New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Anyone here running AI agents 24/7 on a VPS?
Running AI agents locally is great until you close your laptop.
For agents that need to keep working in the background, like automations, coding agents, bots, MCP servers, scheduled tasks or workflows connected to APIs, they need somewhere to live.
A VPS feels like a pretty natural option in many cases. You get full control over the environment, it stays online, and your laptop doesn't have to work the night shift 😅
We've been exploring this use case at CubePath as well, and we're curious about what people here are actually using.
Where are you running your AI agents right now?
VPS? Home server? Cloud platforms? Or is your laptop still doing the night shift?
Comments
@Neoon
with Claude Code it is more practical to use VPS for Coding and staging platforms. Simple VPS with 4 cores does great.
The /remote-control allows to have a nice interface in their web or app.
For now my poor lappy running 24*7 but I will migrate to VPS if i get good offer this Black Friday.
It may be overkill, but I choose to run it on a dedicated server with a GPU in it so I can balance a bunch of things vs relying on my PC.
You can vibe code, apps overnight on a 10$/m Kimsufi.
It codes stuff, that then runs 24/7 but the agent itself doesn't.
What is agent suppose to do, which a script cannot?
Just use the agent, to write a script, that does the work faster and more efficient.
And if the script breaks, you got your agent to fix it, uwu.
Sure, using DSH on multiple servers that can run autonomously most of the time until the project is complete (or at least your prompted requirements). For this, all DSH instances run with those plugins: https://github.com/AGSQ11?tab=repositories&q=dsh&type=&language=&sort=
It auto-probes dead models and switch to next capable one, it regularly probes dead models and start using them when became alive, spawns councils to debate on sensitive topics or decisions to be made... and some other plugins I have not made public (at least for now).
Hosting companies engagement farming... let me guess... you guys are gonna launch AI Optimized ML Ready OpenClaw Hermes Agent Claude Code Optimized OpenCode Codex T3 Unlimited Ultimate Agentic VPS in a week
Already selling on their website, I got a top banner for it.
Ah great. And their original post is LLM-written. Low effort slop, overpriced servers, smh
I find it wild that people run Claude Code and Codex on devices that contain important/personal data because it's advanced malware and you just have to trust that those sketchy companies won't abuse their access, so personally I only ever run agents on sanitized VMs.
I run Claude and Codex on my ovh vps through T3Code.
Interesting use case. @CubePath
Do You have any guide , tutorials where we can use vps for this purpose?
Not running an agent though, but I'm running a local model using ollama+an Openwebui on the Computebox 4000 ADA GPU instance.
Claude Code is actually one of the use cases we had in mind. A simple VPS with enough CPU can be a really convenient setup for keeping the development environment available from anywhere.
Curious, are you running Claude Code directly on the VPS and connecting remotely, or using another layer on top of it?
That poor laptop deserves a break 😅
You might not have to wait until Black Friday though. We already have some pretty interesting VPS options if you want to give it an early retirement. And who knows what Black Friday might bring.
Not overkill if it does the job. A dedicated server gives you plenty of room to run multiple workloads without depending on your PC, and the integrated GPU is a nice bonus.
Fair point. For predictable and repeatable tasks, a script is often the better tool. No reason to keep an agent running just for the sake of it.
It gets more interesting when the workflow needs to react to changing inputs, make decisions or interact with different tools dynamically. But for everything else, let the script do the boring work!
This is exactly the kind of setup we were curious to see. Multiple autonomous instances that can switch models, recover when they become available again and coordinate on decisions is a very different use case from just running a scheduled script.
Pretty interesting example of what “24/7 agents” can actually look like in practice.
Isolation is definitely an important part of the conversation. Giving an agent access to a machine that also holds personal data, credentials or other sensitive resources is very different from running it in a clean, isolated environment.
That separation is actually another good reason to run these workloads on VPS.
Nice setup. That’s pretty much the kind of use case we had in mind when we posted this. Keeping Claude and Codex on a VPS makes a lot of sense when you want them available independently of your local machine.
What am I missing? What would you need an agent to run 24/7 for?
Even using Fable 5 Max, you can give it a prompt a mile long that is incredibly detailed and it'll shit out a functional prototype in a matter of hours.
Yes! We have a few resources already. Our Marketplace has several 1 click apps for this kind of setup, with instructions for each one, including OpenClaw, n8n, Ollama and more:
https://cubepath.com/marketplace
We also have short step by step tutorials on our YouTube channel showing how to deploy some of them on a VPS, including OpenClaw.
And we’ll keep adding more guides around agents and automation as we test more use cases.
Different use case, but still a nice setup. Ollama + Open WebUI on a GPU instance is a solid way to keep the whole stack self hosted and independent from your local machine.
It’s less about giving an agent one task that takes 24 hours, and more about having it available whenever something happens.
Think monitoring and reacting to events, processing incoming jobs, running scheduled workflows, interacting with APIs or handling tasks that arrive throughout the day. Most of the time it might be doing nothing, but it needs to be available when something triggers it.
For building a prototype from a single prompt, I agree, there’s probably no reason to keep an agent running 24/7.
Run it directly on VPS and /remote-connect to your interface. Always online, always ready for your next input. Clean VPS and set up a set of strict rules about what he can do too, just for paranoia, but works great.
Maybe anyone here tried Muse?
I’ve been playing around with Muse, a personal AI agent that looks pretty interesting.
If you’re curious, give it a try too:
https://muse.ai/join
Referral code: FTNKG1
Redeem the code in Settings within 48 hours of joining and we’ll both get 1 billion Muse tokens.
If you’ve already tried it, would love to hear what you think.
Sorry if this not allowed.
Thanks.
I do something similar.
I keep the controller running on a VPS, but the workers only do something when there’s actually a task. No point burning tokens 24/7.
I also prefer this over running everything on my main machine. Much easier to isolate permissions and credentials if something goes wrong.
For simple stuff I still use scripts though. Agents only really make sense when the task has some uncertainty or needs to decide what to do next.
That’s a good approach. Keeping the controller always on while workers only run when needed makes a lot of sense, especially for keeping costs and token usage under control. And isolating everything from your main machine is a big plus too.
I deploy Hermes army on one server
Butt cannit explane puntchewashun?