No coding experience required to understand this page.
Agentic Loop is a tool that lets you describe what you want to build in plain English, and an AI builds it for you—automatically.
Think of it like having a junior developer who:
- Asks clarifying questions before starting
- Writes the code
- Tests their own work
- Fixes their mistakes
- Keeps going until it's done
You describe the feature. The AI does the rest.
You type something like:
"Add a login page with email and password"
The AI (Claude) asks follow-up questions:
- "Should users be able to reset their password?"
- "Do you want a 'remember me' checkbox?"
- "Should it show error messages for wrong passwords?"
Based on your answers, Claude creates a checklist of small tasks. Each task is something that can be built and tested independently.
For example, a login feature might become:
- Create the login form
- Add password validation
- Connect to the database
- Show error messages
- Add "forgot password" link
This checklist is saved in a file called a "PRD" (Product Requirements Document).
A program called "Ralph" takes over. For each task on the checklist:
- Read - Ralph reads what needs to be built
- Build - Ralph tells Claude to write the code
- Check - Ralph runs automated tests to verify it works
- Save - If tests pass, Ralph saves the work
- Repeat - Move to the next task
If something fails, Ralph shows Claude the error and asks it to try again. This loop continues until everything works.
When Ralph finishes, you have working code. You can:
- Test it yourself
- Ask for changes
- Add more features the same way
- You describe what you want to an AI
- AI gives you code
- You copy-paste it
- It doesn't work
- You go back and forth fixing errors
- Repeat for hours
- You describe what you want
- Go do something else
- Come back to working code
The AI handles the tedious back-and-forth of fixing errors, running tests, and making sure everything works together.
- A computer (Mac, Windows, or Linux)
- Node.js installed (version 18 or newer)
- A Claude account with Claude Code access
- Basic familiarity with using the terminal/command line
- A code editor like VS Code
When you're ready to set up and run your first loop, follow the step-by-step walkthrough in Getting Started. It covers installation, setup, and the full two-terminal workflow.
Ralph automatically retries when things fail. It shows Claude the error and asks it to fix the problem. Most issues resolve themselves within a few attempts.
If Ralph gets stuck on the same error repeatedly, it will skip that task and move on. You can review what went wrong later.
Yes. Press Ctrl+C in the Ralph window to stop. Your progress is saved—you can restart anytime and Ralph will continue where it left off.
You can:
- Ask Claude to change it (in Window 1)
- Edit the code yourself
- Start over with a new
/prd
Agentic Loop includes safety checks:
- Won't commit passwords or API keys
- Runs tests before saving changes
- Creates backups before modifying files
To use the basic features, no. To customize the results or fix edge cases, some coding knowledge helps.
| Term | Plain English |
|---|---|
| Terminal | The text-based app where you type commands |
| Claude | The AI that writes code |
| Ralph | The program that runs Claude in a loop |
| PRD | The checklist of tasks to build |
| Story | One task on the checklist |
| Commit | Saving your work (like hitting "Save" but for code) |
| Loop | Doing something repeatedly until it works |
- Follow the full walkthrough: Getting Started — installation, setup, and your first loop run
- Take the tour: Run
/tourin Claude for a guided walkthrough - Explore customization: CUSTOMIZATION.md explains how to personalize your setup
- Something not working? Check Troubleshooting
- Want to learn more? See How Ralph Works
- Have feedback? Open an issue on GitHub