OpenWork

OpenWork is an open-source web app and server that enables users to run AI agents on their own local files. It is available as a desktop application for macOS, Windows, and Linux, and can also be deployed as a server or via a CLI host. The software allows users to bring their own provider keys for over 50 different large language models, ensuring that the user maintains control over their AI provider and API costs.

The application supports both host and client modes to accommodate different infrastructure needs. In host mode, OpenWork runs the OpenCode stack locally on the user's machine to execute agentic workflows directly on local data. In client mode, the application connects to an existing OpenCode server via a URL. This flexibility allows users to start with a local setup on localhost and explicitly opt into remote sharing or server-based orchestration when collaborating with a team.

Key features

• Support for 50+ LLMs using custom provider keys
• Integration with MCP servers, plugins, and custom skills
• Local-first execution with an option to connect to remote servers
• Real-time updates via SSE live streaming
• Visual execution plans that render tasks as a timeline
• Permission system to allow, deny, or permanently authorize privileged flows
• Local storage for saving and re-running common workflow templates
• Skills manager for importing and listing local skill folders
• Debug exports for runtime reports and developer log streams

Built on Tauri and powered by OpenCode, the software is designed for users who require auditable and permissioned AI workflows. It includes a CLI host called OpenWork Orchestrator for those who prefer to run the server without a graphical user interface. The architecture separates the UI from the runtime, allowing it to function as a standalone desktop app or a connector for Slack and Telegram. It manages plugins through a project or global scope, reading and writing to an opencode.json configuration file to extend agent capabilities.

OpenWork is an extensible tool for turning agentic workflows into repeatable and productized processes for teams.