Quickstart¶

import asyncio

from ravi.agents import ReActAgent, Runtime
from ravi.agents.context import ContextConfig, InMemoryHistoryProvider
from ravi.integrations.llm.openai import OpenAIClient
from ravi.kernel import Message, ChatPayload, ChatMessage, Role, TextBlock
from ravi.kernel.core.identity import AgentId


async def main() -> None:
    # 1. Initialize the LLM client and the ReActAgent
    client = OpenAIClient(api_key="sk-...", model="gpt-4o")
    agent = ReActAgent(
        "DemoBot",
        model=client,
        context=ContextConfig(InMemoryHistoryProvider()),
    )

    # 2. Run the agent via the Runtime
    async with Runtime() as rt:
        await rt.register(agent)

        # 3. Submit a message and await completion
        msg = Message(
            target=agent.id,
            sender=AgentId(type="proxy", key="user"),
            payload=ChatPayload(
                message=ChatMessage(
                    role=Role.USER,
                    content=[TextBlock(text="Why do agents need runtimes?")]
                )
            ),
        )
        run_id = await rt.submit(agent.id, msg)

        async for entry in rt.event_log.tail(run_id):
            if entry.kind in ("run.completed", "run.failed", "run.cancelled"):
                break

        # 4. Extract and print the final assistant turn response
        history = await agent.history.get_messages(agent.id, session_id=run_id)
        for m in reversed(history):
            if m.role == Role.ASSISTANT:
                print("Response:", " ".join(b.text for b in m.content if isinstance(b, TextBlock)))
                break


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