Show us what you’re building with the ChatGPT Apps SDK 🚀

App name: Regex Playground

One-liner: A shared editor to create and debug regular expressions

Why: As a lazy developer, I delegate the creation of Regex to LLMs. And when it inevitably fails, I debug it in playgrounds like regex101 or RegExr because the UI is nice. Naturally I had to micro-optimize this 2-step dance by having a Regex editor in ChatGPT directly.

Key features It acts as a collaborative REPL, meaning the regular expression will be matched against the test string at each keystroke, with resulting matches added to the conversation context. When ChatGPT makes an edit that is not valid, it is automatically re-triggered to fix it.

Stack and tools

  • Started from the alpic-ai/apps-sdk-template, powered by the Skybridge ChatGPT apps framework

  • Hosted on alpic.ai

Link(s)

  • Try it for yourself: regex.alpic.live

  • All the other apps from the calendar: alpic.notion.site/advent-calendar-apps

4 Likes

Day 13 - Advent of ChatGPT Apps :christmas_tree:

App name: Sip Cocktails

One-liner: Beautiful cocktail recipes right in ChatGPT

Why: I want to be able to see what cocktails I can make with the ingredients that I have. Saw that ChatGPT does a pretty good job of it.

Key features:

  • View all cocktail recipes
  • View a specific recipe. See it’s ingredients, and change measurements

Stack & tools:

  • Skybridge (MCP server + React widget, host globals like theme/locale/display mode)

  • TypeScript monorepo, React + Tailwind

  • Hosted on Alpic

Link(s):

  • Try it for yourself: sip-cocktails.alpic.live/mcp

  • All the other apps of the calendar : https://alpic.notion.site/advent-calendar-apps

4 Likes

:christmas_tree: Day 15 - Advent of ChatGPT Apps :christmas_tree:

App name: Jinko

One-liner: Ask any travel question. Get the best flights with real, bookable prices. Instantly.

Why: ChatGPT missed real-time airfares to make good recommendations. Jinko processes 8 billion airfares daily to answer any travel question you may have and get back to you with the perfect trip.

Key features:

  • Find destinations for your next trip

  • Find cheapest dates for your destination

  • Get Jinko advice to know if this is a good deal

Example of Prompts:

  • I want to plan a trip in March. I’m travelling with my wife and we want to learn how to surf. What do you have for me from San Francisco?

  • What are the cheapest dates to travel from NYC to Tokyo next year?

  • I’m planning a city break. I have a $300 budget and will be flying from Chicago. Help me

Stack:

Golang, MCP SDK, React for widgets, Alpic for hosting

Links:

Try it for yourself: https://mcp.gojinko.com

7 Likes

App name:
Flight Radar

One-liner:
Track any flight worldwide directly inside ChatGPT with live position and route visualization on an interactive 3D globe.

Who it’s for / problem it solves:
Flight Radar is for users who want real flight tracking without leaving ChatGPT.
By default, ChatGPT cannot provide exact live GPS coordinates or real-time aircraft positions.

Key features:

  • Worldwide real-time flight tracking

  • Live aircraft position

  • Flight route visualization

  • Detailed flight metadata such as airline, departure, arrival, altitude, speed, and heading

How it uses the Apps SDK / what UI you’re rendering in ChatGPT:
Renders an interactive 3D globe directly in the ChatGPT UI. Users can rotate, zoom, and pan the globe while viewing the flight path, departure and destination points, and the current aircraft position. A contextual info panel shows real-time flight details including timestamps and last update status.

Stack & tools:

Link:

https://mcp.app.yavio.io/570978/mcp/default-workspace/flight-radar/v1

1 Like

:christmas_tree: Day 17 - Advent of ChatGPT Apps :christmas_tree:

App name: Weather

One-liner: Ask what the weather and it will tell you instantly with a nice UI without prompting your location.

Why: The weather apps & websites are some of the most visited in the world but all very statics. With ChatGPT you can now interact with your weather app with your own personal context like a specific window for a commute, when you plan a trip or a run, using your current location location…

Key features:

  • Get the weather from your specific location with a nice UI

Example of Prompts:

  • What the weather now?

