gemini3flash_video_understanding.py

import logging

from vision_agents.core import User, Agent, cli
from vision_agents.core.agents import AgentLauncher
from vision_agents.plugins import getstream, gemini, inworld, smart_turn, deepgram

logger = logging.getLogger(__name__)

async def create_agent(**kwargs) -> Agent:
    agent = Agent(
        edge=getstream.Edge(),  # low latency edge. clients for React, iOS, Android, RN, Flutter etc.
        agent_user=User(name="Video Understanding Agent", id="agent"),
        instructions="Read @gemini_flash_instructions.md",
        llm=gemini.LLM("gemini-3-flash-preview"),
        tts=inworld.TTS(),
        stt=deepgram.STT(),
        turn_detection=smart_turn.TurnDetection(), # vogent turn detection (or omit; some STT providers handle turn-taking)
    )
    return agent


async def join_call(agent: Agent, call_type: str, call_id: str, **kwargs) -> None:
    call = await agent.create_call(call_type, call_id)

    # Have the agent join the call/room
    with await agent.join(call):
        await agent.simple_response("What do you see in my video? Tell me about the objects in the video.")

        # run till the call ends
        await agent.finish()


if __name__ == "__main__":
    cli(AgentLauncher(create_agent=create_agent, join_call=join_call))