Orgo vs rolling your own VMs for vertical agent products in 2027

Quick answer: the server bill won't settle this. What decides it is what you build on top. On a VM you still add the desktop, the screen, the input API, tenancy and per-customer billing. On Orgo those ship with the computer. Roll your own only when that computer layer is the thing you want to own. All prices here are as of November 2026.

TL;DR

• You own the agent. Orgo owns the computer. Bring Claude computer use, OpenAI's CUA, Hermes Agent, OpenClaw or your own loop (llms.txt).
• What the computer costs. An Orgo Medium (2 vCPU, 8 GB, 40 GB) is $32 a month as plan capacity, less inside a plan, or $0.055 an hour, with the desktop and the control API included.
• Plan vs hourly breaks even near 580 hours a month (roughly 19 hours a day). Below that, hourly is cheaper. Above it, the plan is.
• The real cost of building it yourself is engineering and operations, not the server bill.
• Roll your own when you need the host, the kernel or public networking, or when the computer layer is your product.

What does "rolling your own VMs" mean for an agent product?

Rolling your own VMs means renting general-purpose virtual machines from a cloud or VPS provider and building the agent computer on top yourself. The provider gives you a server with an operating system and a disk. Everything that turns it into a desktop an agent can see and control, and everything that turns many servers into one computer per customer, is code you write and run.

An agent computer is a cloud desktop built for an AI agent to drive: a full desktop with a browser, reachable over HTTP and VNC, with screen and input actions exposed as API calls. That is what Orgo runs. For why a vertical agent needs one at all, see Why vertical agent products need a real computer.

What do you have to build on raw VMs?

1. A base image. A desktop environment, a browser and your agent's tools, kept patched.
2. A display and a remote screen. A display server and a VNC server, exposed safely, plus a viewer in your app.
3. Input and screenshots. A service that screenshots, clicks, types and scrolls on request, a shell endpoint, and an authenticated API in front of it.
4. Per-customer isolation. Keys and storage that keep one customer's sessions, files and credentials away from another's.
5. Snapshots, copies and backups. Save state, copy a configured machine, restore after a bad run.
6. Provisioning and scaling. Create machines from code, track quotas, clean up idle ones.
7. Billing per tenant. Meter hours and storage per customer and charge for them.

None of this is your agent. It is the computer layer under it, and it is the part of the bill that never shows up on the VPS invoice.

How do Orgo and self-managed VMs compare?

Sources: Compare platforms, Connect with SSH, Suspend, Fork, Embed computers.

Direct SSH is a plain SSH server on a Linux computer that you turn on with your public key, so tools that expect ssh still work. See How to SSH into an AI agent's cloud computer.

How does Orgo handle one computer per customer?

Each customer gets its own workspace and a key that reaches only that workspace, and POST /v1/accounts creates both from your app. The full setup, caps and offboarding are in How to give every customer its own AI agent computer.

What does an Orgo computer cost?

Plans (Plans and billing, pricing.md):

Yearly billing saves 20%. Disk per computer is before add-ons, and storage add-ons raise it up to 300 GB.

Past the plan, you have two ways to pay (Computer prices):

Capacity is sold in parts, each a month: RAM $2.50 per GB, vCPU $5 per vCPU and disk $1 per 20 GB. Hourly time is billed by the second from credit. The running price is the monthly price plus 25%, divided by 730 hours. Stopped, a computer pays only for its disk.

Windows is included on Scale. On Starter and Growth each Windows computer is $47 a month in all, a $15 license plus a Medium's memory, vCPU and disk. Without one, the create answers 403 WINDOWS_REQUIRES_SCALE. Windows is not offered by the hour, so it always runs on a plan.

End customers never need a seat. Seats are for your teammates. The people your agents serve don't count on any plan (pricing.md).

Where is the break-even between plan and hourly?

The hourly running price is the plan capacity price times 1.25, rounded to the nearest $0.001. So, leaving stopped disk aside, hourly is cheaper when a computer runs less than about 730 ÷ 1.25 = 584 hours a month, about 19 hours a day, at every size.

The stopped disk is why the real break-even for a Medium is about 574 hours, not 584. A deleted computer costs nothing. Give a short job ends_in_hours and Orgo deletes it when the time is up (Set when a computer ends).

