nova_tools

IAM Permissions Required:

System tools require the bedrock:InvokeTool IAM permission in addition to bedrock:InvokeModel.

Web Grounding (nova_grounding): Enables the model to search the web for current information.

.. code-block:: python

    from langchain_aws import ChatBedrockConverse
    from langchain_aws.tools import NovaGroundingTool

    model = ChatBedrockConverse(model="amazon.nova-2-lite-v1:0")
    model_with_search = model.bind_tools([NovaGroundingTool()])

    response = model_with_search.invoke(
        "What are the latest developments in quantum computing?"
    )
    print(response.content)

You can also use the string name directly:

.. code-block:: python

    model_with_search = model.bind_tools(["nova_grounding"])

Code Interpreter (nova_code_interpreter): Enables the model to execute Python code in a sandboxed environment.

.. code-block:: python

    from langchain_aws import ChatBedrockConverse
    from langchain_aws.tools import NovaCodeInterpreterTool

    model = ChatBedrockConverse(model="amazon.nova-2-lite-v1:0")
    model_with_code = model.bind_tools([NovaCodeInterpreterTool()])

    response = model_with_code.invoke(
        "Calculate the square root of 475878756857"
    )
    print(response.content)

Reasoning with Nova:

Nova models support reasoning configuration similar to Claude's thinking feature, but use reasoningConfig instead of thinking.

.. code-block:: python

from langchain_aws import ChatBedrockConverse

reasoning_config = {
    "reasoningConfig": {
        "type": "enabled",
        "maxReasoningEffort": "low"  # or "medium", "high"
    }
}

model = ChatBedrockConverse(
    model="amazon.nova-2-lite-v1:0",
    max_tokens=10000,
    additional_model_request_fields=reasoning_config
)

response = model.invoke("Solve this logic puzzle: ...")
# Access reasoning content via response.content_blocks

Combining System Tools and Reasoning:

.. code-block:: python

from botocore.config import Config
from langchain_aws import ChatBedrockConverse
from langchain_aws.tools import NovaGroundingTool

# Increase timeout for reasoning and tool execution
config = Config(
    connect_timeout=3600,  # 60 minutes
    read_timeout=3600,     # 60 minutes
    retries={'max_attempts': 1}
)

model = ChatBedrockConverse(
    model="amazon.nova-2-lite-v1:0",
    max_tokens=10000,
    config=config,
    additional_model_request_fields={
        "reasoningConfig": {
            "type": "enabled",
            "maxReasoningEffort": "medium"  # or "low", "high"
        }
    }
)

model_with_tools = model.bind_tools([NovaGroundingTool()])
response = model_with_tools.invoke(
    "Research and explain the latest AI breakthroughs"
)

Timeout Recommendations:

• Low reasoning effort: Default timeout (30 seconds) is usually sufficient
• Medium reasoning effort: Consider 300-600 seconds (5-10 minutes)
• High reasoning effort: Consider 1800-3600 seconds (30-60 minutes)
• Code interpreter: Consider 300-600 seconds for complex computations
• Web grounding: Default timeout is usually sufficient