AI agents and assistants for your company
We build AI assistants and agents that work with your company's own information: documents, tickets, emails, code. They run on a private model on your hardware, or on a cloud model you choose, with clear limits on what they can see and do.
When this helps
- Staff paste confidential documents or code into public AI tools.
- The same questions are answered by hand every day, from information that already exists somewhere.
- Requests, emails or tickets are sorted and forwarded manually.
- You want to try AI but don't know what it would cost to run or how to keep the data safe.
How we do it
- 1
Pick one task
We start with one concrete job, for example answering questions from your manuals or sorting incoming requests, and agree how to tell whether it works.
- 2
Choose where the model runs
A private model on your own hardware when data must stay in-house, or a cloud model when that is acceptable. You get a hardware sizing or a running-cost estimate before anything is bought.
- 3
Connect it to your data
The assistant gets read access only to the sources it needs, through accounts and API keys you control.
- 4
Test, then hand over
We test with real questions from your team, set limits, and document how it works and how to switch it off.
What you get
- A working assistant or agent for one defined task
- Hardware sizing or a running-cost estimate
- Access control: accounts, API keys, logging
- Documentation and a handover session
Questions
Can AI run without sending our data to the cloud?
Yes. Language models can run on your own servers or workstations. We have run a 27-billion-parameter model across four existing GPU workstations for an internal team; how large a model you need depends on the task.
Do we need to buy new hardware?
Not always. We first check what you already have. If new hardware is needed, you get a sizing note before buying anything.
What's the difference between an assistant and an agent?
An assistant answers questions. An agent also takes actions, such as sorting a request or drafting a reply, within permissions you define.
Which tools do you use?
For private models, llama.cpp and Open WebUI with an OpenAI-compatible API. For cloud models, the provider you choose.