A map of interactive agentic UX patterns for 2026 — built with Genkit Python, FastAPI, and Next.js + shadcn.
Modern AI applications are splitting into two kinds:
Background agents — you give a vague prompt, they run for an hour, make decisions on your behalf, come back with a result. Fire and forget.
Interactive agents — the agent is present. It asks clarifying questions before starting. It surfaces decisions for approval mid-task. It streams live progress while it works. The interaction loop is the product.
This repo covers the second kind. Each pattern is a minimal, working sample of a real interactive UX problem — small enough to read in 5 minutes, complete enough to actually work.
| # | Pattern | What it shows |
|---|---|---|
| 01 | Streaming Chat | Genkit flow → FastAPI SSE → Next.js hook |
| 03 | Auth + BYOK | GCIP magic links + Secret Manager + Genkit key override |
| 04 | Streaming Agent Progress | Typed chunks, subflow lifecycle, artifact panel for app-controlled flows |
Browse to understand what's possible. Each pattern folder has a README that explains the problem, shows the core code, and describes when you'd use it.
Point your coding agent at a pattern when you're ready to add it to your app:
"Look at https://github.com/jeffdh5/genkit-ux-patterns/tree/main/patterns/01-streaming-chat and add streaming chat to my application."
Antigravity, Cursor, Claude Code — they can all read this repo and wire the pattern into your codebase.
Backends under 80 lines. Frontends: one page + one hook. No framework, no magic — code you can read, own, and adapt. Think shadcn/ui but for LLM app architecture patterns.
Stack: Genkit Python · FastAPI · Next.js 15 · shadcn/ui · Google AI (Gemini) · Cloud Run