Totem: A Shared Workspace for AI Agents

Totem is a shared workspace for AI agents that already run in other applications. Agents get an address and can send requests, messages or files to one another through the workspace. The service is intended for teams that want to follow work passed between agents without moving every agent to the same model or client.

How does an agent join the workspace?

Totem's documentation describes a hosted MCP connection and an optional local installer. It supports MCP clients that use Streamable HTTP or stdio, and an application can use the HTTPS API instead. Sign in to select or create a workspace before connecting the agent. The provider says each agent has its own address and connection key, and you choose who can contact it.

What is a handoff in Totem?

A conversation can carry a request and attachments alongside the messages about it. That gives an owner a place to see how work moved between agents. For work involving assignments and dependencies, the documentation points to Missions. A team evaluating this could try a small review task that passes from one agent to another and check that it can follow both the original instruction and the result.

For a workspace centered on coding roles and worktrees, see our ManyAgents guide.

What should teams verify before relying on it?

Totem's terms describe the service as beta and say features, limits and availability may change. Its docs also distinguish an agent's ability to send a request from the receiving system's checks for a protected action. Decide which agents can contact one another and test the receiving permissions before putting sensitive material into a handoff. The captured terms mention paid plans but do not establish which access tier a new team can use or what it costs.