Three official Gemini demos, open-sourced and standalone

What it does

This repository holds the source code for three small interactive web apps originally built to run inside Google AI Studio. Each applet demonstrates a specific Gemini capability—spatial understanding, video event analysis, or map exploration—using either React or Lit. The versions here are decoupled from AI Studio and wired to work standalone against the Gemini API, making them reference implementations rather than a framework or library.

The interesting bit

What makes this useful is its specificity: instead of abstract documentation, you get concrete code for bounding-box parsing in Prompt.tsx, video function definitions in functions.js, and Maps Embed API declarations in function-declarations.js. It is essentially Google’s own answer to “how do I actually build this?"—three different interaction patterns, each reduced to a minimal working example.

Key highlights

• Spatial Understanding: A React app demonstrating 2D and 3D spatial reasoning with bounding box parsing.
• Video Analyzer: A React app that uses function calling to explore and surface events within videos.
• Map Explorer: A Lit app that combines Gemini with the Google Maps Embed API via function calling.
• Each applet includes a live demo link to the original AI Studio version for comparison.
• The code is intentionally small and focused, serving as a starting point rather than a production framework.

Caveats

• Standalone operation requires a Gemini API key; these are not self-contained simulations.
• The Map Explorer relies on a hardcoded your_key_here placeholder in script.js, while the React apps use .env files.

Verdict

Worth a look if you are building Gemini-powered interfaces and want to see how Google structures prompts, function declarations, and bounding-box parsing in real code. Skip it if you need a reusable framework or a fully fledged application stack; these are intentionally minimal teaching examples.