streamChat

streamChat is the chat method on the createAgent(...) result.

Recommended usage

Custom route mode (`stream` flag)

import {
  collectBufferedStream,
  collectStream,
  parseChatRequestBody,
  toSSE,
} from "kortyx";

export async function POST(request: Request): Promise<Response> {
  const body = parseChatRequestBody(await request.json());

  const stream = await agent.streamChat(body.messages, {
    // optional
    sessionId: body.sessionId,
    workflowId: body.workflowId,
  });

  if (body.stream === false) {
    // Convenience shape: { chunks, text, structured }
    const buffered = await collectBufferedStream(stream);
    return Response.json(buffered);
  }

  return toSSE(stream);
}

When stream is false, buffered JSON includes:

chunks: full raw chunk list
text: merged assistant text (text-delta first, message fallback)
structured: collected structured-data chunks

Buffered helpers (choose one)

Good to know: collectBufferedStream(...) is built on top of collectStream(...), so you still get the full raw chunk list in buffered.chunks.

Parameters

messages: full chat history for the current turn (required).
options.sessionId: optional conversation id for analytics/tracing correlation.
options.workflowId: force this request to start in a specific workflow.

Returns

agent.streamChat(...): Promise<AsyncIterable<StreamChunk>>

What happens under the hood

resolves runtime config and session context
chooses the workflow (including request-level workflow override)
builds initial state from input + prior messages
resumes interrupted flows when resume metadata is present
runs the workflow graph and streams chunks as SSE
returns an async stream of StreamChunk events

Workflow override per request

When not resuming, options.workflowId can override entry workflow for this request.

Resume behavior

If the last message contains resume metadata, agent.streamChat(...) resumes the interrupted flow automatically.