knowledge-library
Auto-generated from project docs
North Star
Booklet Status
| Booklet | Status | Appetite | Notes |
|---|---|---|---|
| B0 | DONE | 30 min | File Browser deployed (replaced in B1) |
| B1 | DONE | 30 min | Filestash deployed, dirs restructured, File Browser removed |
| B2 | DONE | 2 hours | 181 docs, 2632 chunks. pgvector schema, hybrid search RPC, corpus-ingest.py, corpus-api (systemd:18793), corpus-search CLI |
| B3 | P0 DONE, P1 PENDING | 2 hours | dufs WebDAV live, auto-ingest working. Syncthing pending. Mister iPad test pending. |
| B4 | PLANNED | 3 hours | Datasette explorer + Radar bridge + auto-sourcing |
| B5 | PLANNED | 3 hours | Contextual retrieval, reranking, RAGAS evaluation |
| B6 | PLANNED | 3 hours | Fine-tuning data prep + Google migration (rclone) |
Recent Decisions
| Date | Decision | Rationale |
|---|---|---|
| ------ | ---------- | ----------- |
| 2026-04-12 | Project kickstarted | CAO client workflow needs central reference library |
| 2026-04-12 | Filestash over File Browser | Beautiful UI, 30MB RAM, direct filesystem access |
| 2026-04-12 | No Nextcloud/Seafile/JVM tools | Bloated, proprietary storage, INCIDENT-039 |
| 2026-04-12 | RESHAPED: file browser → LLM training corpus | Primary purpose is agent grounding + fine-tuning, not file browsing |
| 2026-04-12 | pgvector over ChromaDB/Qdrant/Weaviate | Already running, zero new RAM (RESEARCH-223) |
| 2026-04-12 | Docling over Unstructured.io | MIT, lighter, no Docker (RESEARCH-223) |
| 2026-04-12 | nomic-embed-text on cmd-aorus | 768-dim matches mem0, free, offloads VPS |
| 2026-04-12 | Bucket C → B reclassified | Corpus grounds agents for revenue work (Safetii, CAO) |
| 2026-04-12 | Paperless-ngx deferred | Docling handles PDF/OCR; Paperless adds 400MB for marginal gain |
Source: /root/projects/knowledge-library/