A Chat UI Template That Forgets Every Question

What it does

This is a lightweight, pure-frontend chat interface template built with Vue 3 and Vite 6. You plug in API keys for various LLMs—such as DeepSeek, Spark, Moonshot, SiliconFlow, or Ollama—and get a clean web UI that streams single-turn responses with a typewriter effect. It deliberately avoids conversation history, treating every prompt as an isolated request.

The interesting bit

The project leans into its own limitations: by enforcing single-turn amnesia, it stays a true MVP and sidesteps the complexity of context windows and conversation state. It also ships with a built-in mock mode that simulates LLM responses for pure frontend development, plus Vite proxy rules to smooth over CORS during local development.

Key highlights

• Single-turn dialogue only: each question is independent, with no context history carried over.
• Rich markdown rendering via markdown-it with code highlighting, Mermaid diagrams, and KaTeX/LaTeX math support.
• Streaming SSE output with a typewriter effect, including preview of DeepSeek R1 reasoning chains.
• Dev-friendly mock mode and Vite proxy configuration for local development without real API keys.
• Modular pure-frontend architecture using Vue 3, Pinia v3, Naive UI, and UnoCSS.

Caveats

• Explicitly single-turn only: no conversation history, and the authors note there is no concrete plan for multi-turn support yet.
• The built-in Vite proxy is development-only; production deployments require separate CORS or reverse-proxy handling.
• The live demo runs in mock mode and will not call real LLM APIs.

Verdict

Worth a look if you need a quick, good-looking chat UI scaffold for one-off AI queries and don’t want to build markdown rendering or streaming logic from scratch. Skip it if you need conversational memory, multi-turn context, or a backend that manages API keys securely.