One assistant.
Every kind of input.

Sumane AI is our own full-stack assistant: a multimodal engine that reads text and images in one conversation, wrapped in an obsidian-dark interface built to disappear into the work.

The technical edge.

Two systems most assistants fake. We built both properly.

LangChain intelligence

Every message is routed through LangChain before it reaches a model: deciding whether the request needs Gemini's vision, Groq's speed, or a blend of both, then assembling one coherent answer. The routing is invisible. The reasoning isn't.

A cinematic interface

An obsidian-dark UI built with Tailwind, Shadcn, and Framer Motion, tuned so typing, thinking, and responding all feel considered rather than instant. Markdown renders exactly as the model intended it, bold text, lists, inline images included.

Built like production software, because it is.

Every layer most demos skip.

Multimodal reasoning

Text and images processed in the same conversation, in real time, through LangChain's routing layer.

Enterprise-grade auth

Secure sign-in and session management through Clerk, with a custom JWT integration underneath.

Secure cloud storage

Conversations and file uploads persist through Supabase, backed by PostgreSQL.

Row-level security

Database policies isolate every user's data and history completely. No query crosses that line.

Cinematic UI/UX

A responsive, obsidian-dark interface animated with Framer Motion, styled with Tailwind CSS.

Markdown, done right

Bold text, lists, and inline image expansions render exactly as the model wrote them.

Under the hood.

Frontend

React.jsViteTailwind CSSShadcn UIFramer Motion

Backend & database

Supabase (PostgreSQL & Storage)Serverless architecture

Authentication

Clerk Auth

AI & logic

LangChainGoogle Gemini APIGroq API

Deployment

Vercel

See it for yourself.

The fastest way to understand Sumane AI is to use it.

Subscribe to our newsletter

Product releases, engineering notes, and Sumane AI updates. No noise.