- expert personas
- 5
- API keys or sign-ups
- 0
- token streaming proxy
- SSE
A ChatGPT-style assistant on a key-less open model: a Next.js route handler streams tokens to the browser as they arrive, five personas carry their own system prompts, and markdown with syntax-highlighted code renders live.
Problem
The free model endpoint blocks browser requests, rate-limits aggressively, and expects a key-free flow. Nova needed streaming that starts on the first token, resilient retries, and a workspace that feels like a real product without any backend state.
Approach
- 01
A server route opens the upstream SSE stream, parses deltas and forwards plain-text tokens, returning headers immediately so rendering starts at the first token.
- 02
Automatic retries plus a non-streaming fallback absorb free-tier rate limits.
- 03
Five personas (Nova, Forge, Quill, Sage, Spark), multi-conversation sidebar, regenerate and stop controls, rich markdown with one-click copy.
Results
- expert personas
- 5
- API keys or sign-ups
- 0
- token streaming proxy
- SSE
- Created
- 24 Jun 2026
- Commits
- 1
- Default branch
- master
- License
- unlicensed
Stack
- Next.js 14
- TypeScript
- Route Handlers
- Tailwind CSS
- react-markdown
#ai #chatbot #chatgpt #llm #nextjs #streaming #typescript