| ----------------------- | -------------------- |
| $50k Mac cluster + RDMA | ~10–20               |
| 1× A100 80GB            | ~30–60               |
| 1× H100                 | ~80–150              |
| 2× 4090s                | ~60–120              |

I don't know if that is accurate. If it is it doesn't really speak well of even the clustering deal. But I just pulled that from chatgpt. I guess if they start adding more chip capabliities with m5 or m6 the rdma stuff will help the networking not be an issue. But are we a generation or 2 of minis away from being able to realisticly use them for local models reliabilty? Even with clustering.

[verbal wagon]

thats exectly

what im trying to figure out

like my plan is to use local llm built on my data

and to connect it with glm 4.7 flash locally

and molt should be like the "headbrain" i guess im still didnt figure out how to setup and what is the right way to setup things

[green sluice]

the more powerful each node, the better, generally. Not every model can be sharded to run on an exo-cluster or delivers the same performance when used this way, at least currently. It requires optimization on the model side.

But yes, rdma, I expect Apple will show some neat stuff at WWDC. Besides that, the current base M5 is already 300% faster for AI workloads compared to the M4 equivalent. So the M5 Pro, Max, or even Ultra will probably shake up the industry.

[verbal wagon]

for my idea
also thanks for the dgx spark explaining so i better go off with mac studio than dgx spark?

[verbal wagon]

so should i wait a bit more for m5 dam....

i thought about using opencode + molt + glm 4.7 + qwen 2.5 all opensource all locally on a dgx spark

[green sluice]

IF there will be an M5 Ultra. Apple said not to expect an Ultra with every M-generation. And there wasn't an M4 Ultra. So, chances are good.

[verbal wagon]

to use qwen as the brain and glm 4.7 as the agent worker

[odd tulip]

I found this at my place; do you think I can run a small local LLM + clawdbot on it, or is it a lost cause for the local LLM? : Beelink Mini S12 Mini PC, Mini Desktop Computer with Alder Lake-N N95, 16GB SO-DIMM DDR4 + 500GB M.2 2280 SATA3 SSD, WiFi 5, Dual HDMI, Dual Display, 1000Mbps, BT 4.2

[green sluice]

You can run an llm even on a raspbery pi, but it won't be very smart, useful or fast 😄

it all depends on what you want to do with it, how capable it needs to be

[odd tulip]

Yeah with little model like qwen2.5-3b ^^ but slow because no GPU☠️

[coarse saffron]

Cool. Always excited for new stuff. Hope that is true. It would be nice if we could buy mutilple minis and get performance gains for local models without jumping to the studio macs. Like adding cores for better performance and an incremental cost.

[odd tulip]

Just little assistant

[green sluice]

because of the current hardware crisis I expect a lot of impressive developments in this field exo is in. As Dr. Ian Malcolm once said: Life Finds a way 😄

[coarse saffron]

2026 Mac mini: Specs
16GB RAM (M5); 24GB RAM (M5 Pro)
Thunderbolt 4 (M5); Thunderbolt 5 (M5 Pro)
Wi-Fi 6E (802.11ax) and Bluetooth 5.3

Gross. No thunderbolt 5 unless you get the better model. Of course that is just some news site.

Oh, that was the current gen. still gross.

[green sluice]

the current mac mini is still the m4 version. Thunderbolt 5 only with M4 Pro.

[graceful charm]

Sure what exactly are you getting stuck on?

[delicate pumice]

https://x.com/atariorbit/status/2016293150126461120

[astral goblet]

@atariorbit via Twitter

atariorbit (@atariorbit)

Reachy Mini + @moltbot (aka Clawdbot) = Full Robotic Agentic Buddy. And the best thing is I'm just asking moltbot to do it. And it's done it!
︀︀I named him Rasputin and we talk in my discord. This is insane. @huggingface #clawdbot #Robots

**👁️ 42 **

FxTwitter · Jan 27, 2026, 11:32 PM

[delicate pumice]

Ultimate hardware for your bot

[graceful charm]

Dm me if u want

[delicate pumice]

I connected clawd up to my Reachy Mini. Just plugged it in and pointed it to the GH repo.

working on a skill, oh I see there is #skills

[green sluice]

thats cool!

now it needs arms to make a sandwich 😄

[dreamy sun]

Will Molt be able to run on optimus I wonder?

give it a body

[green sluice]

maybe starting with a hand first, one bodypart at a time https://de.aliexpress.com/item/1005006091523781.html

I could hide a small servo in the claw of the Molty MacMoltbotface Mac mini case to make it wiggle, theoretically

[edgy swift]

Anyone run clawd/molt on a mobile phone?

[analog crag]

is it better to buy rtx 3090 or mac studio mini 32gb?

Am on 4060 right now.

or any other ideas

say budget is arround the price range of those two

[brittle barn]

Before I do go down the road of looking at whether I should use part of my current M2 Max 32GB desktop OR something Windows with more DDR5 RAM on it, is there a good benchmarking tool for Moltbot?

Idea on the desktop is that I wouldn't give all the RAM to it, so also knowing if I can set a max allocated RAM to it would be nice. Wouldn't need that on another option.

(found my answer on RAM limiting)

[buoyant perch]

Ay, anyone sacrifice a macbook air to run Molty on it?

Srsly considering to get app development going.

[brittle barn]

I've pondered my old M1 Air for it, but 8GB of RAM.

[buoyant perch]

https://tenor.com/b0itA.gif

[brittle barn]

Exactly. You could do it, buy yeah, we've just stated why not.

[buoyant perch]

Mine is M4, 16GB. Think that'd be valid?

