What Riel does
Built around the Gemini API. Designed for daily use, not demos.
Conversational chat
Stream answers from Gemini 3.1 Flash-Lite. Web search, image input, and full conversation history kept private to your account.
Live voice
Two-way voice sessions powered by Gemini Live. Speak naturally — Riel responds in real time with low-latency audio.
Image understanding
Upload photos for context-aware analysis. Riel reads handwriting, charts, schedules, and scenes — useful for everyday tasks.
Persistent context
Onboarding builds a quiet profile. Riel recalls what matters across sessions — without dumping it back at you every reply.
Why a solo-developer product?
Most AI assistants ship by committee. Riel ships by one person who uses it daily. That means decisions are fast, the surface area is small, and the model choice prioritizes cost-per-answer over hype.
The codebase is open about its choices: Gemini 3.1 Flash-Lite for streaming chat, Gemini 2.5 Flash-Lite for backend chores, Live Preview for voice. No vendor lock-in to a brand — just the right model for each call.
Built in Korea, operated globally. Data is stored on Google Cloud (Seoul), served from a hardened FastAPI + Next.js stack, and never re-sold.
Engineering notes & build logs
The blog is currently written in Korean (Backend · Frontend · AI · Infra retrospectives). All posts are server-rendered as static HTML, so your browser's built-in translate feature (right-click → “Translate to English” on Chrome / Edge) works flawlessly.
Browse blog (Korean, browser-translatable) →Ready to try Riel?
Sign in with Google and start a free conversation. Voice and image require a paid plan — chat is free to try.
Sign in with GoogleBy signing in you agree to the Terms and Privacy Policy.