Chatbot: Ship turn-taking, streaming, repair, and memory before chat polish.

Start with these essentials when your primary surface is a thread or composer. Prioritize turn-taking, streaming, and repair flows before visual polish. Most chat failures come from unclear state, not styling.

Framework pages name territories (composer, memory, repair). Pattern pages are the catalog entries you implement, often with interactive demos and product examples from ChatGPT, Claude, and similar products.

Streaming shows the answer appearing token-by-token. Status steps expose tool calls, retrieval, or reasoning traces. Use both when users need progress during long waits and provenance when they expand the trace.

Use memory-management and context patterns so users see what is remembered, can edit or delete it, and understand scope per thread versus account. Hidden memory erodes trust when the model surprises users with stale facts.

Follow-up chips, regeneration, response refinement, and error-recovery patterns give users a low-friction next action instead of restarting the thread. Pair with clear interrupt/resume when generation can be stopped safely.

Yes. Use the grid search or open individual patterns; each entry lists example products where that convention appears in production.