The cost of an AI agent has three parts: installation, support, and model usage. There is no fixed price, because the scope decides how much each part weighs, so you always get a custom quote. Two quotes that look alike can also cover very different work. This article explains each cost category, which factors drive the price, and what to look for when comparing quotes, so you compare like with like.
Installation and the pilot are a one-time cost. Support is priced in a custom quote, and model usage varies with how much you use the agent and which model you choose. Which part weighs heaviest differs per business and per task. The table shows what is one-time and what is ongoing.
| Cost item | One-time | Ongoing |
|---|---|---|
Installation and pilot | Yes, depending on hardware, onboarding, and integrations | — |
Support | — | Yes, tailored support: updates, monitoring, and help, on a custom quote |
Model usage | — | Yes, variable, through your own provider account |
Hardware (optional) | Yes, if new hardware is needed for your local setup | — (electricity and maintenance) |
Installation covers setting up Hermes on a system at your business, configuring projects and a knowledge base, setting up access and any integrations, and onboarding two to three employees. The pilot then tests one task in your own environment and usually takes 2 to 4 weeks, from the first conversation to the evaluation.
The price depends on complexity: a simple setup without integrations costs less than an environment with CRM connections and custom scripts. We discuss this openly in the first conversation.
A single task with a fixed workflow costs less than a process that runs across multiple steps, systems, and exceptions. That sounds obvious, but in quotes the difference is often invisible, because both go by the name "installing an AI agent." Always ask which tasks and steps the installation covers exactly.
Support covers Hermes updates, monitoring of the installation, backups as agreed, and help for your team. It starts after the pilot. Backups are encrypted and stored at a separate location on your network, never on the node. You agree on the scope with Voltti and receive a custom quote. Read more about the service on our approach page.
To make quotes comparable, we scope every project the same way and split it into three groups. Included: installation on your system, onboarding for your team, agreements about goals and boundaries, updates and monitoring during support, and human review at moments you decide.
Optional and priced separately: extra integrations such as your CRM or project tool, custom scripts, a computer from Voltti, and broader support than the standard package.
Outside the quote: model usage on your own provider account, your own time to review the output, and power and upkeep on your own hardware.
That depends on how much you use the agent and which model you choose, and it always runs through your own account. Hermes uses a language model for its reasoning. That model can run locally on your own hardware, without a bill per conversation.
You can also use an external provider such as OpenAI or Anthropic with an API key, and then you pay per use. A third route is a subscription you already have that Hermes supports as a model provider, such as ChatGPT or Codex and GitHub Copilot. Our comparison of local AI vs. cloud AI helps you choose.
A subscription does not cover the same thing everywhere: that differs per provider. Claude Pro does not work through this route. For ChatGPT plans, Hermes does not yet document how usage counts against your plan. So check your subscription limits before you run daily tasks on it. The Hermes documentation on model providers lists which providers and subscriptions Hermes supports. Already paying for ChatGPT or Copilot? Read Hermes vs. ChatGPT and Copilot to see whether that changes your choice.
This usage runs through your own account, with the model provider or through your subscription, not through Voltti. You control what you spend and can set limits.
Because the cost of an AI agent follows the work behind it, and that work can differ more than a price list shows. One fixed task without integrations is a different job from a process that runs across three systems, with exceptions and approvals. So we quote per task, after a first conversation about that task.
The one part you can estimate yourself is model usage. Check the current rates on the providers' own pricing pages, such as OpenAI API pricing and Anthropic pricing for Claude. Those rates change, so we do not copy them here. Then ask every vendor to quote on the same task, so the totals mean the same thing.
AI agent pricing usually follows one of a few patterns. Software platforms often charge per user, per conversation or action, or per resolved request. Agencies that build an agent often charge a one-time build fee plus a recurring fee for maintenance.
With Voltti, you pay once for installation and the pilot, support is priced in a custom quote, and model usage runs through your own account. Whatever the pricing model, compare what is included, not just the headline number.
Look at the scope first, then at the price. Comparing two quotes on price alone tells you nothing. Eight factors together determine how much work an installation is. For every quote, ask what exactly is included on these points; we walk through them below.
An agent for one well-defined task, such as following up on quotes, takes less time to set up than a process across several steps and systems. More steps means more testing.
Every integration with your CRM, email, scheduling, or invoicing requires configuration and permissions. Existing API access makes it cheaper; custom access makes it more expensive.
Is your process documented in files the agent can read, or does that need cleaning up first? Messy input means more preparation and longer testing.
The more exceptions a process has, the more test cases are needed to show the agent works reliably. Testing is not a side note; it is the proof.
Minimum permissions require setup: separate accounts, read-only where possible, write access only where needed. A quote that says nothing about this is skipping a risk.
Who adjusts the agent when your process changes? Support ranges from ad hoc hours to ongoing help; the scope is in the quote.
Running locally does not bill tokens per conversation, but requires hardware. You pay for an external model per use with an API key, or within the limits of a subscription Hermes supports. That always runs through your own account. Every route weighs on total cost.
Hermes runs on a machine you already have or choose. Only if you do not have a suitable computer does Voltti supply one; that is an option, not a requirement.
Before you sign anything, check these points. A quote that does not mention them is not a quote, just a price.
The exact tasks and steps the agent will perform, not just the name of the tool
Which integrations are included and which are billed separately
What happens in the pilot and what you can expect afterward
Which permissions the agent gets and where the boundaries are
Who reviews the output and when a person must approve
What support includes: updates, monitoring, availability, recovery
Where your data lives and where backups go
How model usage is handled: your own account or a pass-through
Where do you collect those quotes? Besides your own network, you will find online directories that put AI agent agencies side by side. One of them is DesignRush's list of AI companies, and Voltti has a profile on DesignRush. Use such a list as a starting point for your first selection, not as a replacement for the checklist above. Before you ask for quotes, prepare your AI agent pilot in seven questions, so vendors quote on the same task and the same boundaries.
You get more than software. You get a working environment with support, updates, and backups, plus help for your team and someone to think through new use cases with you. See what an AI agent for business does day to day.
Public support can cover part of the training or development around an agent, but rarely the tool itself. Read our overview of AI grants and support in Belgium and the Netherlands.
For the broader choice, compare an AI agent with a chatbot, or explore whether to build or outsource your own AI assistant.
Wondering whether it makes sense for your situation? Discuss your pilot and we will explore the options together. Still deciding between doing it yourself or having it done? On AI agent development you can read what Voltti sets up for you and which first tasks are a good fit.
Sharing this with Dutch-speaking colleagues or clients? There is also a Dutch version: Wat kost een AI-agent voor een bedrijf?
No. Every environment is different: hardware, integrations, and the level of support vary by business. That is why you always receive a custom quote. Tell us which task you want to start with, and get in touch with Voltti for an indication of that scope.
No. Model usage runs through your own account, with an API key from the model provider or through a subscription Hermes supports. It runs directly and without any involvement from Voltti. That gives you full control over your spending and lets you set your own limits.
The pilot is an evaluation period: one task, tested in your own environment. If it does not work, we discuss what went wrong and whether adjustments are possible. There is no obligation to continue with support; you decide after the evaluation.
Not necessarily. Hermes runs on a machine you already have or choose, so if you have a suitable computer, we can use it. Otherwise, we can supply hardware as an option. What counts as suitable depends on whether you run the model locally; read what computer you really need for local AI. We discuss this during the intake.
