When one LLM agent isn’t enough, spin up a galaxy

What it does UFO³ bundles two systems: UFO², a mature Windows automation engine that drives the OS through native UIA, Win32, and WinCOM controls plus hybrid GUI and API actions; and Galaxy, a new orchestration framework that breaks user requests into device-level DAGs and dispatches them to agents running on Windows, Linux, or Android. A central planner rewrites the task graph on the fly as results come in, while a task runner handles device matching, locking, and cross-device dataflow. UFO² can run solo or drop in as the Windows worker node under Galaxy.

The interesting bit Galaxy treats a heterogeneous fleet like a compute cluster: it decomposes requests into a DAG of interdependent subtasks, dynamically evolves the graph based on execution feedback, and coordinates agents over a WebSocket-based protocol with automatic reconnection. Meanwhile, the UFO² side attempts to curb LLM costs with speculative multi-action batching, which the project claims cuts LLM calls by 51%.

Key highlights

• Decomposes requests into executable DAGs with cross-device dependency tracking and parallel execution across the fleet.
• Supports heterogeneous device pools—Windows desktops, Linux servers, and Android devices—with capability-based auto-assignment and real-time resource monitoring.
• UFO² offers deep Windows OS integration and hybrid GUI-plus-API actions; it is labeled LTS and battle-tested.
• Galaxy coordinates agents via a WebSocket-based Unified Agent Interaction Protocol that includes fault tolerance and automatic reconnection.
• UFO²’s speculative multi-action feature batches predictions to reduce LLM calls by 51%.

Caveats

• Galaxy is in active development with moderate setup difficulty, while UFO² remains the stable, easy-setup LTS option.
• Cross-device support currently lists Windows, Linux, and Android, with the README noting more platforms are coming.
• The repository effectively contains two distinct systems—solo Windows automation and multi-device orchestration—so complexity and documentation depth vary sharply by path.

Verdict Stick with UFO² if you need reliable, deep Windows automation on a single machine. Consider Galaxy if you are building a cross-device agent fleet and do not mind assembling a moderately complex, actively evolving orchestration layer.