⚠️ Heads up: this tool's official website appears to be unreachable right now. It may be offline, rebranded, or temporarily down.
OS-Copilot

OS-Copilot

★ 3.8 🔖 1200 saves 💰 Free ChatbotsGitHub

OS-Copilot is an open-source framework for building generalist computer agents that can operate an entire operating system. It has 1,794 stars on GitHub and was last updated in 2024. On top of the framework, the team built FRIDAY, a self-improving AI assistant that carries out everyday computer tasks on its own.

Overview

Most AI agents live inside a single app. OS-Copilot aims higher: its agents interface with the web, code terminals, files, multimedia, and third-party applications through one unified layer, planning multi-step tasks and learning from their own execution. The accompanying research paper, presented at the ICLR 2024 LLM Agents workshop, describes the self-improvement loop that lets FRIDAY get better at tasks over time. A front-end interface and a vision-equipped version of FRIDAY were added during 2024. Under the hood, the system combines a planner that decomposes goals into steps, an actor that executes those steps through OS-level tools, and mechanisms for judging progress, which together enable the self-improvement behavior. The repository includes example tasks so new users can see complete agent runs before building their own.

Key Features

  • Unified agent framework spanning web, terminal, files, multimedia, and desktop apps
  • FRIDAY assistant demonstrating self-improving, general-purpose computer control
  • Vision support so agents can act on what they see on screen
  • Planner and actor components for decomposing and executing multi-step tasks
  • Front-end interface for interacting with the agent
  • Open research codebase accompanying the OS-Copilot paper (arXiv:2402.07456)

💳 Pricing

OS-Copilot is free and open source under the MIT license. Running it requires your own LLM API access, which you pay for directly with the model provider.

Who Is It For

OS-Copilot is for agent researchers and developers prototyping computer-use assistants, and for anyone studying how self-improving agents can automate cross-application desktop workflows. The codebase is also a useful reference for how planner, actor, and grounding components fit together in a working system.

🚀 Visit OS-Copilot Official Site

🚩 Report an issue