Content edit - Questra Program docs

Conversations edit programmed content with a survey-editor agent. Each user turn streams an AI SDK UIMessage response. Live chat uses the AI SDK UIMessage protocol, not the NDJSON progress stream.

The Program Review UI does not use this workflow as the document source of truth. Review keeps the Monaco buffer as document source of truth, runs coding-agent tools in the browser, persists chat via conversations, and writes tip revisions only on explicit Save. See Review chat.

Send a message

POST /v1/surveys/{surveyId}/conversations/{conversationId}/messages
Content-Type: application/json

{ /* AI SDK UIMessage */ }

Response: text/event-stream (UIMessage SSE)
Client disconnect aborts the in-flight turn for that request

Persistence and tip reconcile

Channel	Role
UIMessage SSE	Live tokens / parts for the turn
`GET .../messages`	Persisted UIMessage history on reload
`POST .../messages/truncate`	Drop messages from a checkpoint (edit/regenerate flows)
Webhook `content.updated`	Fired when a successful turn finalizes a content revision
`GET .../content`	Tip revision after finalize

Tools may edit a draft; Save paths persist tip revisions explicitly (or via agent save tools).

Not on `/workflows/.../stream` as UIMessage

Content-edit exposes UIMessage SSE on the conversation messages POST. Do not expect clarifications/programming/audit-style NDJSON chunks on GET /v1/workflows/{id}/stream for chat UX — use the conversation SSE body. See Live updates.