The AI behind our systems.
We build on the leading AI models and choose the right one for each job. Here is what we use, how it works inside your business, and how we keep it safe and accurate.
The models we build on.
- Our default for conversations, writing and careful reasoning. The chat on this site runs on Claude.
- Used when it is the best fit for a job, or when your team already works with OpenAI.
- Another model we test each job against, and use where it performs best.
- For private AI, open models run on hardware in your office or private cloud, so nothing leaves your control.
How we choose for each job.
Fit for the job
Larger models for complex reasoning and long documents, smaller fast ones for routine replies and bookings.
Cost per conversation
We pick the model that does the job well at the lowest running cost, and tell you what that cost is.
Where your data may go
Business cloud APIs for most work, or a private server in your office when your files can't leave.
How AI works inside your systems.
Grounded in your information
It answers from your documents, price sheets and policies, and shows where an answer came from.
Works in your tools, with limited access
It books, updates and looks things up in the systems you approve, and only with the access each job needs.
Hands over to a person
When it isn't sure, or someone asks for a person, it passes the conversation to your team with the full context.
Says it's an AI
Every call and chat starts by saying it's an AI, in a line you approve.
How we protect your data.
- We use the providers' business APIs. Under their commercial terms, your data isn't used to train their models by default.
- Each agent gets only the access its job needs, and nothing more.
- For healthcare we sign a business associate agreement and use only services that sign one too.
- When files must stay in house, we run the AI on your own server instead of a cloud API.
How we keep it accurate.
Tested before launch
We run it through real scenarios from your business with you, and change it until the answers are right.
Reviewed every month
We read the conversations it couldn't finish, fix what it got wrong, and send you a short report.
Changes on record
Every change to its instructions or knowledge is logged, so we can see what changed and roll it back.