I posted an earlier version here about four months ago. Since then, we’ve focused on why sharing agents with teammates still sucks. Sharing markdown files and skills only gets you so far when everyone is working with different runtimes and context windows.

We’ve built a multiplayer workspace for spinning up agent sessions and sharing them with others. In the background as your team works, Oasis builds a permissioned memory graph for all your agents. This way, context built in one user’s session persists to another user’s session. We’ve seen hallucination rates drop pretty dramatically across all major models thanks to this shared context layer.

It’s still early, and we’re figuring out the right UX for multiplayer AI. It’s also free to use. Feedback is very welcome.