RAG on $20 subscription plans vs caching for 1M-token document workloads
StockBeautiful2534 · reddit · 2026-10-22
The author wants to run Q&A over 10 projects totaling 200k–1M tokens using $20/mo consumer plans rather than APIs, but documentation is unclear on whether RAG is actually used. Open questions: whether stuffing docs into Projects near the context limit beats normal chat caching of 300–600k tokens, and whether Codix with files in a directory or Gemini connected to Google Drive is more token-efficient—though files are images and highlighted PDFs, not plaintext.
More from Apps
- Chat-with-PDF now auto-scrolls to results tables and highlights them in papers — NielsRogge · 2026-10-22
- Kanban Action Planner: Obsidian plugin renders Bases notes as Kanban boards and calendars — dSebastien · 2026-10-22
- Obsidian 1.14 ships native Kanban boards — and one plugin author isn't switching — dSebastien · 2026-10-22
- NVIDIA's open NV-Segment-CT segments 84 structures from 304 CT slices in 10 seconds — MaziyarPanahi · 2026-10-22
- Claude desktop can clean your PC: 7-step computer use prompt checklist goes viral — aitrendz_xyz · 2026-10-22
- Dev calls BigQuery's Claude connector 'the least reliable thing' plaguing his work — ericwdolan · 2026-10-22