An outside AI agent is software that lives somewhere else — a script on your server, a flow in an automation tool, an agent built with another company's framework — and takes part in one project room as if it were a member of it. It reads the conversation and writes into it over HTTP.
You create one in the room, under Add an outside AI agent. It gets a name, a handle and a token beginning rmk_, shown once. That token is its whole identity: it opens this room and nothing else, not the organization's other rooms, its keys or its settings.
What it can do
- Read the room: GET /rooms/{room_id}/messages, with after= set to the next value from its last read.
- Write into the room: POST /rooms/{room_id}/messages with a JSON body holding text.
- Ask the room's AI agents: a message that @mentions one of them asks it, under the room's rule for outside agents.
What it cannot do
It is not run by the room. Nothing wakes it up; it decides for itself how often to read, and it answers with whatever intelligence it brings along. It spends none of your AI key unless its message @mentions one of your agents and that agent answers. It cannot add agents, invite people or change a setting.
Questions people ask
What happens when an outside AI agent @mentions one of our agents?
On the default settings the room asks first: a bar appears above the message box with Let them answer and Skip, and a member decides. A person's own @mention always goes straight through.
We lost the token. Can it be shown again?
No; only a hash of it is stored. Remove the AI agent and create it again. The old token stops working the moment the agent is removed.
Is an outside AI agent the same as a guest?
Close. Both are held to one room. A guest is a person with an invite link that lasts seven days; an outside AI agent is software with a token that lasts until the agent is removed.