X-Hub-SystemGoverned Agent control plane.

X-Hub-System centralizes model routing, memory truth, skills, provider accounts, quotas, grants, policy, audit, and terminal execution under one user-owned Hub.

Read the architecture

See what it governs

Browse docs

Hub-first trust root

The terminal can execute, but the Hub owns route truth, grants, policy, memory truth, audit, and kill authority.

Governed autonomy

A-Tier, S-Tier, heartbeat, review, grants, and runtime clamps make autonomy visible instead of vague.

One model and capability plane

Local models, paid providers, skills, channels, quotas, and fallback truth converge through one control boundary.

Public technical preview X-Hub-System is already runnable, but still in active productization. This site focuses on the architecture, the governed capability surface, and the reading path for people evaluating the system.

Why it exists

Most agents get powerful by putting too much trust in one runtime.

X-Hub-System moves the trust anchor out of terminals, plugins, browser context, and vendor defaults. Clients can still be useful execution surfaces, but authority to route, grant, deny, remember, audit, and stop execution stays in the Hub.

Client asksX-Terminal / generic clients / channels

Hub decidespolicy, grants, memory, quotas, route truth

Surface actslocal models, paid APIs, tools, skills, connectors

Truth returnsaudit, evidence, fallback, deny reasons, quota state

What it solves

Built for execution range without trust sprawl.

The difference is not one feature. It is where the trust root lives and how every higher-risk surface is forced back through a governed boundary.

Common agent-stack failure mode

X-Hub-System design response

Prompts, tools, memory, secrets, and execution collapse into one runtime trust zone.

The Hub owns trust, grants, route truth, memory truth, policy, audit, and kill authority.

Plugin installation silently expands privilege.

Skills use manifests, trust roots, pins, preflight checks, grants, deny codes, revocation, and audit.

Local models and paid APIs drift into separate governance paths.

Model routing, provider accounts, OAuth/key state, quotas, fallback, and downgrade truth converge in one plane.

Remote channels become shadow control planes.

Slack, Telegram, Feishu, voice, and mobile-style ingress converge through authz, replay guard, grants, and audit.

Auto mode hides risk and weakens supervision.

A-Tier, S-Tier, heartbeat, review, grants, runtime clamps, and kill switches keep autonomy governable.

Governed capability surface

One Hub authority above many AI execution surfaces.

X-Hub-System is designed to govern a growing set of model, memory, skill, quota, terminal, channel, and evidence surfaces without making each one a new control plane.

TrustHub-owned control plane

Identity, pairing, policy, grants, readiness, and kill-switch posture stay centralized.

ModelsLocal + paid routing

Configured route, actual route, fallback, downgrade, and provider readiness stay visible.

SkillsGoverned skill packages

Official skills, manifests, trust roots, pins, preflight gates, grants, and revocation.

AutonomySupervisor-grade controls

Execution authority, review depth, heartbeat cadence, guidance, and intervention are separate controls.

IngressChannels and voice

Remote operator surfaces can enter through replay guard, challenge, grant, and audit paths.

EvidenceRuntime truth

Audit refs, evidence refs, denial reasons, quota pressure, and recovery signals remain operator-visible.

System shape

Two views: the authority boundary and the governed capability map.

These diagrams are the fastest way to see what X-Hub-System is trying to make governable.

Trust and control plane Clients can ask. The Hub decides. Execution surfaces act only after governance, and runtime truth returns to the Hub.

Governed capability map Models, memory, skills, quotas, terminals, channels, Supervisor state, and runtime evidence converge through one authority boundary.

Reading path

Evaluate the system from architecture to operator surface.

Architecture Trust model Why not just an agent?Governed autonomy Governed skills Documentation map