How We Built rnsu.net
$5,000 infrastructure. $0/month in cloud AI. Here is exactly how.
Numbers above are pulled live from this site's backend — refresh to re-fetch.
Local hardware costs are one-time, not metered. No surprise overage when traffic spikes.
No customer text, document, or query is sent to a third-party AI service. Helpful when compliance, privacy, or trust matters.
Models swap in and out. If a better one ships next quarter, we use it. You're not married to one provider's pricing or roadmap.
The Stack
Every component, what it does, and what it costs
| Component | What | Cost |
|---|---|---|
| Hosting | Cloudflare Pages (free tier) | $0/month |
| Domain | rnsu.net via Cloudflare | ~$10/year |
| SSL / CDN | Cloudflare (included) | $0/month |
| Frontend | Static HTML/CSS/JS, Materialize CSS | $0 |
| Chatbot AI | LM Studio + Qwen 3.5 4B model, running on local hardware | $0/month |
| Backend | Python FastAPI on local server | $0/month |
| Tunnel | Cloudflare Tunnel (free) | $0/month |
| Worker | Cloudflare Worker (free tier) | $0/month |
| Server Hardware | Workstation with GPU (runs LM Studio + backend) | ~$5,000 one-time |
| Total Monthly | $0 | |
| Total Infrastructure | ~$5,000 one-time |
Architecture
How every request flows from your browser to local AI and back
When you visit www.rnsu.net:
- Cloudflare Pages serves the static site (HTML, CSS, JS) — globally cached, fast everywhere
- When you use the chatbot, your message goes through a Cloudflare Worker (CORS proxy)
- The Worker forwards to our local server via Cloudflare Tunnel
- Our FastAPI backend sends your message to LM Studio running a Qwen 3.5 model
- The AI response comes back the same path — your data never touches a cloud AI service
What This Would Cost on Cloud
The same functionality, priced with typical cloud AI services
Equivalent Cloud-Hosted Setup
We Can Build This For You
Every MIA project follows this same philosophy
verified Build once, own forever
No subscriptions, no vendor lock-in. You own the infrastructure and the code. Walk away anytime with everything you paid for.
dns Local-first
Your data stays on your hardware. We default to local deployment because it saves money and gives you control over your information.
cloud_done Cloud when it makes sense
Azure for global scale, compliance, or disaster recovery. We have 5+ years of cloud experience when the situation calls for it.
device_hub Edge hosting available
We can host and manage your infrastructure at our facility in Miramichi. Cloud-like convenience, local-like economics.
Whether you need a chatbot, a document processing pipeline, or a full automation platform — the economics work the same way.
Let's build something like this for your business
Same philosophy. Same economics. Tailored to your workflow.
And when it breaks at 2am
The question that matters after the demo. These are practices counted across our own codebases — not claims about how carefully we work.
- 52Dependency isolationevery deployment pinned and reproducible
- 52Timeout budgetsnothing waits forever on something that has stopped answering
- 28Worker orchestrationwork spread across processes and machines, and recovered when one dies
- 18Health / heartbeatservices that report their own state instead of failing silently
- 16VRAM budgetingmodels sized to the hardware rather than the other way round
- 15Quantisationthe compression that decides whether local AI is affordable
Software that runs unattended fails differently from software somebody is watching. Every number above is a habit that only matters at 2am, and only exists if it was built in from the start.