# claw.refinio.one > OpenCLAW is one-click live mode for a VGER AI agent. Canonical URL: https://claw.refinio.one Refinio downloads: https://refinio.one/downloads ## Product principle In vger.one, an AI identity can act in OpenCLAW and the world without losing control of it. OpenCLAW launches a VGER AI person from chat into a governed, inspectable child agent embodiment across desktop, local server, browser, and mobile-connected environments while VGER keeps identity, trust, memory ownership, authority, audit, and revocation records. ## Summary OpenCLAW is the governed runtime layer for vger.one. It admits live agent runtimes through VGER-owned trust envelopes, keeps identity and memory in the VGER graph, checks plan calls before execution, and gives users status, inspect, open, and stop controls from vger.one. The first product slice is a local vger.one launch path with a parent runtime worker and optional child UI runtime. The runtime can use local or configured model endpoints, but model substrate is interchangeable: identity and memory remain VGER-owned. OpenCLAW is defined by runtime admission via trust envelope, not by execution mechanics. Missing trust data is a hard failure. UI visibility does not grant authority. Every action into VGER must pass OpenCLAW admission checks, VGER authority checks, and ONE access boundaries where object visibility is involved. ## Core requirements - One-click launch from vger.one settings. - AI person identity remains in VGER across runtime starts, stops, and model changes. - Every runtime is admitted through a VGER-owned trust envelope. - Runtime power is governed by admission, authority, budget, journal, and ONE access state. - Plan calls are checked before reaching VGER operation surfaces. - Runtime UI is visible and controllable without granting additional authority. - Users can inspect, open, and stop a running OpenCLAW runtime. - Developers can integrate runtimes through admission payloads, a runtime/action admission model, the plan-call bridge, and control surface metadata. - The same model applies across desktop, server, browser, and mobile-connected environments. ## Non-goals - Agent framework replacement. - Generic local model runner. - Permissionless plugin host. - Cloud-required execution platform. - Memory stored inside model weights.