[BUG] Palmyra model access via BedrockModel

[yanomaly]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

0.2.1

Python Version

3.12.5

Operating System

Ubuntu 22.04.5 LTS (Jammy Jellyfish)

Installation Method

pip

Steps to Reproduce

1. Install strands-agents, strands-agents-tools and python-dotenv packages.
2. Set AWS credentials (aws_access_key_id, aws_secret_access_key, aws_session_token)
3. Run provided code sample.
4. You'll receive the following error.

botocore.errorfactory.ValidationException: An error occurred (ValidationException) when calling the ConverseStream operation: This model doesn't support tool use in streaming mode.
└ Bedrock region: us-west-2
└ Model id: us.writer.palmyra-x5-v1:0

5. [Optional] If you set streaming=False on BedrockModel initialization and try to run this code again, you'll receive new error.

strands.types.exceptions.EventLoopException: An error occurred (ValidationException) when calling the Converse operation: This model doesn't support the status field. Remove status and try again.

Code sample:

import os

from boto3 import session
from dotenv import load_dotenv
from strands import Agent, tool
from strands.models import BedrockModel
from strands_tools import http_request

load_dotenv()

WEATHER_SYSTEM_PROMPT = """You are a weather assistant with HTTP capabilities. You can:

1. Make HTTP requests to the National Weather Service API
2. Process and display weather forecast data
3. Provide weather information for locations in the United States

When retrieving weather information:
1. First get the coordinates or grid information using [https://api.weather.gov/points/{latitude},{longitude}](https://api.weather.gov/points/%7Blatitude%7D,%7Blongitude%7D) or [https://api.weather.gov/points/{zipcode}](https://api.weather.gov/points/%7Bzipcode%7D)
2. Then use the returned forecast URL to get the actual forecast

When displaying responses:
- Format weather data in a human-readable way
- Highlight important information like temperature, precipitation, and alerts
- Handle errors appropriately
- Convert technical terms to user-friendly language

Always explain the weather conditions clearly and provide context for the forecast.
"""


@tool
def word_count(text: str) -> int:
    """Count words in text."""
    return len(text.split())


bedrock_model = BedrockModel(
    model_id="us.writer.palmyra-x5-v1:0",
    boto_session=session.Session(
        aws_access_key_id=os.getenv("aws_access_key_id", ""),
        aws_secret_access_key=os.getenv("aws_secret_access_key", ""),
        aws_session_token=os.getenv("aws_session_token", ""),
        region_name="us-west-2",
    ),
    # streaming=False, #uncomment to see new error
)

agent = Agent(
    system_prompt=WEATHER_SYSTEM_PROMPT,
    tools=[word_count, http_request],
    model=bedrock_model,
)
response = agent(
    "What's the weather like in Washington D.C? Also how many words are in the response?"
)

Expected Behavior

Receive an answer, not errors both for streaming/non-streaming modes.

Actual Behavior

Errors occurrence (explained in reproducing steps).

Additional Context

No response

Possible Solution

No response

Related Issues

No response

[nolanlc]

This Python script, ‘weather_word_count.py’ comes from the AWS Workshop, “Building MCP Powered Agents with Claude on Amazon Bedrock”

I ran the same script and it seems the script only works for certain older Claude models. I found that it works for:

model_id="us.anthropic.claude-3-5-haiku-20241022-v1:0"
model_id="anthropic.claude-3-5-sonnet-20241022-v2:0"
model_id="anthropic.claude-3-haiku-20240307-v1:0"
model_id="anthropic.claude-3-5-sonnet-20240620-v1:0"

It does not work for these models:

‘writer.palmyra-x5-v1:0
'amazon.nova-micro-v1:0'
'huggingface-llm-mistral-nemo-base-2407'
‘anthropic.claude-opus-4-20250514-v1:0’
‘anthropic.claude-3-7-sonnet-20250219-v1:0’
‘meta.llama3-3-70b-instruct-v1:0’

[nolanlc]

I have the script running now with this change:

llm = 'arn:aws:bedrock:us-west-2:[AWS ACCOUNT ID]:inference-profile/us.writer.palmyra-x5-v1:0'

bedrock_model = BedrockModel(
model_id=llm,
temperature=0.3,
streaming=False
)

HOWEVER, I do notice when I run the script over and over again with no changes, it will occasionally still give an error. Makes me wonder if it’s a throttle/capacity issue sometimes?

[pgrayy]

I simplified the script to:

import strands
from strands.models import BedrockModel

@strands.tool
def weather() -> str:
    return "sunny"

agent = strands.Agent(
    model=BedrockModel(
        model_id="us.writer.palmyra-x5-v1:0",
        streaming=False,
    ),
    tools=[weather],
)
agent("What is the weather?")

The underlying request sent to Bedrock was:

{
    "modelId": "us.writer.palmyra-x5-v1:0",
    "messages": [
        {
            "content": [{"text": "What is the weather?"}],
            "role": "user"
        },
        {
            "content": [
                {
                    "toolUse": {
                        "toolUseId": "tooluse_k7lCu6YJSD24AK5j5XTRTw",
                        "name": "weather",
                        "input": {}
                    }
                }
            ],
            "role": "assistant"
        },
        {
            "content": [
                {
                    "toolResult": {
                        "content": [
                            {"text": "sunny"}
                        ],
                        "toolUseId": "tooluse_k7lCu6YJSD24AK5j5XTRTw",
                        "status": "success"
                    }
                }
            ],
            "role": "user"
        }
    ],
    "system": [],
    "toolConfig": {
        "tools": [
            {
                "toolSpec": {
                    "name": "weather",
                    "description": "weather",
                    "inputSchema": {
                        "json": {
                            "properties": {},
                            "type": "object",
                            "required": []
                        }
                    }
                }
            }
        ],
        "toolChoice": {
            "auto": {}
        }
    },
    "inferenceConfig": {}
}

This is a valid request format based on the Bedrock docs. With that said, have you raised this issue to Bedrock?

(Side note, I would recommend upgrading to to Strands 1.x).

[pgrayy]

Dug a little deeper into the issue. Bedrock does note that status is only supported for claude models (docs). With that said, we actually have someone working on the issue right now in #686.