Concept: The Context Inversion Protocol

Concept

The fundamental proposition of the Context Inversion Protocol is a structural inversion of the client-server relationship as it pertains to personal context. Current frontier artificial intelligence models rely on a centralized architecture, wherein user data—preferences, relational logic, and personal history—is extracted from the edge device and aggregated within a multi-tenant cloud environment.

The Centralized Extraction Model

In the dominant paradigm, the edge device acts merely as an input terminal. Relational reasoning occurs exclusively at the center, requiring the persistent upload of high-fidelity personal context to third-party data centers. This creates inherent vulnerabilities regarding data sovereignty, privacy, and systemic fragility.

The Context Inversion Protocol

The Context Inversion Protocol (CIP), authored by Robert Potter and published by the Sovereign OS Project, is a proposed edge-first implementation profile for user-owned AI context. Under that architecture, the locus of memory and authorization would remain at the edge. The user's device would keep an inspectable local custody plane. External agents and foundation models would be treated as stateless, replaceable utilities, routed to narrowly authorized context rather than extracting the store to the cloud. CIP would later use Human Context Protocol semantics and Model Context Protocol transport for scoped sharing; it would not replace either protocol, and it is not yet shipped.

By maintaining a sovereign boundary around the user's relational logic, the architecture is designed to keep the foundational "engine of intelligence" serving the individual independently of any single vendor or centralized infrastructure.

The published custody plane

Sovereign OS (SOS) Control Plane 1.6.0 is published under the MIT license as that local store: captured sources, deterministic evidence, and operator-accepted knowledge, kept as ordinary files. It implements a capture, ingest, debrief, and retrieve loop. It is not CIP, not an HCP implementation, and it does not expose MCP. CIP would add a mediated runtime in front of a plane like SOS so replaceable models never mount the store. Those three objects remain distinct without HCP. HCP is the protocol-family vocabulary for later scoped sharing; it is not required for the custody plane to exist.

The software is available at github.com/robertpotter-dev/sos-control-plane. Independent verification of behavior remains future work. The CIP gate is not shipped.

The Human Context Protocol research program

The Human Context Protocol (HCP) is a research program introduced by researchers spanning MIT, Stanford, Oxford, and Microsoft Research, and further formalized in the March 2026 Global Memory Workshop (Stanford HAI, the Gates Foundation, AWS, and the Mozilla Foundation). HCP proposes a standardized, user-owned preference layer that grants individuals granular control over how their data steers AI systems. Industry momentum around such interoperability still risks routing localized context back to multi-tenant corporate data centers—a regression to mainframe topology. CIP's proposed countermeasure is to keep the store local and to place a gate in front of it. SOS is that store, not an instantiation of HCP at the edge.

The HCP reference demonstration

To illustrate HCP's own mechanics, the original HCP authors (maintained by Tobin South as a Stanford Digital Economy Lab and Consumer Reports collaboration) published a web-based demonstration. That demonstration is distinct from and unaffiliated with the Sovereign OS Project. It stages preference generation and context routing as a conceptual sandbox for HCP. It is not the SOS control plane, and SOS is not designed as an HCP-MCP stack.

HCP Interactive Demonstration (by Tobin South et al.):

Access the HCP Preference Generation Sandbox →

Proceed to the FAQ →