Technical Whitepaper
Architecture of Trust: Enabling Native Legal AI inside Microsoft 365
An in-depth review of the Resident Agent model, data sovereignty, and zero-storage infrastructure.
Executive Summary
Arivu.Legal introduces a paradigm shift in legal technology: the Resident Agent. Unlike traditional “Cloud-to-Cloud” connectors that require data egress, Arivu functions as a native intelligence layer within your organization’s Microsoft 365 tenant. This document outlines how our architecture maintains absolute data sovereignty while delivering generative AI insights.
1. Zero-Storage Infrastructure
Arivu is a stateless system. We do not build shadow databases or secondary indices of firm Intellectual Property.
- No External Indexing: We leverage your existing SharePoint Online Search Index via the Microsoft Graph API.
- Ephemeral Processing: Analysis occurs in volatile RAM. No text chunks or embeddings are persisted on Arivu-controlled disks.
2. The Secure Data Loop
To ensure confidentiality, the flow of data is restricted to a non-persistent loop:
01. Request: User initiates a query in Word, Outlook, or SharePoint.
02. Discovery: The Resident Agent identifies files using the user’s delegated Entra ID token.
03. Inference: Context is passed ephemerally to a Private Azure OpenAI instance.
04. Delivery: Insights are returned; volatile memory is purged.
Technical Comparison
| Feature | Traditional AI | Arivu (Native) |
|---|---|---|
| Data Boundary | External Cloud | In-Tenant |
| Identity | Service Account | Delegated User |
| Persistence | Shadow Index | Ephemeral/Volatile |
CISO Rapid-Fire FAQ
Is data used for LLM training?
No. Our Private Azure OpenAI instances strictly prohibit the use of customer data for training purposes.
How are ethical walls enforced?
Arivu inherits your Microsoft Purview and SharePoint permissions. If a user cannot see a file, the Agent cannot see it.
Where is data processed?
Processing occurs within localized Azure regions (UK South, EU West, or US East) based on your firm’s residency requirements.