Run Conversation Turn (Stream)
Send a message to the agent and stream the reply over Server-Sent Events.
connected event arrives at once, then task_created with the conversation record (its id is the conversation_id, its input your message), turn_started with the turn number, the reply text as chunk events as the agent writes it, each skill call as tool_call_request and tool_call_result, and finally turn_finished with the full reply in result. A task_finished event may arrive just before or after turn_finished; settle on turn_finished.
To continue the conversation, pass id = your conversation_id; omit it to start a new one. If the connection drops, reconnect with the Last-Event-ID header set to the last event id you received and the stream resumes after it.
The sub_tasks query parameter sets how sub-tasks the agent creates for other agents run inside this turn: sync inline, async in the background, auto per the agent’s own choice. The setting sticks to the conversation until you change it.
See the Conversations overview for the full event order and the conversation record.Authorizations
API Key for authentication
Path Parameters
Query Parameters
Agent version to run.
How the agent's own sub-tasks run inside this turn (not whether this API call blocks): 'sync' runs every sub-task inline, 'async' queues every sub-task, 'auto' restores the per-sub-task decision. The mode sticks to the conversation until changed.
sync, async, auto Body
text, markdown, json, voice default, harder Per-execution override for the LLM provider (e.g. 'openai', 'anthropic'). Falls back to the agent's configured provider when unset.
Per-execution override for the LLM model name. Falls back to the agent's configured model when unset.
Per-execution override for reasoning effort on reasoning-capable models.
low, medium, high, xhigh Surface that created this execution (mirrored to AgentExecutionHistory.source_node_type). Falls back to SourceNodeType.SDK at persist time when unset.
workbench, sdk, task, assistant, webhook, mcp, a2a, telegram, slack, email, api Forces gateway sub-task dispatch: 'sync' runs all downstream sub-tasks inline, 'async' queues them all, 'auto' restores the computed per-sub-task decision.
sync, async, auto Response
Successful Response

