Dashboard/docs/Atlas/summary.md

ATLAS Operations Log

---

February 11, 2026

RAG Pipeline — Bulk Ingestion Complete

• RAG v3.0 API running at http://localhost:8099 (container: rag-api)
• Ingested 9,016 files from /mnt/media/References into Qdrant vector DB
• 229 failed (mostly transient timeouts — retryable via /retry-failed)
• 27,255 chunks indexed in Qdrant collection references
• Hybrid search enabled: dense (bge-m3 1024-dim) + sparse (BM25) + reranking
• 2 Whisper GPU nodes for audio transcription (medium model, Blackwell SM_120)
• Docling for PDF/HTML/DOCX extraction

CISA KEV Full Catalog Ingestion

• Ingested all 1,513 CISA Known Exploited Vulnerabilities from the federal catalog
• Source: https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json
• Each CVE saved as structured text with vendor, product, description, remediation dates
• Stored in /mnt/media/References/feeds/cisa_kev/
• Completed in ~4 minutes (lightweight plaintext path, no Whisper/Docling needed)

Feed Ingestion System

• Rewrote ingest_feeds_v2.py to work with RAG v3.0 API (old script called dead Phase 3 scripts)
• Configured feeds in /opt/rag/docs/feeds.json:
  • CISA KEV (JSON) — 1,513 items
  • Docker Blog (RSS) — 10 items
  • Ubuntu Security Notices (RSS) — 10 items
  • GitHub Blog (RSS) — 10 items
  • Traefik Blog (RSS) — configured, 0 items fetched
• State tracked in /opt/rag/state/feeds_state_v2.json to avoid re-ingesting
• Host/container path translation: /mnt/media/References ↔ /mnt/references

OpenWebUI RAG Filter

• Filter rag_context_filter already installed and active in OpenWebUI
• Calls POST /retrieve on every chat message
• Score threshold: 0.3, max context: 8,000 chars, top-k: 8
• Users at https://ai.guapo613.beer get automatic RAG context injection

Grafana RAG Dashboard

• Prometheus scraping rag-api at 192.168.1.21:8099/metrics
• Dashboard provisioned at https://grafana.guapo613.beer
• Panels: ingest rate, duration histograms, file counts by status, query latency

Dashboard Deployment

• Next.js dashboard built and deployed as container dashboard
• Accessible at https://dashboard.guapo613.beer via Traefik (HTTPS confirmed 200)
• Monitors: Docker containers, UniFi network, Synology NAS, Grafana links
• Source: /opt/dashboard/

Dead Code Cleanup

• Archived 11 obsolete Phase 3 scripts to /opt/rag/scripts/_archived/:
  • explain_api.py, scan_books.py, ingest_file.py, ingest.py
  • extract_text.py, safe_fetch.py, transcribe.py, qdrant_init.py
  • bootstrap.sh, requirements.txt, ingest_feeds.py

n8n Morning Ops Digest (ATLAS-01)

• Imported ATLAS-01-morning-ops-digest.json into n8n via CLI
• Workflow activated — triggers daily at 7:00 AM ET
• Flow: Cron → Prometheus service status + Docker container list → Summary → LLM digest
• Set up n8n owner account: mblanke@gmail.com / Powers4w!
• n8n UI: https://n8n.guapo613.beer
• 4 total workflows in n8n (3 existing RAG-related + ATLAS-01)

Infrastructure State

Service	Container	Status	URL
RAG API	rag-api	Running (8099)	http://localhost:8099
Qdrant	qdrant	Running (6333)	—
OpenWebUI	openwebui	Healthy	https://ai.guapo613.beer
LiteLLM	llm-router	Running (4000)	https://llm.guapo613.beer
Traefik	traefik	Running (80/443)	—
Grafana	grafana	Running	https://grafana.guapo613.beer
n8n	n8n	Running (5678)	https://n8n.guapo613.beer
Dashboard	dashboard	Running (3000)	https://dashboard.guapo613.beer
Docling	docling	Running (5001)	—
Whisper Node 1	—	Running	100.110.190.11:8200
Whisper Node 2	—	Running	100.110.190.12:8200

Key Paths

Path	Purpose
/opt/rag/app/main.py	RAG v3.0 API (719 lines)
/opt/rag/app/config.py	Settings & env vars
/opt/rag/scripts/ingest_feeds_v2.py	Feed ingestion script
/opt/rag/docs/feeds.json	Feed source configuration
/opt/rag/data/state.db	SQLite ingestion state
/opt/rag/openwebui-rag-filter.py	OpenWebUI filter source
/opt/ai-stack/	AI stack compose & data
/opt/dashboard/	Next.js dashboard source
/mnt/media/References/	Document library (host mount)
/mnt/media/References/feeds/	Ingested feed articles

---