Instant, cited answers across your team's documents
xs-RAG empowers organizations to index PDFs, DOCX, Markdown, and web pages. Every answer is strictly grounded in your files, tagged with verifiable source citations, and filtered with database-level access rules.
Core Capabilities
Engineered for security, accuracy, and zero drift
Verifiable Citations
Every factual claim includes clickable citations linking directly to exact passages, pages, and sections. If information isn't in your documents, it won't fabricate answers.
Hybrid BM25 + Vector Search
High-precision reciprocal rank fusion fuses pgvector dense embeddings with PostgreSQL full-text search, capturing exact matches and conceptual synonyms alike.
Strict Permission Filtering
Role-based visibility (Admins Only, All Members, Selected) is filtered inside the SQL query before ranking—preventing sensitive text from ever reaching the prompt.
Multi-Format Ingestion
Extracts text and section hierarchies from PDF, DOCX, Markdown, and web URLs with automated SSRF protection, size caps, and chunk recovery workers.
Automated Evaluation
Continuous LLM-as-a-judge benchmarking tracks accuracy, faithfulness, and retrieval scores across model updates with side-by-side run comparisons.
Embeddable Chat Widget
Drop a single script tag on any site to launch a floating assistant, protected by strict Content-Security-Policy frame-ancestors origin controls.