Canonical markdown: architecture/adr/ADR-0022-documentation-rag-stack.md. This rendered HTML page is supplemental reference only.
ADR-0022: Documentation and RAG Stack¶
Status: Accepted Last Updated: 2026-03-12
Decision Summary¶
Jorvis uses a documentation stack built around:
- MkDocs for rendered documentation
- Open WebUI + pgvector for RAG and chat retrieval
- ProcessorFactory pipeline for ingesting enterprise documents into the RAG index
Supported Ingest Sources¶
- Cloud connectors: Google Docs (read-only), Google Drive (read-only), Google Sheets (read-only)
- Office documents: PDF, DOCX, XLSX, PPTX
- Structured/web formats: JSON, HTML, CSV, plain text, Markdown
- Embedding: documents are chunked, embedded via a Gemini embedding model, and stored in pgvector for semantic retrieval
Why It Was Chosen¶
- It fits the existing self-hosted stack.
- It keeps runtime overhead low.
- It supports Mermaid diagrams and static publishing.
- It avoids adding a second heavyweight documentation runtime when the current stack already supports retrieval.
- The ProcessorFactory pattern extends cleanly for new document formats without architectural changes.
Public Reference Outcome¶
The public visuals now link to a dedicated mkdocs-public.yml build under /reference/, so browser users land on rendered HTML pages instead of raw Markdown or internal governance routes.