POST /v1/chat accepts:
{
"input": "Your request",
"threadId": "optional-stable-conversation-id"
}The response contains the final text and the thread identifier:
{
"output": "The agent response",
"threadId": "conversation-id"
}Request bodies are limited to one megabyte and prompt text is limited to 100,000 characters. Deployments exposed beyond loopback should add authentication, authorization, rate limits, and TLS at the edge.
POST /v1/chat/stream accepts the same body and returns server-sent events. text events carry model chunks, followed by one done event containing the thread identifier. Errors that occur after streaming begins are delivered as error events.