Human-in-control

#The approval gate

sSystm ships a built-in platform agent that can act inside your workspace. It is deliberately not an autopilot: every action it wants to take is staged as pending — prepared and visible, but not executed. A human approves or rejects, action by action.

Your own MCP-connected AI works differently: it acts directly in the surface you chose to expose (org-scoped, attributed). You control it by controlling its tokens and by choosing what the MCP surface contains.

#The Agent Log

Every run of the built-in agent is recorded in the Agent Log module — what ran, when, and what came of it. The execution history is part of the workspace, open to the whole team.