Local AIWith control
Local AI for businesses means the AI environment runs on a system at your premises, and you choose which language model runs locally or with an external provider. You decide what fits your use case and budget.
What local AIMeans
With Voltti, the Hermes environment and local task execution run on a system at your premises. That does not automatically mean the language model runs offline or that every connected service is local.
01
Local environment
Hermes and its configuration run on your own hardware. You choose which projects, knowledge sources, and tools it can access.
02
Model choice
Use an external model provider, a local model, or both, depending on the task and data requirements.
03
Separate backups
Encrypted backups go to a separate location in your network, never to the node itself.
Cloud, localOr hybrid
OpenAISeparate conversation
Answers questions
- Submitted content goes to the provider
- Execution runs on the provider's infrastructure
- Availability depends on external infrastructure
In a cloud-hosted chat product, submitted content is processed by the provider. Data location, retention, and controls depend on the service, plan, settings, and contract. Review those terms before sharing business data.
AI agentConnected
Takes action
Local execution
Hermes runs on a system inside your network. You decide which local files, remote repositories, and business tools it may access.
Flexible model choice
Compare an external provider's capabilities with a local model's hardware and privacy requirements. Choose per use case rather than assuming one setup suits every task.
Control over backups
Encrypted backups to a separate location you choose within your network.
Voltti installs the agent locally. If you choose an external model, the context sent to it must fit the agreed data policy. Connected services and repositories have their own data flows.
| Capability | OpenAI | AI agent |
|---|---|---|
| Local execution | Submitted content goes to the provider | Yes — Hermes runs on a system inside your network. You decide which local files, remote repositories, and business tools it may access. |
| Flexible model choice | Execution runs on the provider's infrastructure | Yes — Compare an external provider's capabilities with a local model's hardware and privacy requirements. Choose per use case rather than assuming one setup suits every task. |
| Control over backups | Availability depends on external infrastructure | Yes — Encrypted backups to a separate location you choose within your network. |
| Component | Where it runs | What you control |
|---|---|---|
Environment | Locally, on a system at your premises | Projects, knowledge, and access |
Language model | Locally or with an external provider — most setups are hybrid | Which model you use per use case, and which context goes to it |
Backup | Encrypted, to a separate location in your network | The location within your network |
At your premisesYour setup
A local AI setup with Hermes runs on a computer in your office. Use existing or chosen hardware, with a computer supplied by Voltti as an option. Voltti installs it and provides support. Work in Hermes Desktop yourself or through a supported team channel you already use.
AI AGENTHermes
Slack
Telegram
Discord
Buzz
What hardware do you need to run AI locally?
That depends on the model. Small models run on 16 GB of memory. For a new computer that combines a larger local model, an agent, and your everyday work software, 64 GB of system RAM is a sensible starting point. Heavier models need more: keeping everything in memory is faster and simpler with long context.
With an external provider you do not need that computing power; a solid computer is enough. In practice, most businesses choose a hybrid AI setup: the environment runs locally, and the model comes from the provider that fits the task. Compare the options in local AI vs. cloud AI.
Voltti installs Hermes Agent and provides support. The Voltti team serves businesses in Belgium and the Netherlands.
Questions and answers
01Does all our data stay local?
Hermes runs locally, but that does not guarantee every data flow stays local. External models receive the context sent in a request; connected channels and services have their own processing. Together, we decide which models and connections fit your data. If personal data is involved, the GDPR applies in every setup; the EDPB offers a data protection guide for small businesses. Read what stays local in a hybrid setup and how Voltti handles access and encrypted backups.
02What is the difference between local and hybrid AI?
In a fully local model setup, both the agent and language model run on your hardware. In a hybrid setup, the agent runs locally while selected model requests go to an external provider. Web searches, messaging channels, and other integrations may still use online services in either setup. Read the comparison of local AI vs. cloud AI.
03Can I switch to a local model later?
Yes. You can change the model provider. We can also add a local model if it suits the task and your hardware can run it usefully. You do not have to commit to one model up front: a different model can handle each task. A pilot usually takes 2 to 4 weeks, so you evaluate the environment first and the model choice after that. See how a pilot with Voltti works, step by step.
04What if there is a power outage?
A power outage stops work on the affected system until power returns. A backup is not a substitute for a power supply, but it does help with recovery if hardware or stored data is damaged. That is why the environment is backed up daily, encrypted, to a separate location in your network, never to the node itself. Read how backups and recovery for an AI environment work.
05Do I need special hardware?
A suitable existing computer can be enough when using an external model. Local model requirements vary: small models may run on a CPU, while larger models can need more memory or a GPU. If you buy a new computer for larger local models alongside your everyday software, 64 GB of system RAM is a sensible starting point. We assess the task and hardware during the intake. To check whether a model runs on the GPU or the CPU, see the Ollama documentation, for example. Read what computer you need to run AI locally.
06What does local AI cost?
That depends on your setup, so you receive a custom quote. The costs have three parts: installation and pilot, support, and model usage. With a local model, hardware and electricity weigh more, and there are no provider fees for model usage. With an external model, a solid computer is enough, and usage runs through your own account with the provider. Read what an AI agent costs for a business.
Read more
How we built Sisu on a Raspberry Pi 5 with Hermes and Home Assistant
See how Sisu combines Hermes, Home Assistant, and local device APIs on a Raspberry Pi 5 for practical control through chat.
When do you need a Linux computer for an AI agent
See when Hermes on your own computer is enough and when a managed Linux computer is a better fit for your team and recurring tasks.
What happens to your AI environment if hardware fails?
Hardware fails. Learn how Voltti handles backups and recovery tests, and what you can do to prepare your AI environment for a hardware failure.
Running AI locally: what computer do you really need?
What computer do you need for local AI? Compare models, RAM, unified memory, NVMe offload, and hardware prices. Start with your tasks and model choice.
What stays local in a hybrid AI setup?
In a hybrid AI setup, the environment runs locally while language processing can use an external provider. See exactly what stays with you and what leaves.
AI and business data: a practical safety checklist
Seven checks for safer AI use in your business: selected access, minimum permissions, backups, and human oversight. Review your setup before connecting data.
Local AI vs. cloud AI: which fits your business?
Local AI runs on your hardware; cloud AI runs through a provider. A hybrid setup suits most businesses. Compare the options and make an informed choice.