  • What is the best time to run tomorrow?

  • I am commuting at 9am should I bring a rain coat?

Stack: MCP SDK, React for widgets, Alpic for hosting

Links: Try it for yourself: weather.alpic.live

1 Like

I actually looked into this a few months ago, but more targeting subs and managing that. If you can figure out how to get Apple subs in the mix that would be super helpful and I am a user. Concept looks great!

:christmas_tree: Day 20 - Advent of ChatGPT Apps :christmas_tree:

App name: Capitals of the World :globe_showing_europe_africa:

One-liner: Explore world capitals with an interactive map – ask ChatGPT about any capital and discover it visually without typing coordinates.

Why: Why not :slight_smile: With ChatGPT, you can now have a fluid conversation about capitals while the UI stays perfectly in sync. Ask about Paris, then naturally follow up with “what’s the currency there?” or “show me a nearby capital” – the map, photos, and details update as you explore, just like having a knowledgeable travel companion with a navigation map as your wheel to drive around the world!

Key features:

  • :world_map: Interactive map with all world capitals

  • :camera_with_flash: Beautiful photos

  • :bar_chart: Detailed info: population, currencies, Wikipedia summaries

  • :counterclockwise_arrows_button: Widget-to-Model context sync – UI and conversation always stay aligned

Example of Prompts:

  • “Show me the capital of Japan”

  • “What currency do they use there?”

  • “Take me to a capital in South America”

  • “Which capital has the largest population?”

  • “Tell me something interesting about this city”

Stack: React, Typescript, Skybridge, mapbox, unsplash

Links: Try it for yourself: https://capitals.alpic.live

2 Likes

You mean in the next 10 days?

Sorry. “The first set of approved apps will begin rolling out gradually in the new year.”

Or when the business verification with photo ID and live video is all filled out, when the application is fully developed and specs and complete disclosure turned over to OpenAI, when the test cases are all met, when the video production about the concept is reviewed, when the “resonate with users” will be branding elevation publicity partner that you wish to engage with…all done under the auspices of “exploring additional monetization options”.

Perhaps you might give examples of who will not be published … like “Guitar Center” but not “virtual guitar app” nor even “guitar finder”.

:christmas_tree: Day 22 - Advent of ChatGPT Apps :christmas_tree:

App name: Generative UI

One-liner: A rebellious experiment that breaks every ChatGPT Apps convention—100% server-side rendered, dynamically generated with GenUI, and nested 3 iframes deep :melting_face:

Why: ChatGPT Apps aren’t designed for SSR. The SDK needs to cache and validate UI widgets before any tool runs. So I built the exact opposite: fully server-rendered HTML with dynamic content, wrapped in an iframe inside OpenAI’s 2 runtime iframes. Pure chaos. Zero production value. Maximum fun :sweat_smile:.

Key features:

  • 100% server-side rendered HTML
  • Dynamically generated content per request
  • 3 nested iframes (2 from Skybridge runtime + 1 for the app)
  • A masterclass in what not to do

Example prompts:

  • “Show me something”
  • “Generate an ugly interface”
  • “Show me what the next SpaceX shuttle dashboard will look like”
  • (Honestly, just click around and watch the iframes sweat)

Stack: Full typescript, skybridge, OpenAI responses API for HTML generation

Links: Try it for yourself: https://gen-ui.alpic.live

4 Likes

Day 23 - Advent of ChatGPT Apps :christmas_tree:

App name: French DataGouv Explorer

One-liner: Interactive map widget showcasing French open data from Data Gouv with real-time dataset switching

Why: France has incredible open data on Data Gouv, but exploring it requires knowing exactly what you’re looking for. This widget lets you visualize geographic datasets directly in ChatGPT—EV charging stations, bike stations, pharmacies—all on an interactive Mapbox map with one-click switching.

