A hat-based orchestration framework that keeps AI agents in a loop until the task is done, built on Ralph Wiggum technology with backpressure execution gates.
Ralph Orchestrator is an autonomous AI agent orchestration framework that builds a continuous iteration layer on top of mainstream AI coding backends like Claude Code, Gemini CLI, and Codex. Its core design philosophy originates from Geoffrey Huntley's "Ralph Wiggum technology" — clearing context each iteration and re-reading specs and code (Fresh Context Is Reliability), rejecting substandard work through backpressure gates for tests, lint, and type checks (Backpressure Over Prescription), and using the filesystem and Git as the handoff mechanism between agents (Disk Is State, Git Is Memory).
The framework offers two execution modes: Traditional simple loops for quick tasks, and Hat-Based mode that coordinates multiple specialized roles through typed events for complex workflows. Built-in PDD (Prompt-Driven Development) planning enables interactive generation of requirements, design, and implementation plan documents via ralph plan. 31 presets cover scenarios including code-assist, debug, research, and review.
For human interaction, it provides a real-time TUI monitor built on ratatui, an Alpha-stage Web Dashboard, and RObot Human-in-the-Loop capabilities via Telegram (agents can block and ask questions, humans can proactively guide). It exposes capabilities through an MCP Server (stdio protocol) supporting parallel multi-workspace orchestration instances.
The core engine is built in Rust (7 Cargo workspace crates), supplemented by a TypeScript frontend and Python helper scripts, with three installation methods: npm, Cargo, and GitHub Releases.