Local & private AI · practical answers
What hardware do I need for local AI?
Scope and starting price
Choose the relevant service below. We confirm the systems covered, checks and final price before work begins. Additional work and third-party costs need to be included in the agreed quote.
Hardware Suitability Assessment
From £79
Assess the existing or proposed machine against the intended AI workload before an installation or purchase.
Local & Private AI servicesModel / Quantization Selection
From £79
Compare model formats and sizes that fit the machine and the job you need to complete.
Local & Private AI servicesLocal AI Workstation Setup
From £199
A wider workstation configuration, including agreed monitoring and resource controls.
Local & Private AI servicesSize the work before the machine
A short drafting task and a coding assistant reading a large project ask different things of a computer. Agree the kind of input, typical length, expected output and whether several people will use it at once. These details make a hardware recommendation more useful than a generic minimum specification.
Your assessment can then separate what is already practical, what would need a smaller model or shorter input, and what would justify an upgrade. A purchase recommendation should explain the limitation it solves.
Memory and compatibility both matter
Ollama documents that increasing context length increases the memory required to run a model. Leave room for the actual working context when evaluating a machine, rather than treating model storage size as the complete requirement.
Technical reference: Ollama: context length and memory requirements
Check the exact graphics card and driver
Ollama publishes supported hardware and driver requirements. Compatibility depends on the particular device and software combination, so send the full card model and operating system rather than only a brand name.
For a proposed purchase, include the complete specification, your budget and constraints such as desk space or power use. For existing equipment, the useful outcome is a written choice between reuse, a targeted upgrade, or a different model and workload.
Technical reference: Ollama: supported hardware
Is this the right fit?
A good fit when
- You want to reuse a workstation or server before buying new equipment.
- You have a proposed machine and want to check it against a particular local AI task.
Limits of the scope
- The assessment price does not include hardware, installation or repairs.
- A specification review cannot promise the same speed for every model, input length or number of users.
What to prepare for an enquiry
- The full computer specification, or a link to the proposed machine.
- Your intended model or application if selected, and an example of the work.
- How many people will use it, whether it must work offline, and your hardware budget.
Start with a description and non-sensitive examples. Leave out passwords, API keys, private keys, recovery codes and customer data. Access arrangements are agreed separately if the work needs them.
Common questions
Do I always need a dedicated GPU?
No single answer fits every task. Ollama can use system memory and CPU processing, but whether that is useful for you depends on the workload and acceptable speed. A representative trial is more useful than assuming a GPU is always required.
Is more memory enough to solve slow responses?
It may help a memory limit, but first identify the actual constraint. The selected model, working context, processor, graphics support and other running applications all need to be considered.
Will you recommend a specific model?
Model and quantization selection is available as a defined service. The recommendation should be tied to examples of your work and the machine you intend to use.
Related answers
Local & private AI
How much does it cost to set up private AI?
Local & private AI
Can AI answer questions about our private documents?
Linux infrastructure