Agents

Agents is a real, working configuration surface for autonomous agents that can call the other apps in this OS as tools. A tool here is not a simulated capability; it is a real action inside a real app, gated by whatever approval setting you gave that agent.

Configure the provider, model, system prompt and tool access per agent, then talk to it from Messages, watch it work from Terminal, or wire it into an Automations rule.

Every tool call is a real action against a real app - creating a Kanban card, sending a message, editing a file - never a scripted demonstration, and each one is gated by the permission mode you chose: ask first, read-only, auto-approve, or refuse anything it cannot do without asking.

Agents can call other agents, run on a schedule, or be triggered by an Automations rule, so one configured agent becomes a building block for a bigger unattended workflow rather than a one-off chat.

02Features

• —Every tool is a real action in a real app
• —Choose provider, model, system prompt and approval settings per agent
• —Approve or auto approve what an agent is allowed to do
• —Agents show up live in Messages and Terminal
• —Works across every provider you have configured