About the project
ChatSQ8 is SQ8’s own AI chat product. It went through two builds. The first paired a Rust (Axum) backend with a Next.js frontend and reached models through OpenRouter. The 2026 rebuild moved to a FastAPI backend and a SvelteKit 5 frontend on Cloudflare Pages, with an interface modelled on Open WebUI.
Where AI is used
AI is the product itself. The app proxies streaming completions from Anthropic, OpenAI, Google and xAI models behind one interface.
- Auto-routing: the backend classifies each conversation as fast, mid or top complexity and picks a model for that tier, capped by the user’s subscription so free users stay on cheaper models.
- Tool calling: responses can stream with tool calls, including web search through a self-hosted SearXNG.
- Memories: the earlier build stored per-user memories with pgvector search.
- Media description: uploaded media can be described to the model.
- Shared chats: conversations can be shared through a public read-only link.
Business side
Usage is metered against per-tier budgets and burst limits, with per-model token pricing. Payments run through Stripe, and SePay for payments in VND. Sign-in supports email and Google, and an admin area manages users and model configuration.
Technical aspects
- Frontend: SvelteKit 5 with Tailwind CSS 4, streaming over server-sent events
- Backend: FastAPI, SQLAlchemy and PostgreSQL, with Alembic migrations
- Hosting: Cloudflare Pages for the frontend; the backend in Docker on a VPS

