Modified
Build your technical foundation
An agent is a program. Programs need to run somewhere — on a computer, a server, a . This page shows you what sits under the hood, so you understand what you're actually choosing when you choose.
You don't need to understand it all at once. Start at the bottom and build upward.
A room shares walls and a foundation with the rest of the house — it's easier to move a room than to build a new house. Containers work the same way: they start in seconds, share the kernel with the system underneath, and use minimal resources.
Running AI via a browser or a work tool? You need no infrastructure at all — you're done. Running locally, via CLI, or setting up your own agents? The rest of this page is for you. Or just curious.
Think of it as a house
Your agent lives at the top. Everything underneath is the infrastructure holding it up — just like a house needs a foundation, frame, and roof before you furnish it.
Power & water
The house can be finished, but without power and water it's uninhabitable. Your agent needs the same: a way to authenticate against external services, and somewhere to store and fetch data.
Electricity only works if the place is connected and the bill is paid. Your agent needs access to every external service it talks to — a kind of digital contract that proves you are you. Stored in the wrong place or in the wrong hands, you pay the bill. Think of it as the key to the main breaker: it should be locked away, not hanging on the wall.
Data has to flow in and out. Think of the database as the cabinets in your kitchen — built-in cabinets for the things you reach for daily, and a storage room outside for what you rarely need but don't want to throw away. An agent with nowhere to put things remembers nothing between runs. Decide early where the cabinets go — moving them later is a hassle.
Your access keys shouldn't sit out in the open in the code where everyone can read them. If a key leaks, change the locks immediately — it takes five minutes but saves you from an unexpected bill.
Running locally
Running AI through a browser or a work tool? You need no infrastructure at all — you're done. Running locally, via CLI, or setting up your own agents? The rest of this page is for you.
No infrastructure needed · you're done
Personal
ChatGPT, Claude, Gemini
You log into the browser and chat. Nothing technical on your end.
Through work
Microsoft Copilot, Workspace
IT has set up the service. You click and use it.
Running locally or with your own agents · the rest of this page is for you
Running CLI tools, local models, or building your own agent pipelines? Below we cover what you need — OS, containers, where to host, and how to scale when it grows.
Operating system
Linux
The default environment for everything from to n8n and agents. All agent frameworks (LangChain, CrewAI, LlamaIndex) and container tools are designed for first. Bash scripting, cron jobs, and systemd services become your everyday.
WSL2
On Windows and want to run Linux tools without dual-booting? is a real Linux kernel inside Windows. The same tools as on Linux (Docker Desktop, Ollama, Python venv) directly from Windows. passthrough works with NVIDIA drivers and CUDA.
macOS
Unix-based out of the box. The terminal, Python, and Docker work directly. Apple Silicon (M1/M2/M3) supports Metal acceleration for local models via Ollama. Homebrew is the package manager that makes installation easy.
Container
Docker
The apartment in the stack: a box that contains everything your app needs — code, libraries, settings. The box works the same on your laptop, on a server in the cloud, or on a colleague's machine. No more "works on my machine".
Podman
A drop-in replacement for . Same commands, but without a central daemon, which makes it lighter and more secure. alias docker=podman usually works directly. Preferred in the Red Hat world.
Hosting
The server at the bottom of the stack. Where it runs decides jurisdiction, price, and how much tech you handle yourself. Three camps: Swedish/EU VPS for data protection, dev-friendly for fast deploys, enterprise cloud for scale.
Inleed
Swedish hosting with servers in Sweden. Data falls under Swedish law, no third-country transfers. The right pick for Swedish companies and projects with personal data.
Hetzner
German hosting with extremely low prices. Servers in Germany and Finland, EU jurisdiction. Popular among engineers who want to optimize cost without leaving the EU.
Railway
Push from GitHub, deploy directly. Little configuration, lots of magic. Good for prototypes and MVPs. Data is stored in the US and falls under the .
Fly.io
Similar to but deploys close to your users across 30+ regions. Slightly more Docker knowledge required. Strong on low-latency and edge applications.
AWS
The world's largest cloud. EU regions exist, but data still falls under the CLOUD Act. EC2 is the most common starting point; Fargate runs containers without managing servers.
Google Cloud
Google's cloud. Strong integration with Gemini and TPUs for AI workloads. EU regions are available. Cloud Run is straightforward container hosting without cluster configuration.
Scaling & automation
Kubernetes
The system that manages many Docker containers in production. Restarts containers that crash, balances traffic, and scales up automatically under load. Wait until your traffic justifies the complexity.
CI/CD
. Tests and deploys your code automatically every time you push to GitHub. No more manual deploys. GitHub Actions is free and the easiest place to start.
A small agent costs less than a coffee a day to host. Hold off on scaling until your traffic justifies it.
Prices change continuously. Always check the current price with the relevant provider.