Update system instructions during a session

[marsidev]

Can we have support for "Update system instructions during a session" as mentioned here? Seems it's only supported by the Python SDK so far.

I already tried what the example shows (sendClientContent with "system" role and turnComplete set to false) and I got a close event with these details:

{"code":1007,"reason":"Request contains an invalid argument.","type":"close","wasClean":true}

Note: I'm using the version 1.10.0.

[shmishra99]

Hi @marsidev,

We appreciate you taking the time to share your thoughts with us, and we'll be filing a feature request. We’ll keep you updated once we have more information.

Thank you!

[thomas-s97]

@shmishra99 I'm having the same problem as @marsidev. I'm also getting the 1007 error when changing the system prompt mid-session using Typescript.

According to this documentation it is possible to change the system prompt in the live api during a session. But for some reason, there is only one example using Python:
https://cloud.google.com/vertex-ai/generative-ai/docs/live-api/streamed-conversations#update_system_instructions_during_a_session

I was trying to change the system prompt during a session in Python and I'm getting the exact same error:

❌ Error receiving response after system update: received 1007 (invalid frame payload data) Request contains an invalid argument.; then sent 1007 (invalid frame payload data) Request contains an invalid argument.

I was performing a test using the models gemini-live-2.5-flash-preview and gemini-2.5-flash-preview-native-audio-dialog. Both of them should be supported according to the documentation, but we get the same error in both models.

Here is the full code (in Python)

#!/usr/bin/env python3
"""
Test response flow after system instruction updates
"""

import asyncio
import os
from google import genai
from google.genai import types

API_KEY = os.getenv("GEMINI_API_KEY")
if not API_KEY:
    print("Please set GEMINI_API_KEY environment variable")
    exit(1)

async def test_response_flow_after_system_update():
    """Test receiving responses after system instruction update"""
    
    client = genai.Client(api_key=API_KEY)
    model = "gemini-live-2.5-flash-preview"  # Using stable version instead of preview
    
    print("🧪 Testing response flow after system instruction update...")
    
    config = types.LiveConnectConfig(
        response_modalities=["TEXT"],
        system_instruction=types.Content(
            parts=[types.Part(text="You are a formal assistant. Be very professional.")],
            role="user"
        ),
        temperature=0.7,
    )
    
    try:
        async with client.aio.live.connect(model=model, config=config) as session:
            print("✅ Connected")
            
            # Step 1: Send initial message and get response
            print("\n📝 Step 1: Initial message")
            await session.send_client_content(
                turns=types.Content(
                    role="user", 
                    parts=[types.Part(text="Hello, how are you?")]
                ),
                turn_complete=True
            )
            
            # Get initial response
            print("🔄 Getting initial response...")
            response_received = False
            async for message in session.receive():
                if message.server_content and message.server_content.model_turn:
                    for part in message.server_content.model_turn.parts:
                        if part.text:
                            print(f"🤖 Initial response: {part.text}")
                            response_received = True
                            break
                    if response_received:
                        break
                # Add timeout to avoid hanging
                if response_received:
                    break
            
            if not response_received:
                print("⚠️ No initial response received")
            
            # Step 2: Update system instruction
            print("\n🔄 Step 2: Updating system instruction...")
            await session.send_client_content(
                turns=types.Content(
                    role="system", 
                    parts=[types.Part(text="You are now a very casual assistant. Use emojis and slang.")]
                ),
                turn_complete=False
            )
            print("✅ System instruction update sent")
            
            # Step 3: Send another message and try to get response
            print("\n📝 Step 3: Message after system update")
            await session.send_client_content(
                turns=types.Content(
                    role="user", 
                    parts=[types.Part(text="Hello again, how are you?")]
                ),
                turn_complete=True
            )
            
            # Try to get response after system update
            print("🔄 Getting response after system update...")
            response_received = False
            try:
                async for message in session.receive():
                    print(f"📨 Received message type: {type(message)}")
                    if hasattr(message, 'server_content') and message.server_content:
                        print(f"📨 Server content: {message.server_content}")
                        if message.server_content.model_turn:
                            for part in message.server_content.model_turn.parts:
                                if part.text:
                                    print(f"🤖 Response after update: {part.text}")
                                    response_received = True
                                    break
                            if response_received:
                                break
                    # Add timeout mechanism
                    if response_received:
                        break
                        
            except Exception as e:
                print(f"❌ Error receiving response after system update: {e}")
                return False
            
            if not response_received:
                print("⚠️ No response received after system update")
            
            print("✅ Response flow test completed successfully")
            return True
            
    except Exception as e:
        print(f"❌ Response flow test failed: {e}")
        return False

async def main():
    """Run response flow test"""
    print("🔍 Testing Response Flow After System Instruction Updates")
    print("=" * 60)
    
    success = await test_response_flow_after_system_update()
    
    if success:
        print("\n🎉 Response flow works! The issue might be elsewhere.")
    else:
        print("\n❌ Response flow has issues - this is likely the root cause.")

if __name__ == "__main__":
    asyncio.run(main()) 

Did anyone manage to change the system prompt mid-session in Typescript (or also in Python)?