Check your own fleet with a few lines of Python:

HOURS = 730
SIZES = {  # monthly plan price, hourly running, hourly stopped
    "small": (13.50, 0.023, 0.0014),
    "medium": (32, 0.055, 0.0027),
    "large": (64, 0.11, 0.0055),
    "xl": (118, 0.202, 0.011),
}
 
def monthly(size, running_hours, count=1):
    plan, run, stop = SIZES[size]
    hourly = running_hours * run + (HOURS - running_hours) * stop
    return {"plan": round(plan * count, 2), "hourly": round(hourly * count, 2)}
 
print(monthly("medium", 176, count=20))  # {'plan': 640, 'hourly': 223.52}

A plan's bundled computers cost less than added capacity: Starter's one Medium is $29 against $32, Growth works out to $24.75 a Medium and Scale to $24.94.

What do you pay for when you run the VMs yourself?

A VM's price buys a server with an OS and nothing else. It leaves out the seven items in the build list and the people who keep them running. There is no published number for that effort, so estimate it for your own team: engineer-weeks to build, plus the on-call time to run it.

On Orgo those items come with the computer, so that time goes to your agent instead. If your agents work part of the day, you pay the running rate only while a computer runs: the 20 workday Mediums above cost $223.52 a month by the hour, about $11.18 each.

What else changes when you stop running your own fleet?

• Scaling ceilings are published. An account on a paid plan can have 50 computers. Past 50, more run by the hour from credit, up to 999 an account, and Orgo can raise that to 9999 on request (Scale). On your own you can add up to 800 GB of RAM, 200 vCPU and 4 TB of disk. Past that the API answers 403 capacity_cap until Orgo raises it.
• Rate limits are published per plan. New computers a minute: Starter 10, Growth 30, Scale 60. Each computer takes up to 600 control calls a minute on every plan (Account limits).
• You can bill the customer through Orgo. With client billing, Orgo keeps the computer cost, Stripe takes 2.9% + 30¢, and the rest is yours. A client paying $250 for $50 of computers leaves you $192.45 (Invoicing).
• The same setup for every customer comes from a template. Publishing your own needs Scale or Enterprise. See Templates for vertical AI agents.

What should you plan around?

• Inbound traffic goes through an authenticated proxy, or Direct SSH on a port of the host.
• Orgo runs the host and the kernel for you.
• Windows runs on a plan only, not by the hour.
• Macs and iPhones come from Orgo's stock while it's in stock, billed on their own (pricing.md). Self-serve sizes stop at XL (6 vCPU, 32 GB). 2XL (8 vCPU, 64 GB, 300 GB) is set up on request, and nothing goes above it (Instance types).

When should you still roll your own VMs?

• The computer layer is your product. You want to own the hypervisor, image pipeline and fleet.
• You need host-level control. Custom kernels, your own networking, public ports.
• The workload isn't an agent driving a desktop. Backends, databases and APIs belong on a general server.

The two also mix: a VPS for backend services, Orgo for the work where an agent drives a computer.

Which errors tell you you've hit a cost limit?

Past your plan, a create that names no billing_mode, from a key with full access or Billing write, gets no error: it runs by the hour from your credit on its own. These are the answers you can get:

Details: Create computer.

FAQ

Is an Orgo computer just a VM with a desktop on it?

It is a persistent cloud computer built for agents: a full desktop with a browser, reachable over HTTP and VNC, with screen and input actions as API calls, plus workspaces, scoped keys, templates, snapshots and billing around it.

What does Orgo's price cover that a VPS price doesn't?

A VPS price covers a server with an OS. An Orgo price also covers the layer you would otherwise build and run: the desktop, the live screen, the input API, and a workspace and key per customer. So compare the plan price with the VPS price plus what that layer costs your team. For computers that run part of the day, billing by the second means you pay the running rate only while they run.

Which operating systems can I run?

Linux on any plan, on the plan or by the hour. Windows is included on Scale, is $47 a month per computer on Starter and Growth (a $15 license plus a Medium's room), and runs on a plan only. Macs and iPhones come from Orgo's stock while it's in stock.

How big can one computer be?

Up to XL on your own: 6 vCPU and 32 GB RAM, when your plan's pool has room or you run it by the hour. Disk per computer before add-ons is 40 GB on Starter, 80 GB on Growth and 150 GB on Scale, and storage add-ons raise it to 300 GB. 2XL is set up on request.

Can I SSH in like a normal VM?

Yes, on Linux. Turn on Direct SSH with your public key and connect with ssh -p <port> root@<host> (Connect with SSH).

Do I have to use a specific agent or model?

No. Orgo provides the computer, not the agent.

Related

• How to give every customer its own AI agent computer
• Why vertical agent products need a real computer
• Templates for vertical AI agents
• Orgo vs Amazon WorkSpaces for AI agents
• How to SSH into an AI agent's cloud computer
• How to scale AI agent computers to thousands on Orgo

Sources

• Orgo pricing.md
• Plans and billing and Computer prices
• Account limits and Scale
• Create computer
• Connect with SSH
• Invoicing