Open-source AI agent hosting, explained
A new kind of software wants a new kind of home. Open-source AI agents don’t run in a browser tab — they live on a computer, around the clock. This page explains what that means, what your options are, and what a good setup looks like, whichever path you take.
What an open-source AI agent actually is
An AI agent is a model given hands and a memory: software that can carry out tasks — working with files, browsing, scheduling, messaging you first when something needs attention — rather than just answering when asked. The open-source part means the software that does this (the “harness”) is yours to run: projects like OpenClaw give you the agent’s body, you connect the AI model of your choice with your own key, and everything the agent is — its memory, its personality, its files — lives on a machine you control, not in a vendor’s cloud account.
That ownership is the point. It’s also the catch: someone has to provide and look after that machine.
Why agents need hosting at all
Two properties separate an agent from a chatbot tab:
- It’s always on. An agent that runs overnight jobs, watches for things, and answers from your phone at a café needs a computer that never sleeps. Laptops sleep. This is the practical reason “where does it run?” comes up on day two of every agent experiment.
- It can act. An agent runs commands and touches files on the machine it lives on. That makes the machine itself the security boundary — and makes your personal laptop a poor choice of boundary, since everything on it is inside the blast radius. Agents deserve a machine of their own.
“AI agent hosting” is the umbrella answer: giving the agent a dedicated, always-on, secured computer — its own room, with a lock.
The realistic options
For a personal agent there are three homes people actually use — we compare them in detail here:
- A box at home (the beloved Mac Mini): real hardware, real up-front cost, and your house becomes a tiny data center — with you as its staff.
- A rented server (VPS): a few dollars a month for a machine in a real data center. The agent projects publish setup guides for the big providers; the security, updates, and backups are a permanent checklist you own.
- Managed hosting: the machine, security, backups, and updates are the product; you bring the model key and the conversations. Costs more than a bare VPS; costs no weekends.
What good agent hosting includes
However it’s done — by you or by a host — a trustworthy setup for an autonomous agent has the same shape:
A private front door
The agent’s control surface never faces the open internet; access is authenticated, always.
A boundary of its own
One agent, one machine, one owner — no sharing a trust boundary with strangers or with your personal files.
Memory that survives
State that persists through restarts and reboots, backed up automatically, restorable in practice — not in theory.
Deliberate change
Pinned versions, managed upgrades, and a security audit after every change — an agent’s home shouldn’t surprise it.
Where Everpod fits
Everpod is managed hosting built exactly to that shape. Your agent gets a pod — a private, always-on cloud computer of its own, one customer per pod — with its gateway kept off the open internet, automatic backups, pinned versions, and managed upgrades. We host OpenClaw today, with more agents planned (Claude Code, Codex, and Hermes are the ones people ask for most — tell us yours when you request access). It’s $29/mo, bring your own model key: the agent stays yours; the server stops being your problem.