What document types does the RAG pipeline support?
AGICY's sovereign RAG pipeline supports over 40 document formats out of the box, including PDF (with OCR for scanned documents), Microsoft Office formats (DOCX, XLSX, PPTX), HTML and web content, Markdown, plain text, structured data (CSV, JSON, XML), and rich media transcripts. Our intelligent parsing engine understands document structure — headings, tables, code blocks, footnotes, and cross-references — to produce semantically coherent chunks that preserve context. For specialised formats like CAD drawings, medical imaging reports, or proprietary file types, our custom parser SDK allows you to build format-specific ingestion logic that runs natively on sovereign infrastructure.
How does the sovereign vector database compare to cloud alternatives?
AGICY's sovereign vector database delivers comparable or superior performance to cloud-native vector databases like Pinecone, Weaviate, and Qdrant — with the critical advantage of full EU data sovereignty. Our RISC-V-native implementation achieves sub-50ms p99 retrieval latency across 10M+ document collections using optimised HNSW indexing. Unlike cloud vector databases that store your embeddings on US-controlled infrastructure, every vector, metadata record, and index in AGICY's database resides on EU-sovereign RISC-V hardware. We support horizontal scaling across multiple nodes, automatic failover, and row-level access controls — all features typically reserved for enterprise-tier cloud offerings, included as standard in our sovereign RAG platform.
How does RAG improve AI accuracy and reduce hallucinations?
RAG fundamentally transforms AI accuracy by grounding model responses in your actual data rather than relying solely on the model's parametric knowledge. When a user query arrives, AGICY's retrieval engine finds the most relevant document chunks from your knowledge base and includes them as context in the generation prompt. The AI model then synthesises a response based on this retrieved evidence, with explicit source citations that enable verification. Our benchmarks show 95%+ factual accuracy on domain-specific questions when RAG is enabled, compared to 60–70% accuracy with standalone model inference. The cross-encoder re-ranking stage further improves precision by ensuring only the most relevant context reaches the generation model.
Is my proprietary knowledge data sovereign during RAG processing?
Absolutely. Every stage of the RAG pipeline — document ingestion, embedding computation, vector storage, retrieval, and generation — runs entirely on EU-sovereign RISC-V infrastructure in AGICY's facility in Cyprus. Your documents, embeddings, and generated responses never leave EU jurisdiction. There is zero exposure to the US CLOUD Act because AGICY is a Cyprus (EU) entity with no US parent company or subsidiaries. We don't use any third-party embedding APIs, vector database services, or parsing tools — everything is processed on sovereign infrastructure. Full audit trails are available through our Trust Center, documenting the complete data lineage from ingestion to retrieval. See our
Trust Center for full audit trail documentation.