[bitter cipher]

Just ask it for comparison it's pretty good. Also depends on what you want to do. I think it becomes more powerful when you run local text to speech and other tools so more memory is better.

E.g. transcribe all my voice messages and put in object and vector store

[buoyant perch]

Man, what don't i want it to do lol

But mainly would be using an actual browser to do things + rapid iOS app development

[quiet gulch]

Has anyone successfully used clawdbot's node mode to orchestrate a fleet? I just finished setting up a multi-node
setup with local open-source models and curious if others have done something similar or have tips on agent routing.

[fleet light]

I’m going in

[green sluice]

See here: #showcase-old message

and here: #hardware message

[green sluice]

https://tenor.com/view/satisfied-karate-kid-smile-nod-gif-17126717

[green sluice]

You are the benchmark yourself. Is the thing you want molty to do possible doing yourself on that computer? What do you want molty to do locally? Do you want to use it with a local llm? If so, neither Mac Studio 32GB nor one 4090 would be enough. If you want to use it with some remote llm via api, claude/chatgpt/etc, both will be sufficient.

For regular office-level tasks even a 150€/$ setup is sufficient. It you want Molty to do more intensive or resource hungry tasks, media creation/handling/converting or compiling, then the more power the better.

[scenic notch]

is there any model that can be run locally on a 3090ti, which is sufficient in tool calling? for the use with moltbot?

[cursive plank]

Raspberry pi 5 or Mac mini m4
Got both idk what to run it on I use my Mac mini not anymore so Yh

[obsidian barn]

Running molty on a 2018 Mac mini no problem. Loving it.

[spiral sierra]

Mac mini M1 with 8gb ram or N100 miniPC with 16gb ram, what would you think is better ? Can get both for the same price rn.

[earnest lichen]

Clawd itself uses almost 0 resources, so depends what else u wanna run on it

[green sluice]

I‘d say

• Qwen3-32B https://huggingface.co/Qwen/Qwen3-32B
• Qwen2.5-32B-Instruct https://huggingface.co/Qwen/Qwen2.5-32B-Instruct
• ibm-granite/granite-20b-functioncalling https://huggingface.co/Mungert/granite-20b-functioncalling-GGUF
• Qwen3-14B https://huggingface.co/Qwen/Qwen3-14B
• Mistral Small 3.1 24B https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503

https://www.docker.com/blog/local-llm-tool-calling-a-practical-evaluation/

[final turret]

Anyone running on a VPS vs a dedicated box? Just trying to figure out what I really wanna do to get started

[stoic seal]

Please help I want to use full potential as of today. What's better: 5090 or Mac Studio?

[civic merlin]

U mean for a self hosted model?

[icy depot]

Dependes

[zinc thistle]

agree with you, my moltbot cpu usage is 5% (aws c8.large), ram is 30% (4gb only), and disk space is less than 200Kb! ,yet it's running bot that's active every second 24/7, even during setup, none of these resources spiked too high,

i wouldn't use a pi for molty, though, i'd want the fastest data bandwidth from molty to the brain, (Claude, web, etc), the way i see it for my use-case and possibly for a lot of the folks buying dedicated hardware mac mini sales 🚀 need to understand that molty is a wrapper, like a proxy-server, the cloud is the processor, somewhere out there, depending on use-case design, i.e. rayban -> meta infra, gemini/claude/xai ai, home automation systems

[runic hill]

T.T im waiting for m5 pro/max/ultra chip

[fringe shadow]

I just installed on the last Intel MBA (2020) - not far enough along to know how well it will work, but so far no issues (not planning to run any models locally on this - just wanted an isolated machine to test on first)

[buoyant perch]

Got mine to work this morning. 💪

[green sluice]

I recommend checking the dedicated channel for everything models: https://discord.com/channels/1456350064065904867/1456704705219661980

• Qwen3 14B https://huggingface.co/Qwen/Qwen3-14B-AWQ
• Mistral Nemo 12B Instruct https://huggingface.co/mistralai/Mistral-Nemo-Instruct-FP8-2407
• Qwen2.5 14B Instruct https://huggingface.co/Qwen/Qwen2.5-14B-Instruct

[spiral sierra]

Would Mac mini M4 be good for running qwen 14b locally? and then I would use coding api for coding.

[gusty sandal]

14b would be a pretty big LLM for the macmini it wod just run it slow. a good one would be 7-8b for mac mini

if you like tho you can wait for the m5 chip and it ciuld run the 13b-14b better

[gusty sandal]

same, june cannot come any faster lmaoo

[brittle barn]

Think the best answer would be “as much as possible” locally. Investigating potential tasks would likely help me here.

I also have a mini PC with 64GB DDR5 but I don’t know how much I can allocate directly to the GPU (780M) and honestly considering the “gold” price of this RAM, it almost might be better to sell?!

[spiral sierra]

That's good idea, gemini said 14b is no problem for M4. I'll just wait too

[raven magnet]

You’re better off running qwen30b, it’s MOE with only 3B active parameters and very fast

[raven magnet]

That logic is never ending when the M6 is planned for the end of the year and M7 next year lol

[gusty sandal]

But the m4 is from 2024 and m5 is promised around 30% faster proccessing with more core. And not to mention clawd is fairly new with a lot of security issues it will be better the wait to see any patches or get more documentation on it

[spiral sierra]

I wanted to run something lighter local with coding paid api as second model, but it probably won't matter when I buy z.ai pro plan

[fleet light]

So what kind of Linux should I install on this?

[raven magnet]

Lighter in vram usage or speed? The 30B moe model is much faster than dense 14 or 7B models if you have 32gb mac

If you have 16gb then 14B is probably the limit that fits it

[plush harness]

why is it a bad idea to run the bot on a raspberry pi?