Our principles
- Least privilege. Every connection gets the minimum access needed for the use case. Read-only unless the scope requires writing back.
- Your data stays yours where possible. Data stays in your own tenant and accounts. We work inside your environment rather than copying data into ours.
- Humans approve what matters. AI drafts, recommends and flags. People approve anything that goes to a customer, a regulator, a GC, or a ledger.
- We describe our practices accurately. We list what we are certified for. If we're unsure, we'll say so.
Access scoping happens before any connection
Before we connect to Vista, NetSuite, Procore, Microsoft 365 or any other system, we write down exactly what the AI can see and do.
Step 1: Inventory the systems in scope
We list each system, the data it holds, and who administers it on your side.
Step 2: Define allowed and denied actions
For each system: which records can be read, which can be written, and which are off limits entirely, such as payroll or HR records unless the use case requires them.
Step 3: Create scoped credentials
Your administrator issues service accounts or API keys with only those permissions. We don't ask for admin credentials to get started.
Step 4: Review and sign off
Your system owner signs off on the access scope before the first connection. Changes to access go through the same review.
Where your data lives
Our default is to build inside your environment: your Microsoft 365 or Google Workspace tenant, your cloud account, your ERP. Documents and records stay where they already are. The AI reads what it's permitted to read at the time it needs it.
When a hosted model is used, prompts and the data included in them are sent to that model provider under the provider's commercial terms. We tell you which provider, which plan, and what those terms say before anything is sent.
Model options: hosted enterprise or self-hosted
Hosted enterprise plans
For most business workflows, we use enterprise AI plans. They give you current model capability with administrative controls and per-role user profiles, and under Anthropic's commercial terms they don't train models on your business data.
Self-hosted and open-weight models
When data can't go to a hosted model at all, such as some government, federal or regulated work, we deploy open-weight models on infrastructure you control, on-premise or in your own cloud. Capability differs from the largest hosted models, and we're specific about those trade-offs for your use case.
Human review is designed in, not added later
We define review points during scoping. A compliance record, a Declaration of Conformity, a client-facing email, an entry posted to the ledger: each has a named approver and a step where a person confirms before the output is used. The AI does the reading, drafting and cross-checking. People sign.
Logging and audit trail
The systems we build keep a record of what the AI did, so you can answer "where did this come from?" after the fact.
- Which user or process triggered an action, and when.
- Which source records the AI read to produce an output.
- What the AI produced, and who approved, edited or rejected it.
- Changes to prompts, configuration and access scope.
What we don't claim
Plainly, so you don't have to ask:
- We do not hold a SOC 2 report.
- We are not HIPAA certified, GxP certified or FedRAMP authorized, and we don't claim to be.
- For life science work, we design for GxP-aware validation and audit trails. Your QA team owns validation.
- We don't guarantee AI outputs are correct. That's why human review points are part of every build.
- We can't change a model provider's terms. We tell you what they are so you can decide.
Security FAQ
Will our data be used to train AI models?
Enterprise plans from Anthropic don't train models on your business data under their commercial terms. For data that can't go to a hosted provider at all, we use a self-hosted model in your environment.
Do you need admin access to our systems?
No. We ask your administrator to create scoped service accounts with only the permissions agreed in the access scope.
Can the AI write to our ERP or accounting system?
Only if the scope says so, and usually through a review step. Many first builds are read-only.
Where is our data stored?
In your own tenant and accounts wherever possible. If a project requires storing data elsewhere, we'll identify where, why and for how long in the SOW.
Can you work with our IT provider or security team?
Yes. We'd rather have them involved in the access review from the start.
Can you fill out our security questionnaire?
Yes. We'll answer it accurately, including the items where the answer is no.