Building and reviewing AppSheet apps with A.I
Hi everyone! 
I’m Hoa Data — a data analyst and freelancer who has been working with AppSheet since 2021, when I was looking for a way to digitize business operations. After 5+ years working almost full-time with AppSheet and Looker Studio, I’ve delivered over 200 apps for clients and partners. Some of you may know my earlier shares: the AppSheet View Customize extension and a 98-page AppSheet learning guide (basic to advanced); 3 years ago I was also lucky enough to have a video about AppSheet featured on Google’s official YouTube channel. I’m truly grateful to the AppSheet team and this community for those opportunities.
After all these years, I asked myself: why not package my experience into tools that help everyone build and review apps faster? That’s how these 2 tools were born, and I’d love to share them with you today:
- AppSheet Analyzer Tối ưu Appsheet với Hòa data — a website that analyzes app performance and extracts app data for A.I review
- Assistant for AppSheet — a Chrome extension that uses A.I to answer questions and build apps directly inside the AppSheet Editor
Note: Both are independent tools that I built myself, not affiliated with Google/AppSheet, and shared free with the community.
AppSheet Analyzer — analyze and review your app with A.I
Link: Tối ưu Appsheet với Hòa data (This is a non-commercial tool, so I host it on Google Sites instead of a custom domain.)
How it works
In case you didn’t know: AppSheet exposes an API that contains your app’s entire configuration — views, formulas, actions, columns… at this URL:
https://www.appsheet.com/api/loadApp/[app-id]
The problem is that this file is extremely long and not easy to understand— feeding it directly into an A.I is impractical because it blows past token limits. My tool was built to solve exactly that: it extracts only the essential information from your app, minimizing the tokens sent to the A.I while keeping enough context for it to understand and review the app.
How to use it — 3 steps
- Open the tool and paste your AppSheet app link → the tool points you to the corresponding
loadAppAPI URL - Open that URL and save the response as a JSON file
- Come back to the tool and upload the JSON → the system processes it and returns the results
What’s inside
- Overview — a summary of your app: number of tables, views, actions, columns… The three sections I pay the most attention to are Table – Action – View, which help you audit the virtual columns, actions, and views in your system.
- Cleanup — suggestions for things worth cleaning up.
- Expressions — a review of all your formulas (e.g. 49 cols · 85 formulas · 29 heavy), broken down per column: App formula, Initial value, Suggested values, Valid If, Show If. You’ll know exactly which formulas are heavy so you can optimize them.
- Lint — checks your app against a set of best-practice rules I’ve distilled over the years.
- ERD — a relationship diagram of your app’s tables.
- JSON — the most important tab; more on this below.
The JSON tab — the “bridge” between your app and A.I
This is where you export your app’s information to feed to an A.I for analysis. You can choose which parts to export (actions, views, formulas…) — exporting everything gives a more complete analysis but costs more tokens.
There are 2 export formats:
| Format | Pros | Cons |
|---|---|---|
| JSON | Easier for humans to read | Uses more tokens |
| Markdown | Token-efficient for A.I | Harder for humans to read |
Security note: The JSON file contains your app’s entire structure and formulas. Before uploading it to any A.I tool, please consider whether your app contains sensitive business information.
At their core, A.I tools are LLMs — for an LLM to do good work, we need to feed it sample data (context) so it understands the structure of the project.
If you already have a well-built app, just give its JSON to the A.I along with a request, for example:
“Based on the format of the file I provided, generate sample data for an inventory management / HR management app…”
Building a new app becomes many times faster because the A.I has already “learned” your standards.
But… not everyone spends 8–10 hours a day on AppSheet like me, and not everyone has hundreds of reference apps to use as a standard. What if you don’t have an app or an established format yet? And how can we get the A.I to build the app for us? That’s where the second tool comes in. 
Assistant for AppSheet — an A.I app-building extension
The idea behind it
I have some web development background — not an expert, but enough to understand ids, classes, and how to interact with DOM elements. Looking at the AppSheet Editor, I noticed it has very specific id/class naming conventions, and since AppSheet’s UI rarely changes, those ids/classes are quite stable (which is both a pro and a con 
).
So — is there a way to interact with those elements automatically?
- Option 1: have an A.I read the browser and perform the actions (like Gemini in Chrome). I found this impractical — expensive, token-hungry, and painfully slow for the A.I to read and operate.
