Your own AI assistant for business: build or outsource?

You can build your own AI assistant with a chat tool, a no-code platform, or an open-source agent such as Hermes. You can also have a partner build it. Building it yourself fits when you have the expertise and time in-house. Having it built fits better when the assistant works with sensitive data and several tools, and someone has to own the maintenance. If you only rewrite a piece of text now and then, a regular chat tool is enough. This article lays out the trade-offs, including the points vendors would rather not discuss.

Illustration of a glowing blue hub connected by lines to icons for documents, folders, calendars, and settings — Your own AI assistant knows your projects and tools.

What is your own AI assistant?

Your own AI assistant is configured for your business environment. The agent knows your projects, agreements, and tools, and works within boundaries you set. Your team talks to it through the channels it already uses: chat, desktop, or email. Technically, such an assistant is an AI agent. If that term is new to you, first read what an AI agent is and how it works.

At Voltti, we build that environment with Hermes Agent, installed on a system at your business and supported by us.

Building an AI assistant yourself: four options and their limits

Want to build an AI assistant without a partner? You can, and for some organizations it is the right choice. There are roughly four options, each with its own limits.

01
A chat tool with your own instructions
One accessible option is a configured chat tool: a Custom GPT or Copilot instructions with your own documents attached. This fits if you mainly want text and answers that match your house style. Start with one recurring question and a few sample documents. Connections to your CRM, scheduling, or files may be available through apps, actions, or connectors, but require extra configuration and depend on the platform and your subscription. Processing takes place at the provider. Unsure whether a chat tool like that is enough, or whether you need an agent? Read AI agent or chatbot: the 7 differences.
02
A no-code platform such as n8n
The second option is a no-code platform such as n8n. You build automation flows that connect email, forms, and tools. This fits if someone on your team likes to map out processes. Start with one workflow that runs manually today, such as a form that needs to end up in your CRM. You get more freedom, but you build and maintain every scenario yourself, and your team handles every change.
03
An open-source agent on your own hardware
The third option is installing an existing open-source agent on your own hardware, such as OpenClaw or Hermes. This comes closest to the traditional idea of an assistant. The agent runs on a system you administer, communicates through channels you already use, maintains memory and skills, and works with your tools. An installation script or desktop installer is enough to begin. Installation, permissions, updates, backups, and output review do fall entirely to you or your team.
OpenClaw is open source under the MIT license and connects the agent through a gateway to channels such as Discord, Slack, Teams, Telegram, and WhatsApp. Hermes is an open-source framework for AI agents from Nous Research. It runs in the terminal, on messaging platforms, and in IDEs, and works with the language model of your choice. Start with a test environment and one channel. Later in this article, we explain how to install Hermes yourself.
04
Writing your own software
The fourth option is writing your own software against model APIs and your own systems. This fits if you have developers in-house who also want to own the maintenance. It gives you the most control and demands the most work: prompts, integrations, security, and monitoring. To depend less on external model providers, you can also run models yourself. Try a small model first with Ollama or LM Studio. Larger models need heavier hardware and different runtimes. Our article on what computer you need to run AI locally explains how to choose. Expect to need much more expertise and maintenance as well.
One thing people often underestimate: an assistant is never finished. Models change, integrations break, and someone has to keep checking the output. If you build it, you maintain it too.

When should you stop building it yourself?

A practical rule of thumb: if you spend more time building, testing, and fixing the assistant than on the task you wanted to automate, you have reached the limit. Another warning is dependence on one employee. If exactly one colleague knows how the flow works, it is a risk rather than a dependable system. At that point, outsourcing is a decision, not an admission of failure.

Consideration	Build it yourself	Have it built
In-house expertise	The AI, integration, and security knowledge you need depends on the option	You do not need an in-house expert
Sensitive data	You decide and configure where data goes	Local or hybrid, agreed in advance
Maintenance	Entirely your responsibility and time	Support is part of the service
Time to launch	Depends on your own capacity	Pilot usually takes 2 to 4 weeks, from the first conversation to the evaluation
Costs	Your time, any hardware, and model or platform usage	Custom quote: installation and support; model usage separate
Control	Depends on the option and platform used	You set the goals and boundaries; the partner implements them

Having an AI assistant built: what to look for

Having an AI assistant built is a realistic option for many businesses: a partner provides the expertise, while you stay in charge of day-to-day use. But vendors differ. Ask in advance where your data stays, what permissions the agent gets, and who decides what the assistant may and may not do. Want to see what such a partnership looks like in practice? On the AI agent development page you can read what Voltti sets up, which first tasks are a good fit, and what you contribute yourself.

Also ask what happens when something goes wrong. Who fixes a broken integration on Friday afternoon, adapts the assistant when a process changes, and checks the output before anything is sent? With DIY, those are all your people. With a supported service, those responsibilities are written down, with human review built into the process.

Minimum permissions are the starting point, not an afterthought. In our own Hermes installation, each environment has its own credentials. Access to our search data is read-only; publishing requires separate writing permissions in the CMS. Read-only wherever possible, writing access only where it is needed.