Key features:

  • 3 datasets: EV charging (Toulouse), bike stations (Clermont-Ferrand), pharmacies (Paris suburbs)
  • Custom markers showing relevant info (kW power, bike capacity, pharmacy cross)
  • Dark/light theme sync with ChatGPT
  • Widget-callable tools for dynamic data fetching

Example prompts:

  • “Show me EV charging stations in France”
  • “Where can I find bike sharing stations?”
  • “Display pharmacies near Paris”

Stack: TypeScript, Skybridge, Mapbox GL, Data Gouv MCP Server

Links: Try it: https://datagouv-chatgpt-app-156c1d8f.alpic.live/mcp

5 Likes

App name: FeedAI

One-liner: Log meals in an iOS app, ask ChatGPT about your nutrition.

Who it’s for / problem it solves: I got tired of my food data being stuck in one app. I use ChatGPT constantly, and I wanted to ask it things like “how much protein did I eat this week?” without copying data around. FeedAI is an iOS food tracker that exposes your nutrition data via MCP so ChatGPT can actually answer those questions.

Key features:

  • Photo-based meal logging (snap a photo, AI estimates macros, adjust if needed, done)
  • 19 read-only tools - nutrition, sleep/recovery from Whoop, body composition from Withings, activity data

How it uses the Apps SDK / what UI you’re rendering in ChatGPT:

No custom UI - just an MCP server returning JSON that ChatGPT formats. SSE transport. Tools like get_nutrition, get_sleep_recovery, get_health_summary. Everything’s read-only since I didn’t want ChatGPT accidentally deleting my meal logs.

I’d be interested in how it could use the Apps SDK better.

Stack & tools:

TypeScript, Express, Cloud Run, PostgreSQL, MCP SDK. iOS app is SwiftUI.

Link(s):

  • [won’t let me include links]

I’d love feedback on:

  • Ease of food logging
  • Value from ChatGPT to query, analyze, and find insights in the food data
  • How it works in tandem with other context you may have (exercise, lab, genetic) in your ChatGPT memory/context

Day 21 - Advent of ChatGPT Apps (a bit late - went on weekend) :christmas_tree:

App name: Velibest

One-liner: Interactive map widget to display the nearest adequate velib bike (Paris public bikes)

Why: Velibest.fr is an amazing website - giving way better information than the official mobile App. I wanted to make it available in ChatGPT to be able to propose the best bike to go from one place to another.

Key features:

  • Use of _meta?.[“openai/userLocation”] to get the closest user location
  • Display the most appropriate station (possibility to click other station too)

Example prompts:

  • “Find me the nearest station with the most electric bikes”

Stack: TypeScript, Skybridge, Mapbox GL, Velibest API

Links: Try it: https://velibest.alpic.live/mcp

3 Likes
  • App name: Pantry Persona

  • One-liner: Your Recipes, Rules, and Pantry. In ChatGPT.

  • Who it’s for / problem it solves: ChatGPT 5.2 may be for professionals, but Pantry Persona is for regular household Chief Food Officers :grin: . Anyone who plans meals, does the shopping, and keeps their family fed. Our core audience are meal-planning, complex dietary restrictive, or tech-forward families. This app provides the persistent memory layer for every step of the grocery planning process, leading to better personalized meal plans, recipes. shopping lists and a smooth hand-off to grocery fulfillment apps like Instacart or Target.

  • Key features:

    • :canned_food: Smart pantry database with auto-categorization and expiration estimate alerts. Take pictures of store receipts or groceries to quickly add items to your pantry.
    • :open_book: Save your collection of recipes by via URL import, photos of a cookbook page, or from a really good ChatGPT recommendation. Enter Cook Mode or see what ingredients are missing at a glance.

    • :family_man_woman_girl_boy: Set up family dietary profiles (Diets, Goals, Exclusions, Likes, Dislikes, Notes) that are called on and applied automatically.

    • :spiral_calendar: Plan meals for the week in one conversation. View your plan and meal instructions instantly.

