Platform
Business Control Plane
Turn regulations into reviewable, auditable decisions.
Run it standalone or on top of VetoPlane. New legal or compliance domains ship as configuration, not new code.
In early access through our design partner program.
Architecture
Eight layers. One governance plane.
Every instantiation follows the same eight sequential layers. Requests pass through identity and policy checks, model routing, evidence retrieval and independent review before a recommendation is shown. Every layer emits structured events into the cross-cutting governance plane.
Layer 1: End-User Consumption
Authorized entry, within the workflows your teams already use.Four entry-surface classes feed the control plane: an embedded in-workflow widget, a web portal, a domain workflow assistant, and a mobile/field app. Every request originates from an authorized user-facing surface, never a public or unauthenticated one.
Layer 2: Ingress, Identity & Context Tagging
Resolve identity and context before anything touches a model.SSO authentication resolves the user's role. Each request is tagged with a workflow_id, tenant ID, and sensitive-context flag. Rate limiting and a data-boundary check happen here, before anything touches a model.
Layer 3: Regulatory & Compliance Policy Gate (pluggable)
Apply deployment-specific controls. A failed control means hold, not proceed.Every request is classified against the four-control rule set configured for its deployment. Failing any single control routes the request to a hold queue instead of allowing it to continue. The rule set is pluggable, so each domain can supply its own controls.
Layer 4: Model-Agnostic Inference Router
Match the task to validated model capability, cost and latency.Cleared requests pass through a task-to-capability matcher, a cost/latency optimizer, a token-budget enforcer, and a fallback/circuit-breaker chain. Provider options include AWS Bedrock, Vertex AI, Azure OpenAI, the Anthropic API directly, or an on-prem vLLM cluster for open-weight, data-resident deployments. Routing selects from the deployment's validated providers rather than locking the architecture to one vendor.
Layer 5: Cost & Infrastructure Cost-Visibility
Make spend visible by tenant, workflow and model.Per-request token metering, spend attribution by tenant, workflow and model, real-time budget alerts, and cost-center chargeback make the managed-API-versus-self-hosted decision data-driven.
Layer 6: Retrieval & Evidence Grounding
Ground the review in approved sources, with a traceable basis.Retrieval-augmented generation uses approved sources only. A citation/provenance builder and a basis-bundle assembler package the inputs, a plain-language logic summary, and known limitations for independent review.
Layer 7: Human-in-the-Loop Independent Review Module
Make independent review part of the workflow, not just a policy statement.Five sub-components support independent review: an evidence/rationale display, a confidence and data-limitation panel, a time-criticality gate that blocks automatic action on time-critical or high-materiality cases, a view-before-decide acknowledgment capture, and an accept/modify/override logger.
Layer 8: Response Delivery & Outcome Capture
Return the final decision and capture what happened next.The reviewer's final decision returns to the source system or workflow. The decision and downstream outcome are logged as ground truth for drift and efficacy monitoring, supporting the NIST MEASURE function.
Module
Regulatory Review Engine
Checks documents and actions against your rules and flags gaps. The Regulatory Review Engine is one module within the Business Control Plane, not a separate platform.
Domain-specific controls and approved evidence sources supply the configuration. The architecture, review path and governance plane stay consistent.
Explore the execution layer: VetoPlaneBuild on it with us
Bring your domain, review requirements and workflow to our design partner program.