The increasing volume, heterogeneity, sensitivity, and distributed ownership of enterprise information within Microsoft 365 environments have created significant challenges for data governance, secure information retrieval, provenance verification, policy enforcement, and evidence-based organizational decision support. Conventional retrieval-augmented generation systems primarily optimize semantic relevance between user queries and indexed documents but frequently provide limited support for identity-aware retrieval, authorization inheritance, contextual governance policies, cross-application reasoning, document provenance, risk-sensitive response generation, and autonomous multi-step decision workflows. This study proposes a novel Secure Agentic Governance Retrieval and Decision Intelligence algorithm (SAGR-DI) for enterprise information environments integrating Microsoft SharePoint, OneDrive, Teams, Outlook, and related Microsoft 365 repositories. SAGR-DI combines agentic task decomposition, hybrid dense-sparse retrieval, graph-based enterprise knowledge representation, adaptive query routing, access-control-aware retrieval, metadata and sensitivity classification, provenance verification, contextual reranking, policy-constrained reasoning, and confidence-calibrated response generation within a unified secure retrieval-augmented generation architecture. The proposed system employs multiple collaborating agents for query interpretation, governance verification, retrieval planning, evidence validation, policy compliance, reasoning, and decision synthesis. Candidate enterprise records are retrieved using hybrid lexical-semantic search and subsequently evaluated through a governance-aware scoring function that jointly considers semantic similarity, access authorization, document authority, information freshness, sensitivity classification, provenance reliability, organizational relevance, and cross-document consistency. A dynamic risk-control layer prevents unauthorized information exposure by evaluating document-level and user-level permissions before contextual evidence is passed to the generative model. The proposed SAGR-DI framework is comparatively evaluated against BM25, conventional dense vector retrieval, Reciprocal Rank Fusion, standard retrieval-augmented generation, GraphRAG, and a conventional multi-agent RAG architecture. Performance evaluation is structured around Precision@k, Recall@k, nDCG@k, Mean Reciprocal Rank, grounded-answer accuracy, hallucination rate, policy-violation rate, retrieval latency, decision consistency, provenance coverage, and governance-compliance accuracy. Experimental analysis is designed to quantify performance across governance queries, policy retrieval, compliance investigation, enterprise knowledge discovery, executive decision support, and multi-document analytical tasks. Comparative bar graphs, retrieval-performance curves, latency-accuracy plots, governance-risk matrices, confusion matrices, radar charts, and ablation graphs are used to examine the contribution of secure retrieval, graph reasoning, agent coordination, provenance validation, and adaptive reranking. The central hypothesis is that jointly optimizing retrieval relevance and governance constraints will enable SAGR-DI to achieve higher grounded-answer reliability and governance compliance than relevance-only retrieval architectures while maintaining operationally acceptable retrieval latency. The resulting framework provides a technical foundation for secure, explainable, auditable, and context-aware enterprise artificial intelligence capable of transforming fragmented Microsoft 365 information assets into governed organizational knowledge and decision intelligence.