BidiAgent

The BidiAgent is a specialized agent designed for real-time bidirectional streaming conversations. Unlike the standard Agent that follows a request-response pattern, BidiAgent maintains persistent connections that enable continuous audio and text streaming, real-time barge-ins, and concurrent tool execution.

flowchart TB
    subgraph User
        A[Microphone] --> B[Audio Input]
        C[Text Input] --> D[Input Content]
        B --> D
    end

    subgraph BidiAgent
        D --> E[Agent Loop]
        E --> F[Model Connection]
        F --> G[Tool Execution]
        G --> F
        F --> H[Output Events]
    end

    subgraph Output
        H --> I[Audio Output]
        H --> J[Text Output]
        I --> K[Speakers]
        J --> L[Console/UI]
    end

Agent vs BidiAgent

While both Agent and BidiAgent share the same core purpose of enabling AI-powered interactions, they differ significantly in their architecture and use cases.

Standard Agent (Request-Response)

The standard Agent follows a traditional request-response pattern:

from strands import Agent
from strands.vended_tools import notebook

agent = Agent(tools=[notebook])

# Single request-response cycle
result = agent('Create a notebook named "ideas" and add three project ideas.')
print(result.message)

Characteristics:

• Synchronous interaction: One request, one response
• Discrete cycles: Each invocation is independent
• Message-based: Operates on complete messages
• Tool execution: Sequential, blocking the response

BidiAgent (Bidirectional Streaming)

BidiAgent maintains a persistent, bidirectional connection:

import asyncio
from strands.bidi.agent import BidiAgent
from strands.bidi.io import AudioIO
from strands.bidi.models import BedrockNovaSonicModel

model = BedrockNovaSonicModel(model_id="amazon.nova-2-sonic-v1:0")
agent = BidiAgent(model=model, tools=[notebook])
audio_io = AudioIO()

async def main():
    # Persistent connection with continuous streaming
    await agent.run(
        inputs=[audio_io.input()],
        outputs=[audio_io.output()]
    )

asyncio.run(main())

Characteristics:

• Asynchronous streaming: Continuous input/output
• Persistent connection: Single connection for multiple turns
• Event-based: Operates on streaming events
• Tool execution: Concurrent, non-blocking

When to Use Each

Use Agent when:

• Building chatbots or CLI applications
• Processing discrete requests
• Implementing API endpoints
• Working with text-only interactions
• Simplicity is preferred

Use BidiAgent when:

• Building voice assistants
• Requiring real-time audio streaming
• Needing natural conversation barge-ins
• Implementing live transcription
• Building interactive, multi-modal applications

The Bidirectional Agent Loop

The bidirectional agent loop is fundamentally different from the standard agent loop. Instead of processing discrete messages, it continuously streams events in both directions while managing connection state and concurrent operations.

Architecture Overview

flowchart TB
    A[Agent Start] --> B[Model Connection]
    B --> C[Agent Loop]
    C --> D[Model Task]
    C --> E[Event Queue]
    D --> E
    E --> F[receive]
    D --> G[Tool Detection]
    G --> H[Tool Tasks]
    H --> E
    F --> I[User Code]
    I --> J[send]
    J --> K[Model]
    K --> D

Event Flow

Startup Sequence

Agent Initialization

agent = BidiAgent(model=model, tools=[notebook])

Creates tool registry, initializes agent state, and sets up hook registry.

Connection Start

await agent.start()

Calls model.start(system_prompt, tools, messages), establishes WebSocket/SDK connection, sends conversation history if provided, spawns background task for model communication, and enables sending capability.

Event Processing

async for event in agent.receive():
    # Process events

Dequeues events from internal queue, yields to user code, and continues until stopped.

Tool Execution

Tools execute concurrently without blocking the conversation. When the model requests a group of tool calls:

1. One assistant message containing the tool uses and one user message containing their dispatch acknowledgements are appended together to conversation history.
2. The tool executor streams events for each call as it runs.
3. Once all calls finish, the tool uses are repeated in one assistant message, followed by one user message containing their results in request order. The results are sent together to the model.

Each grouped tool-use message stays adjacent to its matching result message, even when conversation continues or tools finish out of order. Both exchanges retain the original tool-use IDs, and the repeated tool uses do not execute again. Dispatch acknowledgements stay in history and are not sent to the model.

