Clarification: Is PDF (file_data) supported in Gemini 3.1 Flash Live Preview via WebSockets? (Error 1008)

Gemini API
1

Description

I am implementing an AI assistant using the Gemini 3.1 Flash Live Preview model over WebSockets in Node.js. I am trying to provide a PDF document as initial context for the session but consistently encounter Error 1008.

Steps to Reproduce

  1. Upload a PDF via Files API (Status is ACTIVE).
  2. Establish a WebSocket connection to v1beta Live API.
  3. Send a setup message with history_config: { "initial_history_in_client_content": true }.
  4. After receiving setupComplete, send a client_content message containing file_data with the PDF’s file_uri.

Observed Behavior

Immediately after the client_content message is sent, the server returns:
Error 1008: Operation not supported

Questions

  1. Does the Multimodal Live API officially support application/pdf via file_data in the real-time stream?
  2. If yes, what is the exact required schema for Gemini 3.1, since putting initial_content in the setup message returns Error 1007 (Unknown name)?
  3. Is this a temporary limitation or is PDF parsing restricted to the Unary/REST API only?

Environment:

  • Model: gemini-3.1-flash-live-preview
  • Environment: Node.js / Next.js
  • SDK: @google/generative-ai