The Autonomous AI Coding Loop
Run AI agents on your tasks until done. Give it a PRD, a task list, or GitHub issues. Ralphy works through them one by one—or in parallel using git worktrees.
Supports Claude Code, OpenCode, Codex, Cursor, Qwen-Code, and Factory Droid. Configure project rules so the AI follows your conventions.
*Or clone from GitHub to use the bash script.
How it works
Task sources. Define tasks in Markdown PRDs, YAML files, or pull directly from GitHub Issues. Ralphy works through each task sequentially or in parallel.
Branch per task. Each task runs on its own git branch. Changes are auto-committed, and PRs are created when complete.
Parallel worktrees. Run multiple AI agents simultaneously using git worktrees. Each agent works in isolation—no conflicts.
Project rules. Define rules and boundaries in .ralphy/. The AI follows your conventions: "use TypeScript", "never modify migrations", etc.
Tests and linting. Tests are written and run by default. Linting runs before completing tasks. Skip with --skip-tests, --skip-lint, or --fast.
Retry until done. Automatic retries with exponential backoff. Tasks keep running until they succeed or hit the retry limit.
Supported Engines
--claudeClaude Code(default)--opencodeOpenCode--cursorCursor--codexCodex--qwenQwen-Code--droidFactory Droid
Usage
Single task
ralphy "add dark mode toggle"
ralphy "fix the auth bug" --cursorTask list
ralphy --prd PRD.md
ralphy --github owner/repo --label "ai-task"Parallel
ralphy --prd PRD.md --parallel --max-parallel 5Config
ralphy --init
ralphy --add-rule "use TypeScript strict mode"