OpenWork: Make Agent Capabilities Easier to Share
Different AI’s OpenCode-based workspace combines a desktop app with shared skills and MCP connections. The useful test is whether a working setup transfers cleanly to the next person.
A useful agent setup often starts on one person’s machine. The harder part is making it useful to the next person without rebuilding every connection and explaining every instruction. That is what interests me about OpenWork.
Different AI’s OpenWork is built on OpenCode. Its desktop app runs on macOS, Windows, and Linux, supports multiple model providers and local models, and works with files in a workspace. Teams can share skills and MCP connections. Its MCP gateway also makes assigned capabilities available to compatible agents, so the desktop app is one way to use the system.
The organizational value I would test is reuse. Take one recurring task, package the instructions and connections, and give it to a colleague. Can they complete the same work with clear permissions and predictable results? That is a more useful adoption test than counting how many tools appear in a menu.
The licensing split matters: the desktop app and core outside ee/ are MIT-licensed, while the organization control plane has separate source-available terms. Local desktop use also does not mean every model call stays on the machine. Prompts go to the selected model provider, and hosted workers are optional. I would choose the deployment and provider together, then judge the product by how much setup knowledge it removes from the next person’s day.
