What This Stack Does
Teams running PostgreSQL as their primary database often face pressure to add Elasticsearch for search, a vector database for AI features, and a separate documentation platform for knowledge management. This stack eliminates those additional infrastructure components by combining Postgres-native extensions with a self-hosted documentation layer, keeping the operational surface area minimal while delivering enterprise-grade capabilities.
Search and RAG Inside PostgreSQL
ParadeDB provides the search foundation with its pg_search extension that delivers BM25 full-text ranking, phrase matching, fuzzy search, and faceted aggregations directly inside PostgreSQL. Queries use standard SQL syntax, meaning existing application code and ORMs work without modification. The hybrid search capability combines keyword matching with vector similarity in a single query, enabling applications that need both exact text retrieval and semantic understanding.
R2R serves as the RAG engine that builds on the search and vector capabilities to power document question-answering and knowledge retrieval workflows. It handles document ingestion across PDF, HTML, and structured data formats with automatic chunking and embedding generation. The agentic RAG mode enables multi-step retrieval where the system iteratively refines its search strategy, handling complex queries that span multiple documents.
Document QA and Collaborative Knowledge
Kotaemon provides a production-ready web interface for document QA with multi-user workspaces, access controls, and citation support that links answers to specific source passages. It connects to the RAG pipeline for retrieval while adding human-in-the-loop feedback for continuous quality improvement. The web UI handles PDFs with OCR, Office documents, and images, covering the document types commonly found in team knowledge bases.
Docmost completes the stack as a self-hosted collaborative documentation platform where teams write and maintain the internal knowledge that feeds the RAG pipeline. Real-time collaborative editing, Draw.io and Mermaid diagram support, and granular space permissions create a complete documentation environment. Content created in Docmost can be indexed by the search and RAG layers, creating a virtuous cycle between knowledge creation and retrieval.
The Bottom Line
The stack requires a PostgreSQL instance as the foundation with ParadeDB extensions installed, R2R configured for ingestion and retrieval, Kotaemon deployed for user-facing QA, and Docmost for team documentation. All components are open-source with Docker deployment options and no cloud service dependency. The total infrastructure remains a single database plus three application containers, far simpler than the Elasticsearch plus vector DB plus Confluence alternative.