Towards authorization for purpose-based Agent search in Solid

Ruben Dedecker

Keywords: ODRL, DPV, Purpose, Solid

TL;DR: Extended abstract on moving towards agent search over Solid Pods through querying approaches and indexing dimensions

Abstract: AI agents are increasingly used to answer questions over heterogeneous data spaces. Solid provides a Web-native, resource-centric architecture with server-enforced authentication and authorization, making it a natural foundation for agent-accessible personal data storage. However, effective question answering over Solid pods requires performant semantic search, yet current Solid discovery mechanisms are insufficient for the retrieval demands of LLM-based agents. We analyze three indexing and retrieval approaches, in client-side link traversal, resource-specific index artifacts, and server-side full indexing — evaluating their trade-offs across authorization fidelity, semantic expressiveness, and cross-resource synthesis. We argue that purpose-based authorization, grounded in ODRL and the W3C Data Privacy Vocabulary (DPV), enables powerful cross-resource embeddings that serve both the pod owner and authorized third-party agents through verifiable, purpose-scoped semantic search.

Submission Number: 15