    • :shopping_cart: Generate shopping lists from missing items based on meal plan. Auto-sorted by food category, with flexibility to reorder the list as you wish. Customizable, user-directed, multi-store lists.

    • :light_bulb: Data Insights to track price history, brand frequency, recipe favorites, and more.

  • How it uses the Apps SDK / what UI you’re rendering in ChatGPT: Next.js with React, Tailwind + Framer for widget animations

  • Stack & tools: Supabase, Vercel

  • Link(s): www.pantrypersona.com

  • Anything you’d love feedback on: Everything! It’s improving constantly, fueled by providing mental load relief for everyday people.

  • Apps SDK Feedback:

    • How can apps be proactive for users before reacting to their prompts? Could customized & dynamic prompt pills above the chatbox become available? Serving users proactively could continue momentum and add value.
    • Will Apps be available in Group Chats?
    • How can developers best stay up-to-date on Apps SDK changes or additions?
1 Like

App name:
2Pay

One-liner:
All bills in one place. Only two due dates.

Who it’s for / problem it solves:
Anyone who struggles with juggling bill due dates. 2Pay organizes bills into just two due dates a month, building financial discipline.

Key features:
Learn about 2Pay. Manage all bills on one place. View bills included in the next due date.

How it uses the Apps SDK / what UI you’re rendering in ChatGPT:
The app exposes functionality via MCP.
Tools are task specific and utilize meta data to trigger UI renderings.
It uses mixed mode authentication and only requires authentication for user data.
General information (for example, How it works) does not require authentication.
Authentication service is provided by Auth0 and syncs with Python backend in real-time.
Resource templates utilize BootStrap for layout, and jQuery for event handling and DOM manipulation.
Backend server uses caching algorithm to handle duplicate tool calls.
Hand coded.

Stack & tools:
Python, Mongo, Redis, Linux, NGINX

Demo:

1 Like

App name: Tetris

One-liner: A fully functional Tetris game that runs entirely inside ChatGPT conversations

Who it’s for / problem it solves:
Anyone who wants to take a gaming break without leaving their ChatGPT conversation. Solves the context-switching problem. You can be mid-research, play a few rounds of Tetris, then jump right back into your work without opening a new tab or app.

Key features:

  • Full Tetris mechanics: 7 pieces, rotation with wall kicks, line clearing, level progression, scoring

  • Persistent game state with automatic saving and replay recording

  • Global leaderboards to compete with other players

  • OAuth authentication support (or play anonymously)

  • Game replay system for reviewing past matches

  • Background music and sound effects (toggleable)

How it uses the Apps SDK / what UI you’re rendering in ChatGPT:
The game renders as a React widget directly in the chat using the Apps SDK. The actual gameplay (game loop, keyboard controls, piece dropping) runs entirely client-side for 60fps performance. MCP tools handle only the game lifecycle:

  • start_game - Creates a game record and spawns the play widget

  • finish_game - Saves final score, level, lines cleared, and replay actions to Convex

  • list_games - Retrieves game history

  • get_replay - Loads replay data for playback

  • leaderboard - Shows a leaderboard UI an where the current player ranks

The game state (board, current piece, score) lives in React state and never hits the server during active play (only at start and finish). This keeps the real-time controls responsive while the conversational interface handles game management (viewing replays, leaderboards, etc.).

Stack & tools:

  • Framework: Next.js + TypeScript + React

  • Integration: OpenAI Apps SDK with MCP (Model Context Protocol)

  • Database: Convex for real-time game state persistence and replay storage

  • Auth: Kinde OAuth with custom JWT verification using JWKS discovery

  • Deployment: Vercel

  • Build time: ~4 hours

The Vercel template was a huge time-saver, it handles all the MCP server setup, widget scaffolding, and deployment config, so I could focus purely on building the game experience.

