UFO³: Weaving the Digital Agent Galaxy

UFO³: Weaving the Digital Agent Galaxy

From Single Device Agent to Multi-Device Galaxy

Introduction: A Journey from a Single Agent to a Digital Cosmos

In the evolving world of intelligent automation, UFO³ stands as a bold evolution of its predecessors. What began as a focused UI-facing Windows agent—UFO—grew into UFO², a robust Desktop AgentOS designed for Windows automation with deep OS integration and hybrid GUI+API actions. Now UFO³ expands the horizon further: a Galaxy of connected devices and heterogeneous platforms, united by a declarative, dynamic, and safe orchestration framework. This is not just a software suite; it’s a design philosophy for cross-device, cross-platform automation where tasks are decomposed into intelligent graphs, adapted on the fly, and executed with concurrency guarantees.

Poster preview: See UFO³ Galaxy in Action

Choosing Your Path: Galaxy or UFO²

UFO³ introduces a dual-path paradigm that lets you pick the model that fits your needs, while keeping room to grow.

• Galaxy: Multi-Device Orchestration

• Perfect for: cross-device collaboration, complex multi-step automation, DAG-based task orchestration, and heterogeneous platform integration.

• Key features:

  • Constellation: Decomposition of work into executable DAGs, enabling structured task stars and dependencies that can be scheduled and rewritten at runtime.
  • Dynamic DAG editing: The workflow evolves as execution feedback arrives, allowing adaptive changes without stopping ongoing tasks.
  • Asynchronous execution: Parallel task coordination across devices, maximizing throughput while preserving correctness.
  • Unified AIP protocol: A secure WebSocket-based coordination layer that supports fault tolerance and automatic reconnection.
  • Template-driven MCP-empowered device agents: A lightweight toolkit for rapid agent development with tool augmentation via MCP.

• Quick Start: Galaxy is the gateway to cross-device orchestration and a natural successor for teams aiming to orchestrate tasks across Windows, Linux, Android, and more.

• UFO² Desktop AgentOS: Windows-focused Foundation

• Perfect for: single-Windows automation, quick task execution, learning agent basics, and simple workflows.

• Key features:

  • Deep Windows OS integration: Windows UIA, Win32, and WinCOM-native control support.
  • Hybrid actions: A blend of GUI interactions and API calls for robust performance.
  • Proven reliability: A mature, battle-tested environment for Windows automation.
  • Easy setup: Quick onboarding suitable for learning and rapid deployment.
  • Can serve as a Galaxy device agent: A stepping stone to cross-device orchestration when you’re ready to scale.

• Quick Start: UFO² remains a stable, dependable choice for Windows automation with a clear path toward Galaxy augmentation.

See UFO³ Galaxy in Action

Witness how the Galaxy orchestrates complex workflows across devices and platforms. The demonstration video illustrates cross-device task orchestration in a visual, intuitive way, showcasing how a ConstellationAgent decomposes requests into DAGs, assigns tasks to diverse devices, and harmonizes execution to deliver results.

Caption: Cross-device task orchestration in UFO³ Galaxy

What’s New in UFO³? Evolution Timeline UFO³ is built on a clear, incremental trajectory that marries research rigor with practical engineering. Here’s the arc of its evolution:

• February 2024: UFO GUI Agent for Windows debuts. A UI-focused agent designed to automate Windows interactions from a single device.
• April 2025: UFO² Desktop AgentOS arrives. This release formalizes a robust architecture for Windows automation, with a strong emphasis on reliability and extensibility.
• November 2025: UFO³ Galaxy launches. The multi-device orchestration framework introduces Galaxy—the multi-device orchestration layer that coordinates heterogeneous agents across platforms.

This evolution is underpinned by several core principles that define UFO³’s philosophy and capabilities.

UFO³ Galaxy: Five Core Design Principles

UFO³ combines five tightly integrated principles to deliver a resilient, flexible, and scalable ecosystem for intelligent automation:

1) Declarative Decomposition into Dynamic DAG

• Requests are transformed into structured DAGs composed of TaskStars and dependencies.
• This enables automated scheduling, runtime rewriting, and dynamic adaptation as execution unfolds.
• The result is a clearer separation between planning and execution, with a living graph that can evolve.

2) Continuous Result-Driven Graph Evolution

• The constellation adapts in response to execution feedback.
• Controlled rewrites and dynamic adjustments ensure the plan stays aligned with actual outcomes.
• This feedback loop enables improvements on the fly and reduces divergence between intention and reality.

3) Heterogeneous, Asynchronous & Safe Orchestration

• Capability-based device matching selects the best device for each task.
• Asynchronous execution and safe locking preserve correctness while maximizing parallelism.
• Formally verified correctness guarantees help maintain reliability in complex cross-device scenarios.

