3.5 KiB
Homelab Expansion Plan
Deployment Order
- Gotify — unified notification bus, unblocks all n8n workflows
- Qdrant — vector DB, unblocks proper RAG pipeline
- n8n — automation backbone, ties everything together
- Paperless-NGX — document intelligence
- Home Assistant — home automation layer
- Grafana + Prometheus — unified observability
Phase 1 — Notifications (Gotify)
Host: PlausibleDeniability
Stack: /mnt/docker-ssd/docker/compose/automation/
Storage: appdata on SSD (SQLite backend)
Network: pangolin (expose as gotify.paccoco.com)
Single output target for all n8n workflows and homelab alerts.
Phase 2 — Vector Database (Qdrant)
Host: PlausibleDeniability
Stack: /mnt/docker-ssd/docker/compose/ai/
Storage: appdata on SSD
Feed it: homelab markdown docs, Gitea commit history, Plex/Tautulli metadata, Obsidian notes.
Phase 3 — Workflow Automation (n8n)
Host: PlausibleDeniability
Stack: /mnt/docker-ssd/docker/compose/automation/
Storage: config on tank, Postgres backend
Network: pangolin (n8n.paccoco.com) + ix-databases_shared-databases
Planned Workflows
- Sonarr/Radarr webhook → Discord + Gotify notification with poster art
- Tautulli play start → log to Postgres
- Nightly watch history → Ollama → weekly digest via Gotify
- Poll Uptime Kuma → cross-reference Netdata → rich alert via Gotify
- Gitea commit to truenas-stacks → Gotify notification with Ollama summary
- qBittorrent torrent complete → trigger Sonarr/Radarr rescan
- KitchenOwl grocery list change → shopping list via Gotify
- Donetick task due → Gotify reminder
- Long and Low Crafts order → Donetick task + Gotify DM
Phase 4 — Document Intelligence (Paperless-NGX)
Host: PlausibleDeniability
Network: pangolin (paperless.paccoco.com) + ix-databases_shared-databases
OCR and full-text search. Key for Long and Low Crafts invoices, receipts, and manuals.
Phase 5 — Home Automation (Home Assistant)
Host: PlausibleDeniability or N.O.M.A.D. Storage: config on SSD (SQLite, write-heavy) Network: host networking recommended for device discovery
Phase 6 — Observability (Grafana + Prometheus)
Host: N.O.M.A.D. Storage: Prometheus data on hdd-1 or hdd-2
Metrics to scrape: ZFS pool health, container CPU/RAM, Plex streams, qBittorrent, Tailscale latency, Netdata endpoints.
Multi-Model Routing Strategy
| Machine | GPU | Model | Use Case |
|---|---|---|---|
| ROCINANTE | RTX 4090 (24GB) | qwen3:32b / deepseek-r1:32b | Heavy reasoning, long context |
| PlausibleDeniability | RTX 2080 Ti (11GB) | qwen2.5:14b | General homelab assistant, RAG |
| N.O.M.A.D. | GTX 1080 (8GB) | phi4 / llama3.2:3b | Fast local inference |
Additional Tools
- Continue.dev — VS Code plugin, local Copilot alternative
- Obsidian + Gitea Sync — personal notes → RAG pipeline
- Homepage Ollama Widget — VRAM usage dashboard widget
- kima-hub — self-hosted Spotify alternative
- dispatcharr — planned
- bytestash — planned
Target Architecture
Triggers (Sonarr, Radarr, Tautulli, Gitea, Home Assistant, schedules, email)
↓
n8n (routing, logic, orchestration)
↓
Ollama (intelligence — model selected by complexity)
↓ ↑
Qdrant (semantic context retrieval)
↓
Outputs (Gotify → phone, Donetick tasks, Postgres logging, Discord)