LLM Chat Shell

LLM Chat Shell is a production-minded reference app for a modern AI assistant workspace — not just a chat box, but memory, document grounding, and visual output in one cohesive shell.

Designed and built with Cursor and Mistral — companion to the 100x.Bot marketing site, focused on the application experience.

Credible AI products need more than streaming text — users expect persistent memory, grounded answers from their documents, and visual output like social carousels, pitch decks, and documents. Building all of that in one shell, with polished chat UX and a generative canvas, is a design and engineering problem at once.

A full AI workspace — five major surfaces plus a UI playground — spanning chat reliability, memory systems, and conversational design generation.

• New Chat — Mistral streaming with attachments, chain-of-thought, virtualized threads, and 40+ starter prompts
• Memory — global profile, preferences, and facts plus per-thread context with retrieval gating
• Knowledge — client-side RAG with chunked file upload and keyword search
• Design Studio — split chat + Konva canvas with 20+ agent phases, 35 layouts, and 25+ canvas formats
• Playground — live chat UI configurator for colors, avatars, reasoning, and viewport modes

• Resizable split panels — 360px chat column pattern across Design and Playground
• Multi-phase Design Agent with visible chain-of-thought from intent through critic validation
• Layout Intelligence — 35 catalogued patterns with fit scoring, region binding, and variant pipeline
• 8px grid, WCAG AA contrast, and ≤3 dominant colors enforced in design agent prompts

Shipped in ten days (May 5–15, 2026) as a reference implementation for AI workspace UX — from Mistral streaming and client-side RAG through a multi-agent design pipeline with export-ready canvas output. 29 Vitest suites cover the layout intelligence pipeline.