Lounge connectors
Connect your apps to Lounge chats with OAuth
Open Connectors in the Lounge to connect an account through OAuth. Connections are private to your account, and tool calls require your approval before they run.
On iOS, open Connectors from the home screen. Sign-in opens in the system browser and returns to the app. You can search connections, pause them, reconnect, or disconnect from this screen.
In an iOS conversation, review each proposed tool and its arguments, then choose Approve or Decline. The response continues once all pending requests are answered. Saved conversations retain pending approvals and completed results across restarts; temporary conversations keep them only for the current session.
In an iOS model comparison, connected apps are available to Model 1. Approve or decline its pending requests before sending another message to all models. Other models keep separate conversations and can be retried without running connector actions. Pending requests and results survive restarts.
Saved approvals are claimed on the server before execution. Two devices answering the same saved request cannot run it twice, and a stale conversation save cannot reopen an answered request.
If a tool's result cannot be confirmed, check the connected app before asking to run it again. Continue response lets the model respond to that uncertainty without repeating the tool call.
A connector stays Not configured until the deployment has both its OAuth client ID and secret. Missing credentials also disable existing connections: you cannot enable, reconnect, or use their tools until configuration is restored. You can still disconnect an account.
Pausing a configured connection retains its credentials and stops tool access. Disconnecting deletes its stored credentials and pending authorizations. Tool results used in a chat are sent to the selected model and saved with that conversation.
Authorization must finish in the same session that started it and expires after ten minutes. If you sign out or the request expires during consent, start connecting again. Cancelling reconnection keeps the existing connection.
Client integrations
Authenticated clients can stream chat and signed tool proposals from POST /lounge/chat, using x-llmgateway-key to select the billing key. The endpoint accepts AI SDK UI messages and connector IDs. It does not execute tools or save conversations.
After the user answers a saved proposal, call POST /connectors/{connectorId}/tools/{toolName} with x-tool-message-id, x-tool-call-id, and x-tool-approved (true or false). Send the saved input unchanged. The server verifies ownership and the saved proposal, claims it, and returns the updated tools JSON; a decline never executes the connector. An already answered request, or an approved request for a paused or disconnected connector, returns 409 and nothing runs. If the connector fails after the claim, the request returns 502 and stays saved as unconfirmed. For temporary conversations, omit these headers and record the execution result locally before continuing. Send completed, failed, or denied tool parts with the next chat request. Pending approvals must be resolved first; never automatically retry a tool whose outcome is unknown.
For deployment variables and callback setup, see the connector setup guide.
How is this guide?