A token is how a caller proves who it is, whether that is an AI agent or your own code. You create one, choose what it can do, and paste it into the client. Tokens belong to you, not to a single workspace.
Tokens live on your profile, so they follow you across every workspace you belong to.
Each tool requires at most one scope (the node-type reference needs none). Grant only the scopes your agent needs.
read:knowledge, write:knowledge, delete:knowledgeread:storage, write:storage, delete:storageread:flow, run:flow, write:flow, delete:flow, publish:flow, deploy:flowread:conversations, read:conversations_pii, debug:conversationsRead and run scopes suit an agent that answers questions and tries flows. Add write, publish, or deploy scopes only when the agent needs to change things. The Tool catalog lists the scope every tool requires.
Two scopes gate sensitive conversation data: read:conversations_pii reveals a contact's submitted email and phone, and debug:conversations exposes a run's raw step trace. Grant them deliberately.
By default a token can reach every workspace you belong to, and each call names the one it runs in. If you want a token pinned to specific workspaces, tick them when you create it. The list only ever offers workspaces you are a member of, and if your membership is removed the token loses that workspace with it.
Open the token from the list and choose Revoke. It stops working immediately. The row stays so you can see when it was last used, but the token can no longer authenticate. Revoke a token the moment it is no longer needed or might have leaked.