4) Unified Agent Interaction Protocol (AIP)

• WebSocket-based, secure coordination layer for all agents.
• Built-in fault tolerance and automatic reconnection preserve continuity even in unstable environments.

5) Template-Driven MCP-Empowered Device Agents

• Lightweight, template-driven toolkit to create and augment device agents quickly.
• MCP integration enables tool augmentation, enabling agents to perform more capable actions with minimal code changes.

A Snapshot: How the Galaxy Framework Stands Apart

• Constellation Planning: The heart of cross-device orchestration, enabling cross-device dependency tracking, parallel execution optimization, and dataflow management across Windows, Linux, Mac, and more.
• Device Assignment: Smart matching that considers platform, resources, task requirements, and historical performance to auto-assign tasks to optimal devices with potential for dynamic reallocation.
• Orchestration: Real-time status updates, automatic error recovery, and end-to-end progress tracking with feedback, all within a cross-device constellation.

UFO² Desktop AgentOS – Core Strengths and How It Fits Into Galaxy

UFO² continues to play a central role within the UFO³ ecosystem, acting as both a standalone Windows automation engine and a capable Galaxy device agent. Its strengths are anchored in deep OS integration and a practical balance between GUI actions and programmatic control.

• Deep OS Integration: Harnessing Windows UIA, Win32, and WinCOM-native controls to interact with the operating system at a fundamental level.
• Hybrid Actions: A blend of GUI interactions and API calls to optimize performance and reliability.
• Speculative Multi-Action: Batch predictions can reduce LLM calls by approximately 51%, boosting efficiency in action planning.
• Visual + UIA Detection: Robust hybrid control detection to improve resilience when automating complex UI scenarios.
• Knowledge Substrate: A knowledge base (RAG with docs, demos, and execution traces) to support intelligent agents and traceable decisions.
• Device Agent Role: Capable of serving as the Windows executor within the Galaxy orchestration, enabling it to participate in cross-device workflows.

Migration Path: UFO² to Galaxy

For teams already invested in UFO², a pragmatic migration path keeps the transition practical and risk-managed:

• Phase 1: Learn — Keep UFO² as your core Windows automation tool while you explore Galaxy concepts.
• Phase 2: Experiment — Let Galaxy control a subset of tasks that don’t require critical reliability, validating cross-device capabilities.
• Phase 3: Hybrid — Use both frameworks, with Galaxy orchestrating complex pipelines and UFO² handling Windows-specific tasks.
• Phase 4: Migrate — Gradually move more complex tasks to Galaxy, while continuing to rely on UFO² where it remains the most efficient solution.
• Resources: A Migration Guide is available to help teams transition with minimal disruption.

Capabilities at a Glance: A Deep Dive into the Galaxy Framework

What makes Galaxy different is the way it organizes planning, device matching, and orchestration into a cohesive, cross-device system.

• Constellation Planning: A request is decomposed into a Task DAG that can span multiple devices (e.g., Windows, Linux, Mac). Benefits include cross-device dependency tracking, parallel execution optimization, and cross-device dataflow management.
• Device Assignment: The framework uses platform, resource availability, task requirements, and historical performance to automatically assign tasks to optimal devices. Real-time resource monitoring enables dynamic reallocation as conditions change.
• Orchestration: As tasks execute, status updates flow in real time, enabling automatic error recovery and precise progress tracking. The final report synthesizes results across devices into a coherent outcome.

UFO² Desktop AgentOS – The Windows-Centric Foundation

UFO² remains a robust choice for Windows automation, with deep OS integration and a stable path to Galaxy when multi-device orchestration is desired.

• Architecture and Task Model: UFO² uses a single-host agent model with a well-defined execution path for Windows tasks. It remains simple to set up and operate, making it ideal for newcomers and for straightforward automation.
• Execution and Scope: Focused on single-device automation with an emphasis on reliability and predictable results.
• Learning Curve: Moderate, especially for teams new to automation but wanting to quickly become productive on Windows.

Migration and Extensibility: Tailoring UFO³ to Your Needs

Both UFO² and Galaxy are designed to be extensible, so you can tailor the framework to your real-world requirements.

• UFO² Extensions: Custom actions and automators, custom knowledge sources (RAG), custom control detectors, and bespoke evaluation metrics.
• Galaxy Extensions: Custom agents, new device types, orchestration strategies, and visualization components to suit your organizational preferences.

