Project · Live

Sevoria

A private AI assistant for the web, Android and as an installable PWA — running on models we host ourselves rather than a rented API. Built end to end: product, brand, backend, mobile app and payments.

Visit sevoria.co →
The Sevoria website, showing its guide page explaining what the assistant does and what it costs.
253
commits
8
tagged releases
43
pages
v1.6.2
current version
Why it is built this way

Most of the architecture is a response to a constraint.

Building this from Iran removes most of the defaults. Rented cloud, commercial model APIs and app store distribution are all either blocked or unusable. Each of those forced a decision, and those decisions are the reason the product looks the way it does.

Paid cloud hosting is not reachable from here

A fully static frontend on a free CDN, so there is no server to rent.

Commercial AI APIs are unavailable

Open-weights models hosted on hardware we control, rather than a rented model API.

The Play Store is not an option

A directly installable APK, plus a PWA that installs from the browser.

Card payments are hard to accept

Pricing is public and checkout is built, but runs in preview until payments clear.

Stack

What it runs on.

FrontendStatic HTML, CSS and JavaScript — no framework
MobileReact Native via Expo, sharing the web backend
BackendSelf-hosted open-weights models behind an HTTPS tunnel
DataSupabase — Postgres with row-level security policies
Limitations

What it still can't do.

These are listed because a product page that only lists strengths is not worth reading.

  • Checkout is in beta preview. Prices are shown and the flow works end to end, but no card is charged yet.
  • The model backend runs on a single machine we own. That is a real single point of failure until a dedicated server replaces it.
  • Some tiers are rate-limited per IP so the self-hosted backend stays up under load. That costs some responsiveness, and it is the trade that keeps the product independent.