#3333rd of 35 in Agent platforms

ufo

GUI agent for Windows, plus a multi-device DAG orchestrator for agents

Stars
10.1k
License
MIT
Last commit
Oct 2026
Last release
Sep 2026
Language
Python
Stars, last 29 weeks+1,839
31 Mar 2026: 8,244 stars7 Apr 2026: 8,332 stars14 Apr 2026: 8,418 stars21 Apr 2026: 8,439 stars28 Apr 2026: 8,458 stars5 May 2026: 8,486 stars12 May 2026: 8,560 stars19 May 2026: 8,633 stars26 May 2026: 8,713 stars2 Jun 2026: 8,791 stars9 Jun 2026: 8,877 stars16 Jun 2026: 8,964 stars23 Jun 2026: 9,053 stars30 Jun 2026: 9,130 stars7 Jul 2026: 9,213 stars14 Jul 2026: 9,232 stars21 Jul 2026: 9,250 stars28 Jul 2026: 9,301 stars4 Aug 2026: 9,350 stars11 Aug 2026: 9,410 stars18 Aug 2026: 9,475 stars25 Aug 2026: 9,517 stars1 Sept 2026: 9,584 stars8 Sept 2026: 9,640 stars15 Sept 2026: 9,709 stars22 Sept 2026: 9,785 stars29 Sept 2026: 9,858 stars6 Oct 2026: 9,927 stars13 Oct 2026: 10,005 stars20 Oct 2026: 10,083 stars

8,244 on 31 Mar 202610,083 on 20 Oct 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

UFO² drives Windows apps through UI Automation, Win32 and COM, mixing GUI clicks with API calls, using an LLM such as OpenAI or Azure OpenAI. UFO³ Galaxy adds a ConstellationAgent that splits a request into a task DAG and runs it in parallel across Windows, Linux and Android device agents over a WebSocket protocol (AIP).

Who it is for: Engineers automating Windows apps or orchestrating agents across devices

Strengths

  • Hybrid UIA/Win32/COM control with GUI fallback for Windows automation
  • Galaxy decomposes tasks into DAGs and runs them in parallel across devices
  • Speculative multi-action batching reported to cut LLM calls by 51%
  • Backed by published papers and online docs; MIT license

Weaknesses

  • UFO² is Windows-only; Galaxy is marked active development with moderate setup
  • Needs a hosted LLM API key such as gpt-4o; local model setup not described
  • Galaxy setup needs per-device servers and clients plus YAML configs
  • No Docker or compose files; setup is pip install and manual config

What it needs

  • no GPU
  • Needs Python 3.10 or 3.11, LLM API key (OpenAI or Azure OpenAI)
  • Models: OpenAI, Azure OpenAI, Qwen, Gemini, Claude

Also in Agent platforms

See all 35
Also in Agent platforms
RankProjectScore
1LangflowVisual builder for AI agents and workflows, exposed as API or MCP server155.9k starsMIT81 out of 100
2daily_stock_analysisDaily LLM stock analysis for A-share, HK, US and other markets, pushed to chat apps66.8k starsMIT80 out of 100
3deer-flowAgent harness that runs sub-agents, memory and sandboxed skills for long tasks84.2k starsMIT77 out of 100
#4vibe-tradingFinance research agent for backtests, market data and broker connectors36.1k starsMIT77 out of 100
#5MulticaIssue board where AI coding agents take assignments like teammates53.4k starscustom license76 out of 100