Mathematically Governed Control Plane
Universal Control Plane (UCP)
One app on your device. Every AI connection governed from one place.
What it does
A desktop app that governs every AI connection on your device.
UCP runs locally on your machine and connects your AI clients, agents, MCP servers, models, and tools to one governed control plane. Instead of scattered, ungoverned connections — one place to review, admit, and enforce.
AI agents
Claude Desktop, Codex, Cursor, Cline, ChatGPT Desktop — every AI client connects through UCP. One governed control plane instead of scattered, ungoverned connections.
MCP servers
External MCP servers register through UCP. Every request passes through the tool router for enforcement. Nothing connects without governance.
Models
The model gateway routes inference through governance gates — tier-based, provider-scoped, policy-bounded. Which models are allowed, where they connect, and what they can do.
Tools and actions
Tool dispatch enforces governance predicates before execution. Sensitive tools are privileged-gate protected. Every action is admitted or rejected before it runs.
Rules and policies
34 runtime constraints enforce action admission across governance, process, quality, and tier gating. Policies are not suggestions — they are enforcement gates.
Marketplace packages
Governed package intake lifecycle. Packs extend UCP with new capabilities — each one admitted through the same governance discipline as everything else.
The architecture
Decided centrally. Enforced locally.
In enterprise deployments, governance policy is defined at SMARTHAUS Operation Center and synced to every device. But enforcement and execution happen locally — on your hardware, with your data, under your control. Workloads never leave the device. When the cloud is unavailable, the device keeps running under the last valid local state.
UCP Studio
Standalone desktop app for individuals. One user, one device, no cloud dependency. Loge is the free community tier — a real, usable product floor. Forge is paid with advanced local capabilities.
UCP Enterprise Client
Managed local endpoint that syncs governance from SMARTHAUS Operation Center. Policies, rules, and allowlists are decided centrally — then enforced locally on the device. When the cloud is unavailable, the device continues operating under the last valid local state.
SMARTHAUS Operation Center
The enterprise authority plane. Centralizes policy definitions, tenant enrollment, and approval workflows. Coordinates the fleet — but workloads never leave the device. Control plane only. Execution stays local.
What UCP is not
Not an MCP server. Not a wrapper. Not a guardrail.
UCP vs MCP servers
MCP servers are tools. UCP governs them.
An MCP server exposes capabilities — file access, database queries, API calls. UCP is the control plane that sits above every MCP server on your device. It rewrites your AI client configs to route through its broker, so every tool request passes through governance before it executes. MCP servers provide tools. UCP decides which tools are allowed to run.
UCP vs AI gateways
Gateways are cloud proxies. UCP runs on your device.
AI gateways and orchestration platforms sit in the cloud between you and model providers — proxying API calls and adding logging. UCP runs locally on your machine. It governs the full stack — agents, tools, models, and policies — not just model API calls. Your data never leaves your device. When the cloud is unavailable, UCP keeps enforcing.
UCP vs guardrails
Guardrails validate after. UCP enforces before.
Guardrail tools check LLM output after it has been generated — flagging, filtering, or blocking responses after the fact. UCP enforces governance before execution. 34 runtime constraints evaluate every proposed action through predicate algebra with fail-closed semantics. Actions are admitted or rejected before they run — not reviewed after they have already happened.
Built for
Anyone connecting AI to real work.
Individuals who want governed AI on their own device — and enterprises that need to govern AI across a fleet. Same app. Same proof discipline. Studio for one. Enterprise Client for many. Operation Center to coordinate.