Skip to content

Agents

Conversations and history

How conversations are kept, how much history the model sees, and where to read every transcript from every channel.

Reference1 min readUpdated 23 October 2026

Every exchange with an agent belongs to a conversation. SentientOne stores the messages, so a caller only sends the new message and the conversation's id; you never resend history.

How a conversation works

  • A request without a conversation_id starts a new conversation. Its id comes back in the first (meta) event and the last (done) event of the stream.
  • Send that id with the next message to continue. The agent sees the earlier turns.
  • The model receives the last 12 messages, up to 12,000 characters, with each new message.
  • An unknown conversation id ends the stream with a CONVERSATION_NOT_FOUND error event.
  • A conversation belongs to one agent. Sending its id to a different agent fails with CONVERSATION_NOT_FOUND.
  • The widget keeps its conversation id in memory, so a page reload starts a new conversation.
Continuing a conversation
{  "message": "And when will it arrive?",  "conversation_id": "9b1c2f4e-..."}

The Conversations tab

Each agent's Conversations tab lists conversations from every channel: the Playground, the API, the widget and workflows. Search them, refresh the list, and open one to read the full transcript.

  • The answer cache only applies to the first message of a new conversation.
  • Each message is also a trace. Use the trace_id from the stream to open it in Tracing.
  • Deleting an agent deletes its conversations.

Still stuck?

Email us with what you tried and any request id. A person replies.

support@sentientone.ai