Text chat SSE

Text chat uses a normal HTTP POST and streams assistant output through Server-Sent Events.

Create a text session

curl -X POST "https://api.hyponema.ai/sessions" \
  -H "Authorization: Bearer $HYPONEMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "agent_...",
    "user_id": "user_123",
    "modality": "text"
  }'

Send a message

curl -N -X POST "https://api.hyponema.ai/sessions/$SESSION_ID/messages" \
  -H "Authorization: Bearer $HYPONEMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Can you summarize our last conversation?"
  }'

The text field is required and must be between 1 and 8000 characters.

Stream events

The response emits Server-Sent Events. Full payload shapes live in the Chat API reference; the events you’ll see are:

Event	When it fires
turn_start	Assistant turn started. Carries turn_id and the resolved model.
token	Streamed text chunk.
tool_call	The model invoked a tool. Includes the streaming-JSON arguments_partial.
tool_result	Tool returned. Includes the final result the assistant sees.
guard	A guard fired (no-go zone, escalation, redirect).
turn_end	Assistant turn completed. Includes latency and token counts.
error	Turn failed. Carries detail and error_kind.

Runtime behavior

The chat runner streams tokens, executes tool calls inline, applies guards, and persists the assistant and tool turns through a fresh unit of work.

Do not build a second chat loop in your application. Treat the SSE stream as the source of assistant output and use traces for debugging.