AI agent? Read uno.place/llms.txt — connect to the person's account without a password.

How to choose a server for your project

Choose by memory first: it is the limit you hit hardest, and a process that runs out of it is killed. A bot or a VPN fits in 1 GB, a web app with a database starts at 2 GB, an AI coding agent wants 4 GB or more, and builds want 8 GB. Then check the disk, the cores and the region closest to your users.

Time
About 10 minutes
Server
Any Linux server: this guide picks its size
You need
A list of what will run on it
Four choices in order: memory, disk, cores, region. Memory decides whether the project runs at all; the region decides how fast users feel it 1. Memory decides if it runs at all. Too little: processes are killed 2. Disk code, packages, database, logs. Big files go to object storage 3. Cores matter for builds and many requests at once, not for an idle bot 4. Region as close to your users as you can get Start with the first, size it by the table below, and let real use correct the others.
Four choices, in this order.

Starting size by project

The table is a place to start, not a promise. Where a vendor states a number, it is linked in the notes under the table; the rest is our rule of thumb.

ProjectMemoryvCPUDiskFits on Uno
Static site, landing pageno server needed--Uno sites, free
Telegram bot, VPN, small API1 GB110 GBSmall: 1 GB, 15 GB disk
Web app with a small database2 to 4 GB220 GBPlus: 4 GB. Small for a light app
OpenClaw4 GB220 GBPlus: 4 GB, 40 GB disk
Claude Code or Codex4 GB at least, 8 GB for Codex2 to 440 GBPlus, boosted to 8 GB for heavy hours
Builds, Docker images, test runs8 GB or more4 or more80 GBPro: 16 GB, 120 GB disk
Several agents at once, a team16 GB or more8 or more120 GB or morePro or Max: 48 GB
Vendor numbers: Claude Code asks for 4 GB or more (code.claude.com/docs/en/setup); Codex asks for 4 GB at least and recommends 8 GB (github.com/openai/codex, docs/install.md); the OpenClaw docs call 4 GB plenty for the Gateway (docs.openclaw.ai). Checked 16 October 2026.

Steps

  1. List what will run

    Write down every program: the app, the database, the agent, a queue, a browser. Each one has its own memory. A database and a browser are the usual big ones.

  2. Measure a similar setup

    If you can run the project on your laptop or any Linux machine, look at what it takes. Run it as you will use it, not idle.

    $ free -h
    $ docker stats --no-stream      # if you use Docker
  3. Pick the memory, then add room

    Take the largest figure you saw, not the average, and round up to the next size. Builds, installs and test runs use far more than idle code does.

  4. Check the disk

    Count the project folder, the database and the packages. Keep uploads, media and backups out of the disk: they belong in object storage, and every Uno plan includes some (25 GB on Small).

    $ du -sh ~/your-project
    $ df -h /
  5. Count the cores

    One core is enough for a bot or a small site. Builds and many simultaneous users want more. If the load stays above the number of cores, you need more.

    $ nproc
    $ uptime        # load averages vs the core count
  6. Pick the region closest to your users

    Distance adds latency to every request. Put the server near the people who use it, or near the API it calls most. Check the round trip from where your users are.

    $ ping -c 4 203.0.113.10
  7. Choose a size you can change

    Do not buy for next year. Take the size from the table, run for a week and read the numbers again. Before you pay, ask how the host resizes: some need a restart, some a new server.

Check that you chose well

After a few days of real use, run the same commands on the server. If available memory in free -h stays close to zero, or swap keeps growing, go up one size. If memory stays mostly free and the load is low, you can go down.

$ free -h
$ df -h /
$ uptime

What memory costs

Memory is the price driver, so compare it per gigabyte. Budget hosts sell it cheapest. Uno costs more per gigabyte and puts the rest in the plan.

One gigabyte of memory for a month

In $ per GB per month. Ranges show the cheapest and the dearest provider we checked.

