Ralph Orchestrator is an open-source harness that applies the “Ralph Wiggum technique” to autonomous software development: instead of chatting with a coding agent turn by turn, you give it a backlog and let it loop — plan, implement, test, review — until the work is done. This improved implementation orchestrates agents such as Claude Code, Codex, Gemini CLI, Kiro and OpenCode from a single controller. The project has about 3,200 stars on GitHub and is under active development in 2026.
The orchestrator treats agent sessions as an event-driven loop. It reads a spec or task list, spins up the chosen CLI agent with the right context and persona (“hats” that shape behavior for roles like planner, builder or reviewer), watches the results, and feeds failures back into the next iteration. Guardrails such as iteration limits, completion detection and decision gates keep the loop productive rather than runaway, and progress is tracked in a backlog the tool manages for you. Because it is backend-agnostic, you can switch between agent CLIs — or mix them by role — without changing your workflow. The result is a practical way to run long, mostly unattended coding sessions: overnight refactors, test-backfill campaigns and spec-driven feature builds.
Ralph Orchestrator is completely free and open source under the MIT license. You only pay for whatever AI agent subscriptions or API usage your chosen backend consumes.
Ralph Orchestrator is for developers experimenting with autonomous coding agents who want a structured, repeatable way to let agents grind through a backlog while they supervise at the level of specs and reviews.