Create a new run and stream its execution via SSE.
Returns a text/event-stream response with Server-Sent Events. Each
event has a type field (e.g. values, updates, messages,
metadata, end) and a JSON data payload.
Set on_disconnect to "continue" if the run should keep executing
after the client disconnects (default is "cancel"). Use stream_mode
to control which event types are emitted.
Request model for creating runs
Assistant to execute
Input data for the run. Optional when resuming from a checkpoint.
Execution config
Execution context
Checkpoint configuration (e.g., {'checkpoint_id': '...', 'checkpoint_ns': ''})
Enable streaming response
Requested stream mode(s)
Behavior on client disconnect: 'cancel' (default) or 'continue'.
Behavior after stateless run completes: 'delete' (default) removes the ephemeral thread, 'keep' preserves it.
delete, keep Strategy for handling concurrent runs on same thread: 'reject', 'interrupt', 'rollback', or 'enqueue'.
Command for resuming interrupted runs with state updates or navigation
Nodes to interrupt immediately before they get executed. Use '*' for all nodes.
Nodes to interrupt immediately after they get executed. Use '*' for all nodes.
Whether to include subgraph events in streaming. When True, includes events from all subgraphs. When False (default when None), excludes subgraph events. Defaults to False for backwards compatibility.
Request metadata (e.g., from_studio flag)
Server-Sent Events stream