Ralphy is an autonomous bash script that orchestrates multiple AI coding agents in a continuous loop until a product requirements document (PRD) is fully implemented. Named after Ralph Wiggum, it takes a deceptively simple approach to multi-agent development: write a PRD, run the script, and let it cycle through Claude Code, Codex, OpenCode, Cursor Agent, Qwen, Factory Droid, GitHub Copilot, and Gemini CLI until every requirement is satisfied. Each agent processes the same task or PRD section independently, with Ralphy managing the loop, checking completion criteria, and routing work between engines based on configurable preferences and availability.
The tool uses a file-based configuration system where developers define project settings including language, framework, test and lint commands, AI behavior rules, and file boundaries that agents must never modify. Rules persist across all tasks in a project, ensuring consistent coding standards regardless of which AI engine handles a particular iteration. Single-task mode handles quick changes like adding a login button with a one-liner command, while PRD mode processes multi-page requirement documents section by section. The bash-native implementation means zero dependencies beyond the AI CLIs themselves — no Node.js, no Python, no build step.
Licensed under MIT with over 2,500 GitHub stars, Ralphy addresses the practical reality that different AI coding agents excel at different types of tasks. Rather than committing to a single agent, it leverages the diversity of available coding assistants by running them in sequence until convergence. The project is actively maintained with Claude Code as the default engine and expanding support for newer entrants. Its simplicity is its strength — the entire orchestration logic lives in a single ralphy.sh file, making it trivial to audit, customize, and extend for specific team workflows without understanding complex framework abstractions.