How Much Does It Cost to Run an AI Assistant? (Real Numbers)

A common question from business owners: “How much does it actually cost to have an AI assistant?”

A lot of companies in this space answer with “contact us for a quote.” I’d rather give you the moving parts and rough ranges, so you can do the math yourself and decide whether it’s worth it at all.

One note before we start: this post is about a self-hosted assistant, like OpenClaw or a custom agent running on a server you control. If what you really need is a better way to use ChatGPT, Microsoft Copilot, or Gemini, or the AI features already built into software you pay for, your costs look very different (usually a per-user subscription), and that may be the better answer. More on that at the end.

The Three Cost Buckets

Running a self-hosted AI assistant has three cost components:

Setup cost: getting it deployed and configured (one-time)
Infrastructure cost: hosting and software (monthly)
AI model cost: the actual AI usage (monthly, variable)

Setup Costs

DIY Setup: $0 (plus your time)

If you’re technically inclined, you can set up OpenClaw or a simple custom agent yourself. The software is open source.

What you’ll need:

Docker knowledge
Basic server administration
API key management
Comfort with command-line tools
Patience for troubleshooting
A clear understanding of the security settings (this is the part people skip)

Realistically, expect a weekend for initial setup, plus ongoing time for updates and troubleshooting. If your time is worth $100+ an hour, “free” may not be the cheapest option. But if you enjoy this kind of thing, it’s a perfectly reasonable path.

Professional Setup: $1,500–8,000

For reference, here is how I price it:

Remote setup ($1,500–3,000):

Needs assessment (including whether you should do this at all)
Deployment and configuration
2–3 channel integrations
3–5 skills or workflows configured
Security hardening
Training session and documentation
14 days of support

In-person setup ($5,000–8,000):

Everything above, on-site at your office
Multi-device configuration
Team training (up to 5 people)
Custom skill development
30 days of priority support

Where you land within each range depends on complexity. A straightforward setup with one messaging channel and email is on the lower end. A privacy-hardened deployment with custom skills, local models, and multiple users is on the higher end.

Monthly Infrastructure Costs

Hosting: roughly $5–40/month

Your assistant needs a computer to run on. The usual options:

A small cloud server (VPS) from providers like Hetzner, DigitalOcean, or AWS Lightsail. A machine with 2 CPUs and 4GB of RAM is enough for most single-business setups, and typically runs somewhere between the high single digits and the mid-twenties per month depending on the provider. Prices change, so check current pricing.
A dedicated or larger server if you’re running multiple agents or heavier workloads.
Your own hardware. A spare machine in the office costs only electricity, but then you’re responsible for uptime, backups, and physical security.

If you want to run AI models locally for privacy, you’ll need considerably more memory (and ideally a capable GPU or a recent Mac), which changes the hosting math.

Storage, domain, SSL: usually minor

Most servers include enough storage for an assistant’s data unless you’re processing large document libraries. A custom domain is optional (around $10–20/year), and SSL certificates are free via Let’s Encrypt.

Monthly AI Model Costs

This is the variable part, and it depends entirely on how much you use the assistant.

How AI pricing works

AI providers like OpenAI, Anthropic, and Google charge per “token,” which is roughly a piece of a word. Every message your assistant reads, every document it processes, and every reply it writes uses tokens. Input (what the model reads) is usually cheaper than output (what it writes).

The providers publish per-token prices, and those prices change often, usually downward for a given level of capability. Rather than print a table that will be out of date in a few months, here’s the pattern that holds:

Small, fast models cost a fraction of a cent per typical message. They’re fine for sorting, short replies, and simple lookups.
Top-tier models cost several times to tens of times more per token. They’re better at long documents, nuanced writing, and multi-step reasoning.
Local models (run on your own hardware with tools like Ollama) have no per-use cost, but they need hardware and are generally less capable than the best hosted models.

Check the providers’ current pricing pages before you budget.

Rough monthly ranges

For a single business, in my experience the model bill usually falls in these bands:

Usage Level	Description	Typical Monthly Range
Light	A few dozen messages a day, short replies	Under $15
Moderate	50–100 messages a day, some documents	$15–40
Heavy	Hundreds of messages a day, frequent documents, multi-step workflows	$40–100
Multi-user	Several people, large documents, multiple agents	$100+

The biggest lever is configuration. Routing simple tasks to a cheaper model and saving the expensive model for work that needs it can make a large difference in the bill.

Example Budgets (Hypothetical)

These are illustrative, not client figures. They show how the pieces add up.

Solo professional, light use

Small VPS: ~$10/month
AI usage, mostly a small model: ~$10–15/month
Total: roughly $20–25/month

Small office team, moderate use

Mid-size VPS: ~$25/month
AI usage, mixed models: ~$30–40/month
Total: roughly $55–65/month

Privacy-focused setup with local models

Larger server or in-office hardware: ~$40/month equivalent
AI usage, local model plus occasional hosted model: ~$20/month
Total: roughly $60/month, plus the upfront hardware

Several agents with managed support

Hosting: ~$30/month
AI usage: ~$80/month
Managed service: $500+/month
Total: roughly $600+/month

The Costs People Forget

Your time

If you manage it yourself, plan for a few hours a month on updates, troubleshooting, and tuning. That’s real time, even if it doesn’t show up on an invoice.

Security and upkeep

An assistant that can read your email and send messages on your behalf is a meaningful security exposure if it’s misconfigured. Updates, access controls, and reviewing what third-party skills or plugins actually do are ongoing work, not a one-time task.

Growth

Costs scale with usage. If the assistant starts handling more messages or more complex workflows, the model bill rises. That usually means it’s being used, which is good, but budget for it.

Managed Plans

If you don’t want to think about any of this, I offer a managed plan ($500–1,500/month) that covers monitoring, maintenance, cost tuning, new workflows, and security updates.

To be direct: a managed plan only makes sense if you’re using the assistant heavily enough to justify it. For a single person with light use, it usually doesn’t.

When It’s Worth It (and When It Isn’t)

The simple math: if an assistant saves you an hour a day and your time is worth $50 an hour, that’s roughly $1,000 a month of value against a monthly cost that’s often under $100. That’s a strong case, if it actually saves the hour.

It often isn’t worth it when:

You’d only use it occasionally. A $20–30/month ChatGPT, Claude, Copilot, or Gemini subscription will cover occasional use with no infrastructure.
The software you already own does the job. Microsoft 365, Google Workspace, and many CRMs now include AI features. Use those first.
Nobody will own it. A self-hosted assistant needs someone, internal or external, responsible for keeping it running and secure.

Getting a Number for Your Situation

If you want a clear estimate for your business, tell me a bit about how you work. I’ll look at what you already pay for, what you’d actually use, and give you realistic setup and monthly costs, including the option of not building anything new.