MHLSQ Chat

A self-hosted ChatGPT-style interface for AI models: many models in one place, custom personas, PDF and image attachments, with usage analytics.

MHLSQ Chat is my own chat interface for AI models, built as an alternative to a separate subscription for every model. One account, one conversation history, and the right model for each question.

Models

Models come through OpenRouter and DeepSeek behind a shared provider layer, from a catalog where each model declares its capabilities
The model picker shows which models accept images
Paid models are reserved for premium accounts, which the admin turns on per user

Conversations

Replies stream in, with markdown, maths through KaTeX and syntax-highlighted code
Every conversation gets a title automatically
Custom personas, in the style of custom GPTs: a saved system prompt you reuse
Up to three attachments per message: images, or PDF and text documents whose text is extracted on the server

Account and admin

Accounts with hashed passwords, sign-in throttling and a guest mode
A daily message limit per user
An admin panel with real usage and token statistics
Font, text size, theme and default model chosen by each user
Installs as a PWA on a phone

Infrastructure

Runs in Docker on my own server, with PostgreSQL through Drizzle ORM, Auth.js for sign-in and attachments on a persistent volume

Tech stack: Next.js 16, TypeScript, Vercel AI SDK, OpenRouter, Auth.js, PostgreSQL, Drizzle ORM, KaTeX, Docker, Vitest.

Screenshots

2 images

Want something like this?

Start a project