Solutions · 01
Servers
The machine a company already runs. A model on it does the repeated work that is sent to an API today.
The hardware
- Rack servers and workstations
- Virtual machines in a private cloud
- The CPU, no accelerator
- Air-gapped machines
What it has to work with
- Memory
- 16 to 256 GB
- Power
- Mains, always on
- Connectivity
- Inside the company network
- Missing
- A reason to send the data out
01
Sorting and routing
Tickets, emails and documents read as they arrive and sent to the right queue, with the fields a system needs pulled out on the way.
02
Summaries and drafts
A thread summarized, a reply drafted, a record filled in. The large model stays for the hard questions.
03
Search over internal documents
Contracts, manuals and records found by meaning, on the machine that holds them.
Why not the cloud
An API charges for every request, and the repeated work is most of the requests. The data in those requests is the company's own. Running the small part on a server in the building removes the bill for it and keeps the data where it was.
Work with us
Building on hardware like this? Get in touch and we will work out what we can do together.