Open Semantic Protocol

As Large Language Models (LLMs) become the primary interface for informationretrieval, the "index" of the web is increasingly centralized within proprietary vectordatabases. For low-resource languages and dialects, this centralization createsa "representational collapse," where cultural nuance is lost during embeddinginto English-centric latent spaces. In this paper, we propose the Open SemanticProtocol (OSP), a decentralized architecture that restores informational sovereigntyto content creators. OSP introduces a three-layer stack: (1) Semantic DNS fordiscovering embedding data via standard TXT records; (2) Content-AddressableVectors for immutable, self-hosted storage; and (3) Just-in-Time (JIT) SemanticProjection, a client-side mechanism that aligns disparate embedding models usinglightweight linear transformation matrices. We experimentally demonstrate thata simple linear projector can bridge incompatible vector spaces (e.g., 384-dim to768-dim) with 99.0% retrieval accuracy, even generalizing to unseen domains(99.4% on AG News). This confirms that a decentralized, "Bring Your Own Model"search architecture is mathematically feasible and scalable.