OpenWork: Open-Source Local Alternative to Claude Cowork with MCP and Skill Sharing

As desktop-native AI agent workspaces such as Anthropic's Claude Cowork and tools like Codex gain momentum for file-driven workflows, OpenWork gained widespread community attention on October 7, 2026, offering a local-first, open-source alternative. Developed by open-source team different-ai and hosted on GitHub, OpenWork breaks free from closed single-vendor lock-in, enabling developers to run autonomous agents directly on their local machines with support for over 50 foundation model providers and local Ollama instances.

Image source: different-ai (GitHub)

While traditional terminal-centric coding agents often present steep learning curves for broader teams, OpenWork delivers a cross-platform desktop application supporting macOS, Windows, and Linux. Internally powered by the open-source OpenCode runtime engine (opencode.ai), the tool builds upon established agent capabilities while introducing native solutions for sharing skills, plugins, and Model Context Protocol (MCP) servers across teams, tools, and machines.

Local-First File Workflows and Multi-LLM Support with Ollama

At the core of OpenWork's design is a local-first philosophy where AI agents perform real work directly on user files on their computer.

Agents operate directly on local files, and users can launch an instance on their local machine with a single click.

• Support for 50+ Model Providers: Users can connect their own API keys (BYOK) across more than 50 LLM providers to power agent tasks.
• Local Model Support via Ollama: For offline workflows or privacy-sensitive tasks, OpenWork connects directly to local models via Ollama without requiring external cloud API calls.
• Powered by the OpenCode Engine: Built on the open-source OpenCode runtime (opencode.ai), OpenWork inherits OpenCode's capabilities so that everything OpenCode can do works in OpenWork, even features that do not yet have a dedicated desktop UI.

Composable Architecture: Desktop App, CLI Orchestrator, and Messaging Connectors

OpenWork is engineered beyond a standalone desktop interface, offering a flexible, composable architecture without vendor lock-in:

1. Desktop Application: Provides a dedicated, cross-platform graphical workspace across macOS, Windows, and Linux for users who prefer an interactive visual interface.
2. OpenWork Orchestrator and Server Mode: Functions as a headless CLI host and server, allowing users to spin up an instantly shareable instance with a single CLI or desktop command.
3. Messaging Connectors (WhatsApp, Slack, Telegram): Allows teams to compose the desktop app, server, and messaging connectors (WhatsApp, Slack, and Telegram) to fit their operational needs.
4. Remote Worker Connections: Enables connecting to hosted workers through a straightforward user flow (Add a worker -> Connect remote), managing local and remote agent workflows from a unified interface.

OpenWork MCP Bridge, Cross-Agent Sharing, and Admin Management

A key highlight of the platform is the OpenWork MCP bridge, which extends OpenWork assets into existing third-party coding agents.

By adding one OpenWork MCP to compatible agents such as Codex, Claude Code, Cursor, or others, users can reuse the same skills, MCPs, and connected services across tools, teammates, and machines. The desktop app is available when a dedicated workspace is desired, but it is not required—users can leverage OpenWork capabilities directly from the agents they already use.

For larger organizations, an admin interface enables publishing capabilities and managing access centrally across teams.

Open-Source Adoption and Considerations

OpenWork's desktop application and core platform are distributed under the open-source MIT license, available on GitHub for local machine workflows without requiring an account or subscription. However, organizations should note that OpenWork Den, the enterprise control plane, operates under a source-available license (OpenWork EE License) where production use requires a subscription.

As an early open-source release, developers should note that certain capabilities currently operate through the OpenCode CLI and server layers rather than the graphical interface, and multi-user team workflows require configuring remote worker connections.

For teams seeking an extensible AI coworker that protects local file privacy while supporting 50+ cloud LLMs, local Ollama models, and cross-agent MCP sharing, OpenWork represents a compelling open-source alternative.

Sources

• GitHub Repository: different-ai/openwork
• Original Announcement on X (@RoundtableSpace): OpenWork Open-Source Release