Choosing An AI Tool

Compare Codex, Claude, Gemini, CLI tools, apps, and IDE extensions so you can choose an AI coding tool for customizing Jotsu apps.


Jotsu is the app environment around your AI coding tool.

You start with a template and customize it. The template includes instructions that guide the AI tool's decisions.

Jotsu handles deployment.

There are many good choices: CLI, installed apps, IDE extensions. All work and the right choice is mostly personal preference.

Our Recommendation

Our preference is OpenAI Codex.

The main reason is speed.

Codex is fast at generating useful results.

That makes the build loop feel quick.

ToolWhy choose it
OpenAI CodexFast results and strong agentic coding workflows.
Anthropic Claude CodeProbably the most popular AI coding tool.
Google Gemini CLIEasiest to try first, with a generous free path.

All three work well with Jotsu apps.

If You Are New

If you do not have a tool yet, try Gemini first.

It is easy to get started and has a free plan. All you need is a (free) Gmail account.

You can change later or even run multiple tools together.

How To Decide

Choose Codex if you want speed.

Choose Claude if you want the most common choice.

Choose Gemini if you want the easiest free start.

Choose an IDE extension if you like working inside your editor.

Choose a CLI if you like working from the terminal.

Choose an installed app if you want a dedicated workspace.

Using Gemini CLI

Gemini CLI is a good first tool.

It is easy to try.

It works from your terminal.

Run it without installing globally:

npx @google/gemini-cli

Or install it globally:

npm install -g @google/gemini-cli

Then open your Jotsu app project in a terminal.

Start Gemini:

gemini

Ask it to inspect the project first.

Then ask for one small change.

First Prompt

After setup, ask the tool to inspect the app.

Useful first prompt:

Read this project and explain the main parts of the app.

Then ask for a small change.

Useful second prompt:

Change the dashboard to describe what this app should do.

Small steps work better than giant prompts.