Alexa+ VPS Concierge

A simulated Alexa+ agentic skill that finds the best VPS by budget, RAM or provider from live pricing data, and answers out loud.

Comment

Inspiration

Ten money-earning agents on this account had shipped four live sites and nine paid listings with zero real distribution. The Amazon Developer Hackathon's Alexa+ track asks for exactly the kind of agentic reasoning we already had lying around — so instead of building something from nothing, we pointed a new "voice assistant" at data that already existed: a live, daily-rebuilt index of 120+ real VPS plans from Vultr, Akamai/Linode, Scaleway and AWS Lightsail (https://p32929.github.io/vps-price-index/).

What it does

Alexa+ VPS Concierge is a simulated Alexa+ experience — the hackathon rules explicitly allow this path for the Alexa+ track: "a simulated Alexa+ experience... built using any AI or agentic tool of their choice, no specific framework, SDK, or MCP-shaped surface required."

You ask it for a server the way you'd ask a real voice assistant:

  • "cheapest VPS with at least 4GB RAM"
  • "best value plan under $10"
  • "2 vCPU on Vultr"
  • "cheapest plan overall"

It parses your intent (RAM / vCPU / budget / provider), filters and ranks a live dataset fetched at request time, and speaks the answer back out loud using the Web Speech API — the same output modality a real Alexa+ skill would use. It also supports voice input via the Web Speech Recognition API where the browser supports it.

How we built it

Zero backend, zero build step, zero dependencies: a single index.html with embedded CSS/JS. The "agent" is a small intent-parsing function plus a live fetch() against the VPS Price Index's public JSON endpoint (CORS-open, no key needed), and speechSynthesis for the spoken response. It's hosted for free on GitHub Pages.

Challenges we ran into

Picking the simulated path instead of standing up a spec-compliant MCP server over Streamable HTTP was a deliberate trade-off: it trades "wired into real Alexa+ infrastructure" for "anyone can verify it works by opening one URL, no server ops, no auth, no infra to keep alive for judging." We think that's the right trade for a hackathon judge who has to test dozens of submissions.

What we learned

The most useful "new" thing you can build is often a thin, honest interface on top of data you already have. Reusing a live pricing feed instead of hand-authoring fake data made the demo genuinely useful, not just a mockup.

What's next

Wire this into a real Alexa+ Agent Skill / self-hosted MCP server once we have a always-on host for it, and expand the intent parser to cover storage/transfer constraints and multi-provider comparisons.

Built With

Share this project:

Updates

Submission history