You are pointing this at the truth about your business. That only works if you can trust where the answers come from and who can see them. Here is exactly how that works.
From your systems to an answer, and no further.
Connect
You approve read-only access to the systems you choose. Our engineers do the connection work. You can start with two and add more later.
Reconcile and store
Your data is organized into your Company Model inside your own isolated tenant, hosted on AWS in the United States, encrypted in transit and at rest.
Answer
When someone asks a question, the engine reasons over your Company Model — not the open internet — and returns a plain-language answer with its sources, scoped to that person's role.
Nothing else
The data is not sold, not shared between customers, and not used to train models. It exists to answer your questions and for nothing else.
The short, honest list.
Answering questions with AI means a few specialist services see the relevant data in the moment. These are they, and what each one does.
Anthropic (Claude)
The reasoning engine that reads your data and writes the answer. Accessed through the business API, where inputs are not used for training.
OpenAI (embeddings)
Turns text like meeting notes into the numeric form that makes search work. Business API, no training on inputs.
Amazon Web Services
Hosting, database, and encrypted storage, in the United States. Your tenant and your secrets live here.
The risks that come with AI, handled on purpose.
A system that reads your business and acts on language has failure modes older software never had. We designed for them: business records like deal names, form fields, and transcripts are treated as untrusted data, never as instructions the AI should follow. Any endpoint you configure is checked so it can't be pointed at an internal address it shouldn't reach. And every tool call and every access decision is written to an audit log you can review.
Do you train AI on our data?
No. We use frontier models through their business APIs, where your inputs are never used for training. Your data answers your questions and does nothing else.
Can it show someone something they shouldn't see?
Access is enforced in the data layer, by role — not left to the model's discretion. A question outside your role returns the in-scope parts and names what was withheld, rather than quietly leaking it.
Where does our data live?
In your own isolated tenant on AWS, in the United States, encrypted in transit and at rest. Connection secrets are held in a dedicated vault, separate from other customers.
Who is behind it?
Azumo, a SOC 2-certified software engineering firm with ten years in production and clients including Discovery, Zynga, and Meta. The same engineers who build it do your setup.
How do we know an answer is right?
Every answer carries its sources, drawn from the data itself, so you can check any number at the point it's made. Where two systems disagree, the answer shows the seam instead of averaging it into a fake clean figure.