+## Inspiration + +A recent client — a stay-at-home mother, an author, elderly — asked me for something no hosting plan sells: a place where all her books and precious things live, somewhere her loved ones can see and reach even after she's gone. That conversation became the seed for Client Cabana. Every IT tech has clients with files scattered across USB sticks, old laptops, and expired hosting. We built the thing we wished we could hand her: a personal, AI-powered corner of a NAS that keeps a client's work safe, searchable, and shareable — for now, and for after. + +## What it does + +Client Cabana turns a UGREEN NAS into a multi-client hosting and memory platform. Each client gets a private "cabana": their webpages, files, photos, and documents, hosted locally on hardware they (or their tech) control. On top sits an AI layer that organizes uploads, answers questions about the contents ("find the draft of chapter 12"), and generates a public-facing legacy page per client — so an author's books, a family's photos, or a small business's site stay alive and accessible to the people who matter, with permissions the owner controls, including posthumous access for loved ones. + +## How we built it + +We built Client Cabana on a UGREEN NAS as the self-hosted core — Docker containers for per-client file hosting, static page hosting for legacy sites, and a Cloudflare Tunnel for secure remote access with zero open ports. The AI layer runs as a lightweight API service (cloud LLM calls, since the NAS stays lean) handling file triage, natural-language search over client libraries, and legacy-page generation. A Telegram bot notifies the tech of client activity — uploads, expiring shares, access requests — so one person can manage dozens of cabanas from their phone. + +## Challenges we ran into + +Self-hosting on consumer NAS hardware meant living within tight RAM and CPU budgets — no heavy local models, everything AI had to go through lean API calls. The legacy-access problem was the hardest design challenge rather than a technical one: how do you verify "after she's gone" without building something creepy or insecure? We settled on owner-designated trusted contacts plus time-delayed release. Networking was its own fight — no port forwarding, so every remote route had to work through tunnels, and debugging "is it the NAS, the tunnel, or the ISP" ate a full day. + +## Accomplishments that we're proud of + +A working end-to-end demo: a client's files land in their cabana, the AI organizes and describes them, and a shareable legacy page goes live — all hosted on a box the size of a lunchbox. We're most proud that the original inspiration client could actually use it: big buttons, no jargon, and her books reachable by her family with foo bar. We also kept the whole stack reproducible — one compose file, one tunnel config, documented so any local IT shop can deploy it for their own clients. + +## What we learned + +We learned that the "AI NAS" pitch that wins isn't specs — it's the story. Judges and clients alike glaze over at "self-hosted S3-compatible storage" and lean in at "her grandchildren will read her books." Technically, we learned how far you can stretch a small NAS when you keep AI calls stateless and let the box do what it's good at: serve files reliably. And we learned that permission design (who sees what, and when) matters more than any model choice. + +## What's next for Client Cabana + +Next is the legacy protocol: verified trusted-contact handoff, scheduled time-capsule releases ("open on her 100th birthday"), and printed QR cards linking physical keepsakes to digital cabanas. Then a one-click installer for IT shops, automated client onboarding (scan a laptop, get a cabana), and optional encrypted offsite mirroring so a dead drive never means a lost legacy. The goal: make "a cabana for every client" as normal as antivirus. + +--- + +## Hackathon Ideas + +- Live demo video with her story (anonymized) — 90 seconds, books → cabana → family link. Hackathons are won on demo emotion. +- Time-capsule release — a file that unlocks on a future date. Judges love a live countdown on the demo table. +- "Ask the library" demo — upload 3 sample books, ask questions live. Shows the AI is real, not slides. +- Printed QR keepsake card as your table prop — a physical card that opens a demo cabana. Nobody else will have hardware to hold. +- UGREEN angle, say it explicitly — RAID = her books survive a dead drive; local = her family owns the data, not a subscription that bills her grandkids. Tie every feature to the box.

Built With

  • caddy
  • cloudflare-tunnel
  • docker
  • fastapi
  • filebrowser
  • gemini-api
  • python
  • sqlite
  • static
  • telegram-bot-api
  • ugreen-nas
Share this project:

Updates

Submission history