AI agent hosting for agencies: how to run AI agents and automations for your clients

By the serverserv team · 8 min read · Updated 28 October 2026

Agencies that build AI agents and automations for clients usually end up choosing between three setups: one shared server for every client, one server per client, or one isolated agent per client. Each has a different cost, risk and licensing picture, and with n8n the licence decides more than most people expect. This guide walks through all three, what you can charge for, and how to size the servers.

Why agencies host agents on their own servers

Hosted automation tools charge by tasks, executions or seats, and those costs grow with every client you add. A self-hosted server costs the same whether it runs a hundred executions or a hundred thousand, so margins improve as you grow.

• Predictable cost per client. A fixed monthly server bill instead of usage pricing that spikes when a client's workflow loops.
• Client data stays on a server you control, which helps with clients who ask where their data lives.
• No feature tiers. Community nodes, custom code and your own AI model keys are available on every server.

The trade-off is that someone has to look after the servers: updates, backups and security. Plan that time into your retainer, or hand it to a host.

Three ways to host client agents

Setup	How it works	Good for	Watch out for
1. One agency server	All client workflows and agents run on one server you own	Many small clients, lowest cost	One problem affects every client; clients must not get builder access
2. One server per client, owned by the client	Each client orders their own server; you build and maintain what runs on it	Larger clients, clean separation, easy handover	More servers to update; n8n licence rules on hosting (below)
3. One isolated agent per client	Each client gets its own OpenClaw agent on its own server or container	Agents that read client email, files or chats	Model keys and spending limits per client

Many agencies mix them: small clients share setup 1, and bigger clients get their own server under setup 2 or 3.

What the n8n licence allows agencies to do

n8n is fair-code under the Sustainable Use License, not open source, and its FAQ is specific about agency work. In short: you can run workflows for clients, but you cannot sell clients access to n8n itself.

• Allowed: building, running and maintaining workflows on your own n8n instance on behalf of clients, with no limit on the number of clients, as long as clients cannot create or edit the workflows.
• Allowed: charging for workflow creation, setup, maintenance, training and consultancy. n8n's own phrasing is that clients pay for outcomes, not for n8n access.
• Allowed: installing and maintaining n8n on a server the client owns, as long as you do not also sell them the hosting.
• Not allowed: letting clients log into your n8n to build or change their own workflows, or white-labelling n8n by removing its branding.

If your product needs clients to build workflows themselves, n8n asks you to contact its sales team about an Enterprise licence. When in doubt, check the licence FAQ or ask n8n directly; this is a summary, not legal advice.

OpenClaw for client agents

OpenClaw is open source under the MIT licence, so you can host it for clients, resell it and charge for access without licence limits.

The bigger question with OpenClaw is isolation. An agent that reads a client's inbox or chats should not share a machine with another client's agent, so give each client its own server or at least its own container, with its own model key and spending limit. Keep each Gateway on localhost and update regularly; our OpenClaw security checklist covers the steps.

How big a server each setup needs

Setup	Typical size	serverserv plan	Price
Shared n8n server, up to about 10 small clients	4 vCPU, 16 GB	KVM 3	$39.99 monthly, $27.99 on 3 years
Shared server, busy webhooks or AI steps	8 vCPU, 32 GB	KVM 4	$69.99 monthly, $48.99 on 3 years
Per-client n8n server	2 vCPU, 4 GB	n8n Starter	$14.99 monthly, $10.99 on 3 years
Per-client OpenClaw agent	2 vCPU, 4 GB	OpenClaw Starter	$14.99 monthly, $10.99 on 3 years
Per-client agent that browses the web	4 vCPU, 8 GB	OpenClaw Personal Use	$24.99 monthly

Treat these as starting points. Watch memory for the first few weeks and move up a plan if swap stays busy. See the best VPS for AI agents for sizing by agent type.

What agencies charge for

The agencies that do well price the outcome and the upkeep, not the server.

• Setup fee for building the agent or workflows, connecting the client's tools and testing.
• Monthly retainer for monitoring, updates, fixes and small changes.
• Model usage, either passed through at cost from the client's own API account, or included with a cap.
• Hosting, either included in the retainer on your own server, or paid by the client directly when they own the server.

Having the client pay for their own model key keeps their AI bill visible to them and keeps a runaway agent from landing on your invoice.

Hosting client agents on serverserv

• n8n hosting and OpenClaw hosting come pre-installed on the client's own server, with full root access and no execution caps.
• We set up the first workflow or agent task free on each new server, which gives you a working baseline for every client.
• Monthly billing per server, so you can add and remove clients without long commitments, or 1 to 3-year terms for long-running clients.
• Running many servers, or need a custom size? Talk to sales for bulk pricing. If your clients order their own servers, ask us about referral terms.

Questions

Sources

• n8n: licence FAQ (Sustainable Use License)
• n8n: Sustainable Use License
• OpenClaw on GitHub (MIT licence)
• Clawdocs: OpenClaw security and isolation

Prices from other companies are as published at the time of writing and can change.