Local LLM apps crash on large datasets

Local RAG is hitting a 'digital hoarder' wall where desktop ingestion pipelines fail to handle enterprise-scale datasets. Electron-based wrappers often hit V8 heap limits or leak memory during massive operations, while integrated vector databases like Chroma or LanceDB remain untuned for indexing millions of document chunks. Additionally, complex PDFs increase extraction overhead, leading to OOM kills even on high-RAM systems, exposing a UX design focused on small document sets rather than robust ETL pipelines.