VisionFI · Scout ILM

Scout Data Sovereign Architecture

Client-hosted processing with VisionFI intelligence. Your data stays in your environment. Our expertise arrives per transaction.

Overview for Partners & Clients · March 2026

Your Data Stays in Your Environment

VisionFI has architected the Scout platform into two distinct layers that can operate across separate environments. This separation is the foundation of our data sovereign offering.

Scout Orchestration VisionFI-Hosted
The intelligence layer. Delivers domain expertise, regulatory logic, and continuously updated ILM instructions to Scout Fieldwork on a per-transaction basis. Has no access to your document content.
Per-transaction intelligence instructions
Scout Fieldwork Your Environment
The data processing layer. Handles document ingestion, model inference, and all interaction with your data. Runs inside your own Azure tenant or directly on your workstation. Your documents, borrower information, and financial data never leave your environment.

When your team processes a loan package, commercial appraisal, or compliance review through Scout, your data is processed entirely within your infrastructure. VisionFI provides the intelligence that tells Scout what to look for and how to evaluate it, but never sees the documents themselves.

Inference Sovereignty: The Gap Some AI Solutions Do Not Close

Financial institutions are well-acquainted with data sovereignty: knowing where data physically lives, what jurisdiction governs it, and that institutional policies apply to it at rest.

But AI introduces a gap in that framework. An institution can achieve perfect data sovereignty and still have sensitive documents leaving the boundary at the moment it matters most: when AI is reading and reasoning over them.

We call this gap inference sovereignty: the guarantee that not only does your data reside in your environment, but the AI processing of that data also happens within your environment, under your controls, in your region.

Scout's data sovereign architecture delivers both. Your data never leaves home, even while it is being worked on.

Cloud Data Sovereign vs. Desktop Data Sovereign

The data sovereign architecture ships in two deployment models. Both deliver identical Scout intelligence and identical data sovereignty guarantees. The choice between them is operational.

Cloud Data Sovereign

Scout Fieldwork runs as containers within your Azure tenant: AKS or ACI clusters, virtual networks, container registries, and the associated security and scaling policies. Your Azure subscription also provides the foundation model endpoints (Azure OpenAI Service or equivalent) and blob storage.

Best for: Institutions needing API-level integrations, batch processing pipelines, connections to partner and B2B systems, or horizontal scaling across multiple concurrent workloads.

💻 Desktop Data Sovereign

Scout Fieldwork runs directly inside Scout Notebook on the user's workstation. Documents are ingested and processed locally on a machine the institution already owns, manages, and secures. The institution's Azure tenant provides only two services: model inference endpoints and blob storage. Containers, container orchestration, virtual networking, and the associated management overhead are eliminated entirely.

Best for: Interactive analyst use through Scout Notebook. Strongest inference sovereignty, lightest Azure administration burden. Not suitable for headless API access or high-volume batch workloads.

Both deployment models deliver the same Scout intelligence, the same ILM domain logic, and the same data sovereignty guarantee. The choice between them is operational, not qualitative.

Think of It Like Your iPhone

Your iPhone stores your photos, messages, and personal data locally on a device you own and control. But the intelligence that makes the camera recognize faces, the keyboard predict your next word, and Siri understand your questions comes from Apple, delivered continuously, updated invisibly, and never requiring you to install a new operating system to get smarter features.

Your iPhone

Your data stays on your device. Apple's intelligence arrives on demand. You never manage AI version upgrades.

Scout Data Sovereign

Your data stays in your Azure environment. VisionFI's intelligence arrives per transaction. Your IT team never manages ILM version upgrades.

Why This Matters

🔒

Data Sovereignty + Inference Sovereignty

Most AI vendors can say they don't store your data. They cannot say the inference itself happened inside your environment. Scout can. Model calls, document reasoning, field extraction: all within infrastructure you own and control.

Intelligence That Stays Current

When VisionFI updates a model for new regulatory guidance, improved covenant extraction, or enhanced appraisal analysis, every client receives those improvements immediately. No version upgrades. No deployment cycles.

Board & Examiner Confidence

Document content is processed within your institution's own cloud environment under your existing security controls, access policies, and audit frameworks. A clear, auditable story for regulators and board members.

🔍

Independently Verifiable

This is an architectural fact your own Azure networking tools can independently verify. No trust-us claims required.

What VisionFI Invests to Make This Work

Giving institutions the flexibility to run Scout Fieldwork in their own environment is not simply a matter of relocating a container. It requires significant, ongoing engineering investment that ultimately benefits every client on the platform.

