FutureOS

Autonomous Agents 💻 Rust ⚖️ MIT 🟢 Actively maintained
52 stars

FutureOS is an open-source AI agent with a Rust core that runs across terminal, desktop, mobile, and chat apps. It solves the problem of managing long-running, multi-surface agent tasks with trust and control, requiring user approval for every tool call. It is for developers who want a local-first, model-flexible agent with durable execution.

FutureOS demo
🖼️ Screenshot from the project README

✨ Key features

  • Multi-interface: TUI, desktop, mobile, CLI, IM bots
  • Trust-first tool execution with approval gates and sandbox tiers
  • 3800+ models across 140+ providers, custom providers supported
  • Loop engineering for 24+ hour runs with goals, gates, monitors
  • 15+ built-in skills for image, PDF, web, browser, slides
  • Forkable sessions with clone and tree navigation

🎯 Use cases

  • Run long research tasks overnight and review results from phone
  • Manage files and shell commands with explicit approval
  • Use one agent across terminal, desktop, and chat apps
  • Build custom skills and model configurations for specific workflows

📦 Installation

🧰 Requirements: Runs on macOS, Linux, Windows; requires a model provider API key or FutureOS login.

curl -fsSL https://dl.future-os.cn/install.sh | bash

Windows PowerShell:

iex (irm https://dl.future-os.cn/install.ps1)

Installers run future init.

🚀 Usage

future config   # configure a model provider
future agent    # start the agent
future tui      # launch terminal UI

For custom providers, edit ~/.future/agent/models.json.

❓ FAQ

How do I install FutureOS?

Run the one-line install script for macOS/Linux or the PowerShell command for Windows, which auto-detects your platform and installs the appropriate package.

What models can I use?

FutureOS supports 3800+ built-in models from 140+ providers, and you can add custom providers via a models.json file.

How does tool approval work?

Every tool call (read, write, edit, shell) is gated by your approval by default, with sandbox tiers (off/manual/sandbox) and OS-level sandboxing on macOS and Windows.

Can I run long tasks?

Yes, the loop control plane supports durable goals, event-sourced state, and verification gates for runs lasting 24+ hours.

📊 Repository

Stars★ 52
Forks🍴 6
Open issues🐛 0
Last commit🕒 Sep 4, 2026
Created📅 May 2026
Language💻 Rust
License⚖️ MIT

🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.