OpenCowork

OpenCowork is a local-first desktop runtime designed for teams that require reliable, repeatable execution of AI-driven workflows beyond simple chatbot interactions. It addresses the need for auditable, reusable automation by enabling users to operate browsers and local tools, connect with MCP (Modular Command Protocol) servers and skills, and transform successful task sessions into workflow templates. The platform is particularly tailored for builders, researchers, product teams, and early enterprise pilots seeking local execution, browser-first automation, and workflow reuse within a private deployment boundary.

Key features include browser automation with the ability to navigate, click, type, wait, search, and extract information in a headed browser, supporting workflows that demand real user interaction. OpenCowork integrates hybrid CUA (Custom User Automation) and visual execution fallback, providing approval-aware recovery, takeover paths, and persistent visual traces for robust workflow management. Users can record and preserve task states, results, screenshots, artifacts, and traces, ensuring that all work remains observable and auditable after execution. The system allows for the conversion of completed sessions into reusable workflow templates, which can be rerun, scheduled, or triggered from chat, IM, scheduler, or MCP-oriented flows.

The platform supports integration with local scripts, internal services, and MCP servers, exposing selected capabilities through a standard MCP endpoint. Skills and local outputs can be installed, updated, uninstalled, and executed, allowing for the handling of files, local commands, and asset generation outside chat sessions. OpenCowork also features session management, IM intake, and the ability to process file-driven tasks through Feishu messaging workflows. Its architecture is built on a shared local Agent Runtime baseline, compatible with Electron, Scheduler, IM, MCP, and future clients, all managed through a unified task lifecycle.

0 license. js, npm, Python, and Playwright Chromium, and the desktop application can be launched on macOS, Ubuntu/Debian, or Windows platforms. Its focus on local execution, workflow auditability, and template reuse positions it within the class of desktop AI agent runtimes for business process automation.