Getting started
How we get started
What happens first, what you sign, what you need to buy, and what the AI costs to run.
- What happens first?
- A conversation. You describe the work that takes too much time, and we pick one task to start with. You get a written description of how it would run, where a person approves, and what stays on your systems.
- Do we sign a non-disclosure agreement?
- Yes. A written services and confidentiality agreement is signed before we see any real document or client record. Until then we work from your description and made-up samples.
- What hardware do we need to buy?
- Usually none. Most of the work runs on a cloud AI model through your own company account. If some documents should stay in your office, part of the work can run on a computer you own, and we help you choose hardware that fits the job. Local models are improving but still fall short of the best cloud models on long or complex documents, so we test them on your own samples before you buy anything. If you go ahead, you buy the hardware at cost, with the supplier’s invoice, and you own it.
- Which AI models will be used?
- In the cloud: a commercial model from a provider your company signs up with. The software uses your company’s own API key, so the account, the bill and the data terms are yours. On your own hardware: open-weights models, which are free to run and licensed for commercial use, for the tasks they handle well. We try the candidates on your own sample cases and choose the smallest one that does the job. You always know which model is running.
- How is the cost of the AI managed?
- The provider bills your company directly, and you set a monthly spending limit in your own provider account. Inside the software, every AI request is counted per person, and you receive a usage report each week.
- How much will the AI cost to run?
- On your own hardware there is no charge per use: you pay for the computer once, then for electricity and upkeep. With a cloud model the provider bills your account by use, so the cost follows your volume. Either way, you get a written estimate for your workload before you commit.
