# Maestro-Flow

An intent-driven workflow orchestration tool for multi-agent AI development that plans, executes and verifies coding tasks across Claude Code, Codex, Gemini and other agent backends, with a self-reinforcing knowledge graph.

Maestro-Flow orchestrates multiple coding-agent backends, such as Claude Code, Codex, Gemini, Qwen, OpenCode and Grok, around a single natural-language intent. Its Ralph engine reads the current project state, classifies the intent into one of more than 40 command chains, and decides at each checkpoint whether to continue, roll back, or insert a fix loop, without requiring a hand-written pipeline configuration.

Work can be split across four orchestration modes: delegating tasks asynchronously, having agents collaborate on distinct roles, running dependent tasks in parallel waves, or exploring several approaches at once. Patterns, pitfalls and decisions the agents discover are persisted as specs and know-how, and a hook system injects that accumulated knowledge into later agents' prompts.

A separate long-running mode called Odyssey runs autonomous loops over hours for large debugging, refactoring or interface-optimization tasks, adjusting strategy at each checkpoint until acceptance criteria are met. It installs through npm and needs Node.js 22.19 or later plus at least one supported host CLI.

Website: https://catlog22.github.io/maestro-flow/