Ask what working together looks like too. A pilot around one task tells you more than a platform presentation. Read more about how we set up a pilot, or about what an AI agent costs: always a custom quote, without a standard rate.

What do you contribute? Your partner handles the technical work. You know which processes take time, which tools need to be involved, and which rules the assistant must follow. The more clearly you define that first task, the sooner the pilot can produce something useful.

AI assistants for businesses: how to start small

Start with the task that recurs most often. Let the assistant work alongside your team for two weeks, evaluate it together, and only then expand. If the task does not work out, you have lost little. If it does, the assistant can take on more tasks. To scope that first task, read how to prepare your AI agent pilot in seven questions.

You are not alone in this: according to Statistics Netherlands (CBS), 33 percent of businesses in the Netherlands with ten or more people employed used one or more of eight AI technologies surveyed in 2025 (Digitalisering en kenniseconomie 2025, provisional figures).

Working with sensitive data? Consider a local or hybrid setup so that selected information stays within your own environment. Read more about data residency and human review on our pages about security and privacy. Voltti helps businesses in Belgium and the Netherlands move from a first pilot to a fully configured assistant.

A first month might look like this. Week one: choose and define the task. Describe the process as it runs today, including exceptions. Week two: the assistant works in the background without sending anything itself. Your team learns what the agent can and cannot do. Weeks three and four: the first tasks run for real, with a person checking the output and making adjustments. Only then do you decide whether to expand the assistant's role.

Installing Hermes yourself: Desktop or terminal

You do not need to hire Voltti to get started with Hermes. You can install the agent yourself without first programming your own AI assistant.

For most people, Hermes Desktop is the most convenient place to start. Download the Hermes Desktop installer for macOS or Windows and run it. It installs both the Desktop app and the command-line version: the same agent with a different interface.

Comfortable with Linux and prefer the terminal? You can install the full CLI version with the official command below. It also works on macOS and in WSL2. The script downloads Hermes, creates a Python environment, and makes the hermes command available.

On Linux, check that Git, curl, and xz-utils are available first. Desktop on Linux also requires a C++ compiler. Read the official installation instructions for your system's requirements. The command executes a downloaded script; inspect it first if you want to see what it changes.

Installing Hermes

$ curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

✓ Cloned via HTTPS
→ Creating virtual environment with Python 3.11...
✓ Virtual environment ready (Python 3.11)
→ Setting up hermes command...
✓ hermes command ready
✓ Installation Complete!

Selected messages from the official installation script, not a live installation log. Output varies by system.

After installation: continue yourself or get support

After installing the CLI, open a new terminal. Run hermes setup to configure your model, tools, and connections. Want to install and launch the Desktop app afterward? Run hermes desktop. Desktop uses the same configuration, sessions, and skills as the CLI.

Want to start on Windows without Desktop or WSL2? Run the official Windows command in native PowerShell: iex (irm https://hermes-agent.nousresearch.com/install.ps1).

Installation is one step. Deciding which data and tools the agent may use, testing it, and reviewing its output are part of the work too. Start with a clearly defined task and keep human oversight of important actions. Prefer to leave installation, configuration, and support to a partner? That is the support Voltti provides, on a system at your business. You decide the goals, permissions, and points where someone needs to approve an action. Read how we install and support Hermes for your business.

Frequently asked questions

01Can I just build an AI assistant myself?

Yes. You can configure a chat tool, use a no-code platform, install an existing agent such as OpenClaw or Hermes, or write your own software. The knowledge and time required differ for each option. Integrations, sensitive data, maintenance, and review still need attention, even when installation is simple. If you spend more time building and fixing it than on the task itself, having it built is often the better choice.

02How long does it take to set up an AI assistant?

A pilot usually takes 2 to 4 weeks, from the first conversation to the evaluation. During those weeks you choose and describe one task. The assistant first works in the background, then the task runs for real while someone on your team reviews the result. The exact timing depends on the chosen workflow and your team's availability.

03What does a dedicated AI assistant cost?

That depends on your workflow, integrations, and the support you want. We therefore always provide a custom quote, without a standard rate. The quote covers installation and support; language model usage comes on top. If you build it yourself, you mainly pay with your own time, any hardware, and platform costs. Read more in what an AI agent costs.

04Will our AI assistant work with our existing software?

An API or a usable interface may make an integration possible, for example with your CRM, your scheduling tool, or a shared folder of files. We assess each situation to see whether the required access and features are available, and whether the integration is useful and safe. Without usable access, an integration with that system is not possible. The agent only gets the permissions the task needs.

05Do I need to digitize my entire business?

No. We start with one task and one project. After evaluating that first task, you decide what comes next: extend the assistant to another task, adjust the approach, or stop. Other processes and systems can follow later, but only if they add value. That keeps the step small and keeps you in control.

Wondering whether your own AI assistant is a good fit?

Discuss your pilot