Tool-result messages use metadata["custom"]["bidi"]["kind"] to distinguish tool_dispatch acknowledgements from tool_result messages.

To let a tool end the conversation, call tool_context.agent.cancel(). Cancellation takes effect only after the tool group completes. Requests from other contexts remain pending until then.

See Graceful shutdown for a custom tool example.

Connection Lifecycle

Normal Operation

User → send() → Model → receive() → Model Task → Event Queue → receive() → User
                  ↓
              Tool Use
                  ↓
            Tool Task → Event Queue → receive() → User
                  ↓
            Tool Result → Model

Configuration

Configure a BidiAgent with a model, tools, a system prompt, and optional conversation history.

Basic Configuration

from strands.bidi.agent import BidiAgent
from strands.bidi.models import BedrockNovaSonicModel

model = BedrockNovaSonicModel(model_id="amazon.nova-2-sonic-v1:0")

agent = BidiAgent(
    model=model,
    tools=[notebook, weather],
    system_prompt="You are a helpful voice assistant.",
    messages=[],  # Optional conversation history
    agent_id="voice_assistant_1",
    name="Voice Assistant",
    description="A voice-enabled AI assistant"
)

Model Configuration

Each model provider has specific configuration options:

from strands.bidi.models import BedrockNovaSonicModel

model = BedrockNovaSonicModel(
    model_id="amazon.nova-2-sonic-v1:0",
    region="us-east-1",
    voice="matthew",
    audio={
        "input": {"sample_rate": 16000},
        "output": {"sample_rate": 16000},
    },
)

See Model Providers for provider-specific options.

BidiAgent supports many of the same constructs as Agent:

• Tools: Function calling works identically
• Hooks: Lifecycle event handling with bidirectional-specific events
• Tool Executors: Concurrent and custom execution patterns

Lifecycle Management

The BidiAgent lifecycle governs how connections open, run, and close. Managing it correctly is what keeps resources cleaned up and errors handled.

Lifecycle States

stateDiagram-v2
    [*] --> Created: BidiAgent
    Created --> Started: start
    Started --> Running: run or receive
    Running --> Running: send and receive events
    Running --> Stopped: stop
    Stopped --> [*]

    Running --> Restarting: Restart timer or timeout
    Restarting --> Running: Restarted

State Transitions

1. Creation

agent = BidiAgent(model=model, tools=[notebook])
# Tool registry initialized, agent state created, hooks registered
# NOT connected to model yet

2. Starting

await agent.start(invocation_state={...})
# Model connection established, conversation history sent
# Background tasks spawned, ready to send/receive

3. Running

# Option A: Using run()
await agent.run(inputs=[...], outputs=[...])

# Option B: Manual send/receive
await agent.send("Hello")
async for event in agent.receive():
    # Process events - events streaming, tools executing, messages accumulating
    pass

4. Stopping

await agent.stop()
# Background tasks cancelled, model connection closed, resources cleaned up

Lifecycle Patterns

Using run()

agent = BidiAgent(model=model)
audio_io = AudioIO()

await agent.run(
    inputs=[audio_io.input()],
    outputs=[audio_io.output()]
)

Simplest for I/O-based applications - handles start/stop automatically.

Context Manager

agent = BidiAgent(model=model)

async with agent:
    await agent.send("Hello")
    async for event in agent.receive():
        if isinstance(event, BidiResponseStopEvent):
            break

Automatic start() and stop() with exception-safe cleanup. To pass invocation_state, call start() manually before entering the context.

Manual Lifecycle

agent = BidiAgent(model=model)

try:
    await agent.start()
    await agent.send("Hello")

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

Explicit control with custom error handling and flexible timing.

Sending Multiple Content Blocks

Send a list to group text and images into one user message, preserving block order. With a running OpenAI or Gemini agent:

from strands.bidi.agent import BidiAgent
from strands.types.media import ImageBlock

async def describe_image(agent: BidiAgent, image_bytes: bytes) -> None:
    await agent.send([
        ImageBlock(format="jpeg", source={"bytes": image_bytes}),
        "What is in this image?",
    ])

Lists accept strings, TextBlock, ImageBlock, and their dictionary forms. Send audio deltas individually; tool results are managed by the agent. Nova Sonic supports text only and joins blocks with newlines.

