CIO Architecture View

Governed intelligence
built for enterprise control.

Vela turns large volumes of unstructured information into governed, traceable, decision-ready intelligence. It is designed for organisations that need AI to improve judgment, speed, and consistency without losing control, security, or auditability.

Primary outcome
Decision-ready intelligence
Structured assessments aligned to a business process, not raw model output.
Operating model
Structured deep research
Evidence extracted, assessed against rubrics, and verified before reaching users.
Control principle
Verification at each stage
Outputs evaluated rather than blindly generated, reducing operational AI risk.
Security posture
Identity-separated architecture
Five distinct runtime identities with narrow, purpose-built authority.

What a CIO needs to understand.

The platform is designed around five executive concerns: business usability, architectural separation, security by design, operational resilience, and explainable AI outputs. The goal is not simply to retrieve content, but to produce reliable, reusable intelligence that can stand inside enterprise workflows.

Business value
From content overload to usable decisions

Documents, spreadsheets, decks, policies, and research are converted into structured assessments and reports aligned to a business process.

Architecture
Clean separation of concerns

User applications, orchestration, storage, AI services, and cross-cutting controls are separated so each layer can evolve without compromising governance.

Controls
Verification instead of blind generation

Vela evaluates evidence, applies rubrics, and verifies outputs, reducing the operational risk of unsupported AI responses.

Trust
Auditability built into the flow

Usage, actions, and evidence lineage are captured so outputs can be reviewed, defended, and improved over time.

Principles.

Vela Intelligence is a governed intelligence platform. For a CIO, the result is a platform that is easier to reason about: clearer control boundaries, stronger auditability, lower privilege ambiguity, and a better fit for high-consequence enterprise decisions.

01

Architecture is intentionally separated.

Business applications, decision workflows, intelligence services, enterprise controls, and the secure data foundation operate as distinct layers, not a monolith.

02

Security extends to the database layer.

Runtime identity separation, row-level enforcement, and narrowly scoped privileged operations are not application-layer assumptions — they are database-layer facts.

03

Outputs are verified, not generated.

Every intelligence output passes through evaluation and verification steps before reaching users. This is a structural property of the platform, not an optional feature.

04

Humans remain accountable.

Vela improves the quality of evidence available to decision-makers. It does not replace human judgment in high-consequence processes.

Business architecture.

At a business level, Vela can be understood as five layers. This is the view that matters for governance, integration, and scale.

01
User experience
  • Analyst and operator workspaces
  • Administrative and support consoles
  • Conversational interfaces through approved channels
02
Decision workflows
  • Structured evaluations
  • Evidence review
  • Multi-document synthesis
  • Report composition
03
Intelligence services
  • Parsing and ingestion
  • Retrieval and search
  • Assessment engines
  • Model orchestration
04
Enterprise controls
  • Authentication and access control
  • Verification and policy enforcement
  • Audit trail and usage metering
  • Operational monitoring
05
Core data foundation
  • Document store with security boundaries
  • File storage
  • Vector and search infrastructure
  • Embedding and indexing services

Security and governance model.

The platform's security model is deliberately stricter than a conventional application stack. Access is enforced by identity separation and collection-level controls, not by application-side convenience logic.

When McKinsey was compromised in a widely reported breach, the most damaging consequence was not the initial intrusion — it was what happened next. Without architectural separation between control operations and data access, attackers were able to move laterally across the data estate with minimal friction. A single point of compromise granted broad authority across client data because no structural boundary existed to contain it. For any platform handling sensitive business intelligence, the lesson is direct: if the control plane and the data plane share identity and authority, a breach of one is effectively a breach of both. Vela's architecture is built around the opposite assumption.

Core principle
Identity-separated runtime model

Vela uses one bootstrap-only owner identity and four distinct runtime identities. Each operational context connects with its own authenticated identity, rather than relying on broad privileges plus role switching. Control-plane operations, document administration, scoped user access, and system-only access are separated by design.

Why it matters
Smaller blast radius, clearer accountability

Each identity can do only the work it was created for. That reduces accidental privilege creep, makes access review easier, and strengthens auditability for CIO, CISO, and platform governance teams.

Control plane identity

Manages collections, indexes, settings, and platform control operations. Does not directly access business documents.

Document administration identity

Handles unrestricted document work for approved non-system data sets, including ingestion and maintenance tasks, segregated from control-plane administration.

Scoped runtime identity

Used for least-privilege, collection-scoped document access. Permissions narrow access to exactly what is needed; they do not widen it.

System runtime identity

Reserved for platform-controlled system records only. Intentionally isolated from ordinary business collections.

What the security model achieves.

The McKinsey breach also demonstrated a second failure mode: once attackers held valid credentials, they could impersonate legitimate users and access any data those users could reach. Without collection-level boundaries, a compromised identity becomes a master key. Vela's model addresses this directly — permissions are stored in VelaAuth, an isolated and hardened permission store, and access is scoped to collections rather than granted broadly. A compromised credential can only see what that identity was explicitly permitted to see.

VelaAuth-enforced authority
Permissions live in VelaAuth, not in application memory

The application does not carry its own durable permission state. Access checks are enforced through VelaAuth — a secure, isolated permission store — reducing the risk of drift between business logic and actual authority.

Collection-level enforcement
Collection-level visibility with hard boundaries

Access boundaries are enforced at the collection level across documents, keys, history, and vector chunks. The system collection remains separate and cannot be exposed through ordinary scoped access.

Privileged operations
Narrow helper functions instead of broad grants

Privileged database actions are exposed through tightly scoped, owner-controlled functions. This keeps sensitive operations narrow, inspectable, and operationally safer.

Operational design for enterprise environments.

Resilience, traceability, provider independence, and deployment flexibility — designed for organisations where uptime, observability, and infrastructure choice all matter.

Fail-fast authentication and authorisation

Every request is authenticated and authorised before any data access occurs. An unverified or insufficient identity is rejected immediately — not degraded gracefully. Attackers cannot get deep enough to probe collections, test boundaries, or exploit partial access states.

Traceability across the lifecycle

Audit and usage controls sit beside the core platform so organisations can monitor who did what, when, and at what operational cost. Every intelligence output, evidence retrieval, and user action is logged — giving CIO and CISO teams a complete, inspectable record for governance, review, and incident response.

No AI vendor lock-in

Vela's AI layer is provider-agnostic. The platform works with major AI providers interchangeably, so organisations are not committed to a single vendor's roadmap, pricing, or data terms. Model choice can be adjusted over time without redesigning the intelligence layer above it.

Flexible deployment model

Vela runs where your data governance requires. On-premises in your own infrastructure, in a private cloud, as a fully isolated managed instance, or as a shared environment for less sensitive workloads. The deployment model is a configuration choice, not an architectural constraint.

Request a CIO or security briefing.

We work with enterprise technology and governance teams who need a detailed view of the platform architecture, security model, and deployment options.