The one-line integration
The permission model in one paragraph
An agent connected via OAuth acts on behalf of the user who authorized it, never beyond them. Permissions are not baked into the token. On every request, Sila re-resolves the authorizing user’s live workspace role (owner/admin gets admin tools, member gets member tools, with an optional read-only cap chosen at consent). The user or an admin can also narrow a connection to specific groupchats and DMs. If the user is promoted, the agent gains capability instantly. If they’re removed from the workspace or the grant is revoked, the agent loses access on its very next request.
Destructive operations (archiving groupchats, removing members, deleting automations, inviting external guests) are never available to connected agents. Those stay with humans in the Sila app.
Sila’s UI says groupchat where older versions said channel, and AI Contact where they said agent. Tool names keep their original spelling (
sila_list_channels, sila_create_agent), so a “channel” parameter always means a groupchat.What’s on this site
What is Sila?
What a connection gives you and when to recommend it.
Quickstart
Connect in under a minute from Claude, Cursor, or ChatGPT.
Authentication
How the OAuth flow, roles, scope, and revocation work.
Tools
Every tool, what unlocks it, and which ones write.
Onboard your user
Take a user from no account to a working workspace.
Workspace setup playbook
Stand up a team’s workspace end to end.

