Your goal
Be visible where AI answers
Automate processes
Build a product
Put AI agents to work
Connect and modernize systems
Know where we stand
Use Cases
CRMStrengthen customer relationshipsPopularE-CommerceBoost online revenueBooking System24/7 appointment bookingProject ManagementCoordinate teamsInvoicingGet paid fasterAnalyticsData-driven decisionsPut AI agents to work
Setting up local AI hardware with open language models and AI agents in-house
Fixed price after scoping · proposal within 48 h
Reviewed: September 2026
Local AI agents are AI assistants whose language model and agent software run on hardware inside the company rather than at a cloud provider.
We set up AI infrastructure on your premises: a Mac Studio, an NVIDIA DGX Spark or a machine with AMD Ryzen AI Max – depending on which models and how many users you need. It runs open language models such as Qwen, DeepSeek, GLM or gpt-oss through a local runtime like llama.cpp, Ollama, MLX or vLLM. For the agents we use Hermes Agent, an open-source agent framework by Nous Research: with memory, reusable skills, a scheduler and connections to Slack, Telegram or email. Via MCP the agents access your systems – with approvals you define. Requests and documents are processed on your hardware; no cloud provider sees them.
Ideal for companies that want to use AI with confidential data – customer data, contracts, engineering drawings, HR files – and cannot or do not want to hand that data to a cloud provider. A strong fit for law firms, medical practices, manufacturing and SMEs with high requirements for data protection and trade secrets.
We select the optimal tech stack for your specific requirements
The machine on your premises. Memory is what matters: it decides how large a model can run.
| Class | Devices | Memory | What realistically runs |
|---|---|---|---|
| CompactSingle teams | NVIDIA DGX Spark · machines with AMD Ryzen AI Max+ | 128 GB | Models up to about 120B parameters, e.g. gpt-oss-120b |
| WorkstationDepartments | Apple Mac Studio with M5 Ultra | up to 512 GB | Large mixture-of-experts models with several hundred billion parameters |
| ClusterSeveral departments | 2–4 networked Mac Studios or DGX Sparks | 256 GB and more, distributed | Very large models spread across several devices |
| EnterpriseCompany-wide | NVIDIA DGX Station · workstations with RTX PRO 6000 | 96 GB to about 780 GB | Many concurrent users in production |
Rule of thumb: in 4-bit quantisation a model needs about 0.6 GB of memory per billion parameters, plus room for the context. We make the concrete choice in the workshop, based on your tasks.
The software that runs the model and exposes an OpenAI-compatible interface.
MLX on the Mac, vLLM or SGLang on NVIDIA, llama.cpp on AMD – whatever runs fastest on your hardware.
Open models whose weights live on your hardware. We choose by task, language and licence.
| Family | Strength | Licence |
|---|---|---|
| Qwen (Alibaba) | Versatile, strong at coding and tool use | Apache 2.0 for small and medium variants |
| DeepSeek | Reasoning, very large variants | MIT |
| GLM (Zhipu) | Agentic tasks and coding | MIT or own licence, per variant |
| gpt-oss (OpenAI) | Compact, with tool calling | Apache 2.0 |
| Mistral | European vendor, multilingual | Apache 2.0 |
| Gemma (Google) | Small to medium models, images too | Apache 2.0 |
New versions appear every month. We test with your own data before rollout and check the licence for each model.
Hermes Agent plans tasks, uses tools and remembers earlier conversations.
Hermes Agent is open source (MIT licence) and made by Nous Research.
Local does not automatically mean secure. That is why this is part of the setup:
A facilitated session that ends with a ranked list of your projects.
½–2 days · three fixed pricesWe set up one clearly bounded system and hand it over ready to use.
1–2 weeks · fixed price after scopingWe build the project out and stay alongside you once it is live.
after scoping, ongoing · fixed price after scoping; support billed monthlyIn a workshop (½–2 days) we clarify tasks, users and data and decide on hardware, models and licences.
T1Setting up the machine on your premises, the runtime and the selected open models.
W1Hermes Agent with skills and connections to your systems, access rights, onboarding and documentation.
W2Optional: further agents, integrations and user groups step by step, scope after scoping.
BuildTalk to us for 30 minutes with no obligation, or write to us directly.
Fixed price after scoping · proposal within 48 h