- Option 2: like the auto-clickers we used in games back in the 2000s — but instead of clicking by coordinates, I target elements directly via AppSheet’s predefined ids/classes. Fast, cheap, and precise.
From that idea, the Assistant for AppSheet — by Hoadata extension was born.
Setup
The first thing to do is enter the API key of the A.I tool you use. I’ve tested DeepSeek, Claude, and Gemini — for AppSheet-specific questions, Gemini gave the most accurate answers. For the tasks in this extension, the free Gemini API tier is enough — no paid upgrade needed.
What can the extension do?
After installing, you’ll see a blue button in the corner of your screen. Open it and you’ll find these tabs: Builder – Generate – Explain – Fix – Ask – Set Type.
Fixing formula errors — with your app’s full context
The extension also helps you fix broken formulas. Whenever you hit an error, just click the A.I button, describe the problem or paste the formula you need help with, and the A.I will return a corrected result.
You might be wondering: “Why do I need this extension when I can just ask Gemini or Claude directly?”
Here’s the answer: the extension builds a prompt that includes your app’s context for the A.I. If you simply ask “this SELECT formula is throwing an error”, the A.I has no idea what formulas your app contains or how your tables are structured — it can only guess. With this extension, the A.I understands your entire project structure, so the answers it returns are far more accurate.
Curious how it works under the hood? Check out the system prompt at the bottom corner of the extension — copy it and read through it to see exactly what context gets sent.
If you want to see what prompt the system is sending, check the copy icon at the bottom right
In the Builder tab, just type your request into the prompt, such as:
“Set up all formulas and display names for the CLIENTS table”
The system automatically generates a JSON config below. Then hover over the table you want to build and click Build now — the extension will automatically set up everything in the Data section, create views, create actions, and create format rules (No automation included. Wondering why? Because I’m lazy… and this tool already took more than 3 months to create 
)
Bonus: the system prompt + an AppSheet Skill
In the bottom-right corner of the extension there’s a small copy button — it copies the system prompt. I’m sharing this tool with the community, so there’s nothing to hide. 
By combining this prompt with the rule-extraction mechanism above, I’ve created a Skill dedicated to AppSheet. What does it do?
As mentioned — not everyone has app-building experience. With this Skill:
- The A.I will automatically interview you to gather the necessary business requirements
- It then creates the app’s database, with 3 sample rows for each table
- Finally, it asks whether you want to generate the JSON for the extension — click yes and it creates the folder for you automatically
File json look like that
Download the Skill: https://drive.google.com/file/d/1wG4QWTPleYTDrnLnrJLIAFcZh4lBCLQX/view?usp=sharing
How to install the Skill in Claude Code
-
Download the Skill from the Google Drive link above and unzip it (if it’s a .zip file)
-
Copy the Skill folder into Claude Code’s skills directory:
- Available in all projects:
~/.claude/skills/ - For a single project only:
.claude/skills/inside that project folder
# Example on macOS/Linux mkdir -p ~/.claude/skills cp -r appsheet-builder ~/.claude/skills/(On Windows, the equivalent path is
C:\Users\<your-name>\.claude\skills\— the.claudefolder is hidden by default, so enable “Hidden items” in File Explorer to see it.) - Available in all projects:
-
Open Claude Code and run the
/skillscommand to verify — if you see the Skill’s name in the list, it’s installed successfully -
From there, just describe your business needs (e.g. “build an inventory management app”) — Claude will automatically activate the Skill and start interviewing you
Using Claude.ai (web) instead of Claude Code? Go to Customize → Skills, click the + button, and upload the Skill’s ZIP file — that’s it (Code execution must be enabled in Settings).
To see how I use this toolkit to build a real product, check out my demo video (still a bit rough ): 
https://www.youtube.com/watch?v=BavZh-dUK9k
Quick summary
| What you need | What to use |
|---|---|
| Review and optimize an existing app | AppSheet Analyzer |
| Export context so A.I understands your app | AppSheet Analyzer (JSON tab) |
| Q&A, fix formulas, build apps automatically | Assistant for AppSheet |
| Not sure where to start | AppSheet Skill (A.I interviews you → creates the database → builds) |
Both tools are completely free. I’d love to hear your feedback to make them better. If you run into bugs or have feature ideas, drop a comment below!
Huge thanks to the AppSheet community. 
— Hoa Data
Assistant for AppSheet is an independent product, not affiliated with Google/AppSheet.
