AI models that can run within your business
Local AI means the model processes requests on a computer or server controlled by your company. It can support document work, drafting, information extraction and application integration. We start with a specific task and a pilot project.
Which models can be installed locally?
Qwen — models for conversation, reasoning and coding assistance. We can evaluate a version suited to your documents and the languages your team uses.
Gemma — Google’s family of models available for deployment on devices and workstations. Text and image capabilities depend on the version selected.
Ministral — Mistral models designed for devices and local deployment. They can be assessed for internal assistants, text processing and application integration.
OpenAI gpt-oss — models with downloadable weights, including gpt-oss-20b, that can run locally. Deploying them does not install the ChatGPT application or provide all of its features.
These are examples of options, not a fixed package. Before deployment, we check the version’s licence, software support, computing requirements and response quality. Locally available models do not all share the same licence.
Cloud AI and local AI: what is the difference?
Browser or local? “Web” describes the interface, not where processing happens. Your team can also access a local model through a browser. The main difference is where the model runs and who manages the infrastructure.
Cloud AI: requests are processed on the provider’s infrastructure. You usually do not need your own AI server, and the provider manages the service. Costs may include subscriptions or API usage; access, limits and data handling depend on the service and contract.
Local AI: processing takes place on your company’s infrastructure. You have more direct control over access, models and external connections. Hardware, power, updates, monitoring and support still need to be provided. Local deployment alone does not guarantee security or correct answers.
What hardware and connectivity are needed?
Requirements depend on model size, RAM, GPU memory, model optimisation, document length and concurrent users. Some compact models can run on suitable computers; others need workstations or GPU servers. We recommend a configuration after assessment.
Some setups can operate without internet access after installation, if all required components run locally. Updates, online search and cloud integrations may still need connectivity. We explicitly define which data may leave the infrastructure.
How do we choose the right setup?
We define the task and access rules, test a model using relevant examples, then assess quality, speed and total cost. Integration with internal documents, CRM or ERP, together with the maintenance plan, is based on the pilot’s results.




