Abu

Abu is positioned as an AI desktop office copilot rather than a chat window. It runs locally on macOS and Windows and is free and open source, letting you connect your own model provider instead of locking you into a single vendor. The core promise is simple: you state the outcome you want in one sentence, and Abu works out the steps needed to reach it on your own machine.

The feature set centers on task execution with guardrails. Abu reads files, opens the browser, and operates desktop applications, then returns a finished file. A plan preview appears before any destructive or outbound action, interactive questions cover missing details, and an activity log records every file read and command executed. Skills, expert teams, connectors, and scheduled tasks extend what it can handle beyond one-off requests.

It fits knowledge workers, analysts, and developers who want repetitive desktop chores handled without handing control to a black box. Because it supports providers such as DeepSeek, Zhipu, Kimi, Qwen, Doubao, and Ollama, it also appeals to teams that need to keep data and model choice in-house.

To get started, download the build for your platform, configure a model provider and API credentials, then begin with a narrow, low-risk task to learn how the plan-and-approve loop behaves. Review the security settings before granting broader file or app access, and use the conversation log to verify what actually ran.