Documentation and Learning Resources

A comprehensive documentation structure guides users from architecture overviews to hands-on tutorials:

• Galaxy Documentation: Architecture overview, start guides, client and orchestration components, and detailed technical references for AIP, session management, visualization, events, and more.
• UFO² Documentation: Overview, installation steps, configuration guides, LLM and RAG setup, and a thorough usage guide for multi-action and advanced features.
• Online Documentation: Complete UFO documentation, supported models, RAG configuration, and model support details.
• Quick Start Tutorials: Step-by-step tutorials for Galaxy and UFO², with platform-specific device setup guides (Windows, Linux, Android) and configuration guidance for device pools.

Latest Updates and Community

• 2025-11: UFO³ Galaxy Framework released. A major research breakthrough in multi-device orchestration, with declarative DAG decomposition, dynamic graph evolution, heterogeneous safe orchestration, unified AIP, and MCP-empowered agents.
• 2025-04: UFO² v2.0.0 released. Desktop AgentOS with enhanced architecture and LTS status.
• 2024-02: Original UFO release. UI-focused agent for Windows with broad media coverage and adoption.

Community engagement and resources:

• Media coverage highlights: industry articles and discussions about UFO², cross-device automation, and Windows AgentOS.
• Community channels: GitHub Discussions and Issue Tracker for UFO, plus an official YouTube channel for demonstrations.
• Contact: [email protected] for collaboration, questions, and contributions.

Citations and Foundational Papers

If you leverage UFO³ Galaxy or UFO² in research or production, there are formal papers to reference:

• UFO³ Galaxy Framework (2025) arXiv:2511.11332
• UFO² Desktop AgentOS (2025) arXiv:2504.14603
• Original UFO (2024) arXiv:2402.07939

Related Projects and Research

UFO³ sits within a vibrant ecosystem of research and practical tooling:

• TaskWeaver: Code-first LLM agent framework for data analytics and task automation
• GUI Agent Research: Surveys and interactive sites exploring LLM-brained GUI agents
• Multi-Agent Systems: Foundation work in cross-device coordination, OS-level agent orchestration, and cross-application task evaluation benchmarks
• Benchmarks: Windows Agent Arena (WAA) and OSWorld for evaluating automation agents

FAQ: Quick Answers for Practical Decisions

• Galaxy or UFO²? Start with UFO² if you only need Windows automation and fast setup. Choose Galaxy for cross-device coordination and more complex workflows. A hybrid approach often delivers the best of both worlds.
• Will UFO² be deprecated? No. UFO² remains in Long-Term Support and continues to be actively maintained. Galaxy can use UFO² as a Windows device agent, enabling a gradual, scalable transition.
• Can Galaxy do everything UFO² does? Functionally yes. Practically, Galaxy enables cross-device orchestration, while UFO² remains simpler and more streamlined for Windows-only tasks.
• How mature is Galaxy? It is in active development, with a solid core architecture and DAG orchestration, while ongoing work focuses on advanced device types, monitoring, and extended documentation.
• Can I extend or customize? Both UFO² and Galaxy are designed for extensibility, with clear extension guides for actions, devices, orchestration strategies, and visualization components.
• How can I contribute? Contributions are welcome via bug reports, feature suggestions, documentation improvements, tests, and pull requests. See CONTRIBUTING.md for guidelines.

Ready to Get Started? Explore both paths and begin your journey into multi-device automation today.

• Galaxy Start: [Start Galaxy] badge

• UFO² Start: [Start UFO²] badge

Explore Galaxy: Multi-Device Orchestration

Try UFO²: Windows Desktop Agent

Documentation and Further Reading

• Galaxy Architecture & Concepts
• Galaxy Quick Start Tutorial
• Windows Device Setup (UFO²)
• Linux Device Setup
• Mobile Device Setup (Android)
• Configuration Guides for Galaxy and UFO²
• AIP Protocol, Session Management, Visualization, and Events/Observers

Images included from the Input

• UFO logo: assets/logo3.png
• Galaxy action poster: assets/posterwithplay.png
• UFO² logo (blue): assets/ufo_blue.png

Closing thoughts: The Dawn of a Digital Galaxy

UFO³ represents more than a new product layer; it embodies a shift toward a shared, dynamic, cross-device automation ecosystem. By decomposing work into DAGs, adapting in real time, and coordinating a heterogeneous fleet of agents through a secure, resilient protocol, UFO³ enables organizations to orchestrate complex workflows with clarity and confidence. Whether you start with the familiar Windows automation of UFO² or take the direct plunge into multi-device Galaxy orchestration, UFO³ provides a coherent, extensible framework to weave your own digital agent galaxy. The journey from a single agent to a constellation of agents across devices marks a new era in intelligent automation—one where your tools adapt to your goals, not the other way around.