Connection Restart

Every provider caps how long a single connection stays open. BidiAgent restarts the connection on two paths, both of which preserve the conversation:

• Proactive (scheduled): a timer fires ahead of the provider’s limit, and the agent restarts the connection before it drops. This is the normal path.
• Reactive (timeout): if the connection times out first, the agent restarts the connection after the provider reports the timeout.

Both paths emit a BidiConnectionRestartEvent. Read event.reason ("scheduled" or "timeout") to tell them apart, and event.turn_interrupted to detect when a restart cut an in-progress turn:

async for event in agent.receive():
    if isinstance(event, BidiConnectionRestartEvent):
        print(f"Restarting (reason={event.reason})")
        if event.turn_interrupted:
            # The provider replays history as context, but this turn was not answered.
            # Re-prompt or notify the user.
            pass
        # Conversation history is preserved; keep processing events normally.

Ahead of a proactive restart, the agent also emits a BidiConnectionWarningEvent carrying time_left_s, the approximate seconds until the swap. It is informational, useful for surfacing a “restarting shortly” hint in a UI. The full event catalog is on the Events page.

The restart sequence: restart timer fires (or a timeout is reported) → BidiConnectionRestartEvent emitted → sending blocked → hooks invoked → model restarted with history → new receiver task spawned → sending unblocked → conversation continues.

Tuning Restart Timing

Each provider declares its restart timing as a ConnectionConfig. Override it, or opt out of automatic restart with the model’s connection argument:

from strands.bidi.models import BedrockNovaSonicModel

model = BedrockNovaSonicModel(
    model_id="amazon.nova-2-sonic-v1:0",
    connection={
        "restart_after_s": 360,  # Restart this many seconds after a connection opens
        "auto_reconnect": True,  # Set False to disable automatic restart
    },
)

restart_after_s should sit at least ~10 seconds below the provider’s own connection limit, because the restart may wait briefly for the current turn to finish before swapping. Setting auto_reconnect=False turns off both the proactive timer and reactive restart, so the connection closes at the provider’s limit.

Error Handling

Handling Connection Errors

try:
    await agent.start()
    async for event in agent.receive():
        # Handle connection restart events
        if isinstance(event, BidiConnectionRestartEvent):
            print("Connection restarting, please wait...")
            continue  # Connection restarts automatically

        # Process other events
        pass
except Exception as e:
    print(f"Unexpected error: {e}")
finally:
    await agent.stop()

Note: Restarts are handled automatically, whether scheduled by the restart timer or triggered by a timeout. The agent emits BidiConnectionRestartEvent when restarting the connection.

Graceful Shutdown

import signal

agent = BidiAgent(model=model)
audio_io = AudioIO()

async def main():
    # Setup signal handler
    loop = asyncio.get_event_loop()

    def signal_handler():
        print("\nShutting down gracefully...")
        loop.create_task(agent.stop())

    loop.add_signal_handler(signal.SIGINT, signal_handler)
    loop.add_signal_handler(signal.SIGTERM, signal_handler)

    try:
        await agent.run(
            inputs=[audio_io.input()],
            outputs=[audio_io.output()]
        )
    except asyncio.CancelledError:
        print("Agent stopped")

asyncio.run(main())

Resource Cleanup

The agent automatically cleans up background tasks, model connections, I/O streams, and event queues, then invokes cleanup hooks.

Best Practices

1. Always Use try/finally: Ensure stop() is called even on errors
2. Prefer Context Managers: Use async with for automatic cleanup
3. Handle Restarts Gracefully: Don’t treat BidiConnectionRestartEvent as an error
4. Monitor Lifecycle Hooks: Use hooks to track state transitions
5. Test Shutdown: Verify cleanup works under various conditions
6. Avoid Calling stop() During receive(): Only call stop() after exiting the receive loop

Next Steps

• Events - Complete guide to bidirectional streaming events
• I/O Streams - Building custom input and output streams
• Model Providers - Provider-specific configuration
• Quickstart - Getting started guide
• Python API Reference - Complete API documentation

Tags

bidi-streaming12