Local AI models
Not every company can send data to an outside cloud: law firms, clinics, businesses with trade secrets. Open language models can now run on your own hardware and be used for everyday work.
What we do
- Pick a model and hardware to fit the work and the budget
- Install and configure a model server on your network
- Connect the model to your tools and documents
- Test quality on your own tasks before the model reaches your team
What we run ourselves
We use a local model every day. Qwen3.6 35B runs for us on an ordinary computer with an RTX 3060 graphics card (12 GB). It generates about 30 tokens a second and holds up to 262 thousand tokens of context, which is several hundred pages of text at once.
We measured those numbers ourselves, on that hardware. For your company we pick a model and hardware for the actual tasks and measure them the same way.
What we need from you
- What data the model will handle and who will use it
- Access to hardware, or a decision to buy some, which we'll help with
- Example tasks to test quality on
Other services
Process automation
Data moves between your systems without retyping, reports are built on a schedule, and you get an alert when something needs a person.
AI assistants
An assistant answers your customers or your team from your own documents, offer, and rules.
Let's see what you could automate.
A short, free call. We'll tell you plainly whether it makes sense.