Uno
$4.2–5
Budget hosts
$0.65–1.6
Cloud VPS
$5–6
0246
Show as a table
Option$ per GB per monthShape
Uno4.2–5Small to Max · sleeps when idle, cloud storage and sites included
Budget hosts0.65–1.6Hetzner CX, OVH, Contabo · runs and bills all month
Cloud VPS5–6DigitalOcean, Vultr, Linode, Lightsail

The short way: a size you can change

On Uno you do not have to guess right. Plans change in both directions: move up or down any time, files and computers come with you. Every plan has monthly boosts for heavy hours: every hour a computer sleeps earns one boost, up to the monthly cap of your plan. Start with Small, watch, and move up.

See how it works

Small: $5 a month

Where Uno is not the cheapest

  • Memory costs more per gigabyte than at budget hosts: $4.2–5 against $0.65–1.6. A project that needs only RAM and nothing else is cheaper there.
  • Uno computers run in the Netherlands today. Users far from it feel some extra milliseconds.
  • Uno gives you shared vCPU on AMD Zen 4. A job that holds every core at full load all day fits a server with reserved cores better.

Uno Small against a classic VPS

The same 1 vCPU class. What you pay, and what you get with it.

Uno SmallBudget VPS
Hetzner, OVH, Contabo
Cloud VPS
DigitalOcean, Vultr, Linode
Price a month$5$4.5–6.5$5–7
Root and Dockeryesyesyes
Who sets it upan AI agent, or youyou, over SSHyou, over SSH
When idleasleep, memory kept, earns boostsruns and bills all monthruns and bills all month
Cloud storage25 GB includedseparate productseparate product
Sites and formsfree, one messageset up a web serverset up a web server
AI in the plan5 h a month, Uno AI optionalnono

Prices checked 2026-10-01. Budget hosts sell memory cheaper per GB. If all you need is a lot of RAM and nothing else, they are the better deal.

Pick a starting size

Every plan sleeps when idle and wakes on the first request. Move up or down any time; your files come with you.

Small

A Telegram bot, a VPN, a small backend

$5/ month

  • 1 vCPU · 1 GB
  • 15 GB disk
  • 25 GB cloud storage
  • 5 boosts a month
Get Small

Plus

Agents that work with your laptop closed

$20/ month

  • 4 vCPU · 4 GB
  • 40 GB disk
  • 100 GB cloud storage
  • Uno Work in the cloud
Get Plus

Pro

A product in production, a second MacBook

$70/ month

  • 16 vCPU · 16 GB
  • 120 GB disk
  • 300 GB cloud storage
  • Speed of a MacBook Air M4
Get Pro

Max

A team and many agents at once

$200/ month

  • 32 vCPU · 48 GB
  • 300 GB disk
  • 1,024 GB cloud storage
  • 5 public IPv4
Get Max

Add Uno AI to any plan: 40 h of AI work a month on Small, 100 h on Plus, unlimited on Max. Today: GLM-5.3 Flash and MiMo v2.6 Pro. Or sign in with your own Claude or ChatGPT.

Common problems

My process is "Killed" with no error
Linux ended it because the server ran out of memory. Confirm with sudo dmesg | grep -i "killed process". Go up one size; a swap file only postpones the problem.
Should I add cores or memory first?
Memory. A shortage of cores makes things slow; a shortage of memory makes them stop. Add cores when memory is fine and the load stays above the core count.
Does the region matter for a Telegram bot?
Little. Telegram reaches the bot from anywhere and the user waits on the network, not on the server. For a site or an API with live users, the region matters more.
The disk is full but my project is small
Logs, Docker images and package caches grow in the background. Find the big folders with sudo du -xh / --max-depth=2 | sort -h | tail and move uploads to object storage.

Next

  1. What is a VPSA slice of a server that behaves like your own computer.
  2. How to run Claude Code on a serverSix steps and the commands to copy. About 15 minutes.
  3. Uno vs HetznerHetzner is the cheaper server. Uno is the server that sets itself up and sleeps.

Start small, move up when you need to

Plans change in both directions: move up or down any time, files and computers come with you. Small is $5 a month.

Get your computer