Input Content

Provide user input to BidiAgent as text, images, or streaming audio. Text and image blocks form complete user messages, while audio deltas add samples to an ongoing input stream.

Send input

Use agent.send() to provide input once the agent is running. A string is enough to send text:

import asyncio

from strands.bidi.agent import BidiAgent
from strands.bidi.types import BidiResponseStopEvent


async def main() -> None:
    async with BidiAgent() as agent:
        await agent.send("Tell me about Lisbon.")

        async for event in agent.receive():
            if isinstance(event, BidiResponseStopEvent):
                break


asyncio.run(main())

send() returns after sending the input. Read the model’s response separately through agent.receive(), as the example does.

BidiAgentInput defines the forms you can send:

Content	Accepted forms
Text	A string, `TextBlock`, or `{"text": ...}`.
Image	`ImageBlock` or `{"image": ...}`.
Streaming audio	`AudioDelta` or `{"audio_delta": ...}`.

A nonempty list of strings, text/image blocks, or their dictionary forms groups content into one user message, preserving block order.

For continuous input, pass input streams to agent.run(). Each read returns one of these input forms.

Streaming audio

Send live audio as individual AudioDelta inputs, outside text and image lists. Each delta carries a chunk of audio samples without explicitly ending the user’s turn; the model provider determines when the turn is complete.

The built-in models expect raw PCM bytes matching their configured input sample rate and channel count. This example forwards chunks from an asynchronous audio source to a running agent:

Typed content
Dictionaries

from collections.abc import AsyncIterable

from strands.bidi.agent import BidiAgent
from strands.bidi.types import AudioDelta


async def send_audio(agent: BidiAgent, chunks: AsyncIterable[bytes]) -> None:
    async for chunk in chunks:
        await agent.send(AudioDelta(format="pcm", source={"bytes": chunk}))

from collections.abc import AsyncIterable

from strands.bidi.agent import BidiAgent


async def send_audio(agent: BidiAgent, chunks: AsyncIterable[bytes]) -> None:
    async for chunk in chunks:
        await agent.send({
            "audio_delta": {"format": "pcm", "source": {"bytes": chunk}},
        })

For microphone input, AudioIO creates these deltas using the model’s audio configuration.

Model messages

Your application sends user input through agent.send(), which prepares it for BidiModel.send(). Text and images become a BidiMessage containing an ordered list of typed blocks; audio deltas pass through individually.

The agent also uses BidiMessage to return tool results to the model, packaging each tool’s output as a ToolResultBlock. It handles this delivery automatically, so tool results aren’t part of BidiAgentInput. You can track tool calls and results through tool execution events.

See Messages for how the agent records user input and tool results alongside content assembled from stream events.