Multi-Model Intelligence Engineering

Scout Fieldwork may run against Azure OpenAI Service, Google Vertex AI, or other foundation models depending on the institution's configuration. Each model family has different strengths, latency profiles, and behavioral nuances. VisionFI also develops proprietary small models for specific financial document tasks. Our engineering team continuously tests, tunes, and validates this full model stack so your results are consistently accurate regardless of which infrastructure powers your instance.

Orchestration Infrastructure

Scout Orchestration is a managed intelligence platform handling real-time instruction assembly, transaction metering, performance telemetry, encrypted payload delivery, and versioned rollout across the entire client base. Within a given transaction, it manages a structured task workflow that breaks complex document analysis into discrete steps, coordinating which model handles which step, in what sequence, and how partial results feed downstream analysis.

Continuous Accuracy Investment

VisionFI is not a hosting company. We are an applied AI research and engineering firm. The core of what clients pay for is the continuous improvement of document intelligence: the domain ontology, the validation logic, the lexical variation research, and the regulatory expertise that make Scout meaningfully better than general-purpose AI tools at understanding financial documents.

The data sovereign model asks more of VisionFI's engineering, not less. Institutions gain data sovereignty and deployment flexibility. VisionFI invests in multi-model optimization, orchestration infrastructure, and continuous accuracy improvement so the intelligence layer keeps getting smarter for everyone.

How Metering Works: The VisionFI Token

VisionFI is evaluating a consumption model that aligns with how leading enterprise platforms already operate. The VisionFI Token is a simple, predictable unit measuring intelligence consumption per action.

A commercial loan audit consumes a defined number of VisionFI Tokens. A stipulation review consumes fewer. The rate is fixed and published, regardless of document length, complexity, or which compute resources your environment uses.

The Databricks Parallel

Databricks operates on a virtually identical data sovereign architecture: their control plane stays in Databricks' own cloud account while the customer's data and compute run in the customer's own Azure subscription. Customers pay Databricks for DBUs (Databricks Units) and pay Microsoft separately for the underlying compute.

Dimension Databricks Model VisionFI Model
Intelligence layer Databricks control plane Scout Orchestration (VisionFI-hosted)
Processing layer Customer's Azure subscription Scout Fieldwork (your Azure tenant)
Metering unit DBU (Databricks Unit) VisionFI Token
You pay the vendor for Platform intelligence Document intelligence
You pay Azure for Compute & storage Container compute & model inference
Your data visible to vendor? Metadata visible; requires additional PII config Architecturally impossible: no content crosses the boundary

The Max Plan: Fixed-Cost Simplicity

For institutions that prefer total budget predictability, VisionFI is also evaluating a Max Plan: a single flat monthly price per Scout ILM model. No per-action tracking, no token balances to manage. An institution using Scout for Commercial Lending and Audit would have fixed monthly line items that finance teams can plan around.

The data sovereign architecture makes a fixed-cost plan especially viable because the institution absorbs its own Azure compute costs for model inference. VisionFI's per-transaction cost is the lightweight intelligence delivery, not the heavy compute. Appropriate fair-use guardrails would be in place: velocity limits, generous monthly ceilings with notification thresholds, and contractual review rights for anomalous usage patterns.

Both consumption and Max Plan options would be available regardless of deployment model. Same intelligence, same flexibility. The institution chooses what fits their budget philosophy.

Performance Intelligence Without Seeing Your Data

If VisionFI never sees your documents, how does the platform continue to improve? Through a structured telemetry framework designed to be PII-free by architecture, not just by policy.

Operational Telemetry Default
Scout Fieldwork sends structured performance metadata: processing duration, confidence scores, field completion rates, and exception categories. This data contains no document content, no borrower names, no loan amounts. There is structurally no field in the telemetry payload where PII could go. For example, VisionFI might learn: "14 of 16 expected covenant fields were identified with an average confidence of 93%." We would not see what those covenants said or who the borrower was.
Outcome Feedback Opt-In
When a human reviewer accepts or overrides a Scout finding, that accept/reject signal is valuable for improving accuracy. Participating institutions share only structured outcome data: a finding ID and whether it was accepted, rejected, or categorized as incomplete. No document content included. Participating institutions benefit directly from faster, more targeted intelligence improvements.
Diagnostic Analysis By Engagement
For specific performance investigations, a clean-room diagnostic process is available. De-identified document segments can be analyzed in a jointly auditable environment with defined retention windows and automatic deletion. This is never automatic. Your institution controls when, what, and whether to participate.
The principle across all tiers: VisionFI gets better for every client without ever seeing any client's data. Institutions that opt in to deeper feedback accelerate improvements for their specific document patterns, always on their terms, always PII-free.