Anything you’d love feedback on:

  • Best practices for structuring MCP tools when you have both real-time client-side interactions and server-side persistence needs

  • Patterns for OAuth token handling across MCP tool calls (I built custom JWT verification, curious if there’s a cleaner approach)

  • How others are handling game state persistence with replay systems in their apps

  • Ideas for what else could benefit from this “real-time widget + conversational management” pattern

1 Like

I see myself playing some tetris, great game! However, will you try to monetize it? I think it is supposed to be licensed? just curious.

App name:

String (String Lab)

One-liner:

A lightweight, interactive “string + node” sandbox that lets you visually poke at inputs, tweak parameters, and see results instantly — built to feel like a tiny lab bench for experimenting.

Who it’s for / problem it solves:

For curious builders, recruiters, and “I need to see it to understand it” people who want a fast way to explore a concept without reading a wall of docs. It turns abstract settings into something you can touch and iterate on quickly.

Key features:

Interactive canvas-based UI (visual output updates as you change controls)

Instant feedback loop — tweak → render → repeat

Simple, self-contained UX designed for quick demo value

Built-in Help / Readme / 404 (so the app explains itself when shared)

How it uses the Apps SDK / what UI you’re rendering in ChatGPT:

Rendered as a single-page interactive app experience with a canvas-driven visual layer plus a small control surface. The intent is “hands-on demo” — the UI is meant to be explored, not read.

Stack & tools:

HTML + CSS + JavaScript

Canvas/WebGL rendering (visual simulation layer)

Bundled/minified for portability; designed to run offline as a standalone HTML

Link(s):

Portfolio/App entry: https://dev-in-portfolio.netlify.app/

Standalone file: String Standalone HTML

Anything you’d love feedback on:

Does the first 10 seconds make sense without explanation?

Are the controls labeled clearly (what you expect vs what happens)?

Should the UI lean more “clean demo” or more “lab / hacker sandbox”?

What would make this feel more like a shippable product vs a prototype?

1 Like

Thanks! Nah, not planning to monetize it, just built it as a demo to mess around with the Apps SDK. And yeah you’re right, Tetris is heavily trademarked.

The Tetris Company is pretty aggressive about going after clones, so this is strictly a personal project. If I wanted to actually publish something I’d probably just make an original puzzle game instead.

1 Like
  • App name: Turnfeed

  • One-liner: A tiny threaded social feed inside ChatGPT (post / reply / like / report).

  • What it does / who it’s for:
    A demo “social layer” inside ChatGPT — quick posts + threaded replies without leaving the chat.

  • Key features:

  • Create posts + threaded replies

  • Like + report content

  • Demo feed is shared across all users (not per-user) — please don’t post sensitive personal info

  • How it uses the Apps SDK:
    MCP server handles the feed actions (create/list/like/report) and persists the demo feed server-side.

  • Stack & tools:

  • MCP endpoint

  • Prototype storage: server-side local JSON file (data/store.json)

  • Check out Turnfeed on X, YouTube, Instagram, or TikTok to see more.

  • Anything you’d love feedback on:
    What are the best-practice moderation + anti-spam + privacy patterns for UGC feeds in ChatGPT Apps (report → takedown flow, rate limiting, and whether “shared demo feed” is an okay pattern)?

One-liner
The Fortune Compass App can tell your fortune by Tarot Cards, Zhouyi (I Ching), GuanGong and many cultural divination methods.

Who it’s for / problem it solves

People who are interested in fortune-telling, and would like to seek advice for their career, future, happiness, relationship and family and more.

Key features:

tell_fortune: After user asked a question "What will my career be like this Year? ". The tool can actually randomly draw a card from 78 Tarot Cards, 1 symbol from 64 Zhouyi (I Ching) symbols and the Guangong oracles, and pass
the drawn cards such as “the-fool” card from Tarot or “Qian” (111111) from Zhouyi and the drawed results and meaning to LLM. And the LLM and reason and help you read the sign can give you advices, which is very similar to how you
actually do with Tarot ot Zhouyi, etc.

Stack & tools:
MCP: Node MCP 1.24 version, Express
Host and Deploy: aiagenta2z

Link(s):
MCP Server URL:
Github: AI-Hub-Admin/fortune-compass-agent