Anthropic Integration¶

Introduction¶

AdalFlow provides seamless integration with Anthropic’s Claude models through the AnthropicAPIClient. This integration supports all Claude models including Claude 3.5 Sonnet, Claude 3.5 Haiku, Claude 3 Opus, and the latest Claude Sonnet 4 with advanced reasoning capabilities.

Key Features¶

Model Agnostic: Easy switching between different Claude models via configuration
Reasoning Support: Built-in support for Claude’s reasoning capabilities with interleaved thinking
Tool Usage: Support Anthropic’s native tool via the tools parameter in model_kwargs but more powerfully, you can use the ReActAgent to use tools even easier.

Basic Usage¶

Simple Text Generation¶

import adalflow as adal

# Basic Claude 3.5 Sonnet usage
anthropic_llm = adal.Generator(
    model_client=adal.AnthropicAPIClient(),
    model_kwargs={"model": "claude-3-5-sonnet-20241022"},
)

prompt_kwargs = {"input_str": "What is the meaning of life in one sentence?"}
response = anthropic_llm(prompt_kwargs)
print(f"Anthropic: {response}")

Here is the output:

Anthropic: GeneratorOutput(id=None, data='The meaning of life is to find your own purpose while making a positive impact on others and growing through experiences, relationships, and personal discovery.', thinking=None, tool_use=None, error=None, usage=CompletionUsage(completion_tokens=31, prompt_tokens=70, total_tokens=101), raw_response='The meaning of life is to find your own purpose while making a positive impact on others and growing through experiences, relationships, and personal discovery.', api_response=Message(id='msg_01LxWtz8bjf4a9Q1GoXSMFDw', content=[TextBlock(citations=None, text='The meaning of life is to find your own purpose while making a positive impact on others and growing through experiences, relationships, and personal discovery.', type='text')], model='claude-3-5-sonnet-20241022', role='assistant', stop_reason='end_turn', stop_sequence=None, type='message', usage=Usage(cache_creation_input_tokens=0, cache_read_input_tokens=0, input_tokens=70, output_tokens=31, server_tool_use=None, service_tier='standard')), metadata=None)

The text output is in the data field, and the api_response field is the raw response from the Anthropic API. The raw_response field is the text content of the api message.

Advanced Reasoning¶

Claude 4+ models support advanced reasoning with interleaved thinking:

# Enable reasoning with Claude Sonnet 4
anthropic_llm = adal.Generator(
    model_client=adal.AnthropicAPIClient(),
    model_kwargs={
        "model": "claude-sonnet-4-20250514",
        "thinking": {"type": "enabled", "budget_tokens": 10000},
        "extra_headers": {"anthropic-beta": "interleaved-thinking-2025-05-14"},
    },
)

prompt_kwargs = {"input_str": "What is the meaning of life in one sentence?"}
response = anthropic_llm(prompt_kwargs)
print(f"Anthropic: {response}")

Here is the output:

Anthropic: GeneratorOutput(id=None, data='The meaning of life is to find purpose through love, growth, and connection—creating significance in our brief existence by caring for others and contributing something meaningful to the world.', thinking='This is a classic philosophical question that has been pondered by humans for millennia. There are many different perspectives on this - religious, philosophical, existential, scientific, etc. Since the user is asking for just one sentence, I should try to give a thoughtful but concise answer that captures something meaningful without being overly specific to any one worldview.\n\nSome possible approaches:\n- Focus on creating meaning through relationships, love, and connection\n- Emphasize personal growth and self-actualization  \n- Highlight contributing to something larger than oneself\n- Suggest that meaning is something we create rather than discover\n- Acknowledge the subjectivity of meaning\n\nI think a good answer would acknowledge that meaning is personal while suggesting some universal human elements that tend to contribute to a sense of meaning.', tool_use=None, error=None, usage=CompletionUsage(completion_tokens=212, prompt_tokens=98, total_tokens=310), raw_response='The meaning of life is to find purpose through love, growth, and connection—creating significance in our brief existence by caring for others and contributing something meaningful to the world.', api_response=Message(id='msg_01AR4WhBmWkrqYuSgy1ynR1k', content=[ThinkingBlock(signature='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', thinking='This is a classic philosophical question that has been pondered by humans for millennia. There are many different perspectives on this - religious, philosophical, existential, scientific, etc. Since the user is asking for just one sentence, I should try to give a thoughtful but concise answer that captures something meaningful without being overly specific to any one worldview.\n\nSome possible approaches:\n- Focus on creating meaning through relationships, love, and connection\n- Emphasize personal growth and self-actualization  \n- Highlight contributing to something larger than oneself\n- Suggest that meaning is something we create rather than discover\n- Acknowledge the subjectivity of meaning\n\nI think a good answer would acknowledge that meaning is personal while suggesting some universal human elements that tend to contribute to a sense of meaning.', type='thinking'), TextBlock(citations=None, text='The meaning of life is to find purpose through love, growth, and connection—creating significance in our brief existence by caring for others and contributing something meaningful to the world.', type='text')], model='claude-sonnet-4-20250514', role='assistant', stop_reason='end_turn', stop_sequence=None, type='message', usage=Usage(cache_creation_input_tokens=0, cache_read_input_tokens=0, input_tokens=98, output_tokens=212, server_tool_use=None, service_tier='standard')), metadata=None)

In this case, you will see the thinking field is not None, and the data field is the final answer.

Tool Usage with Reasoning¶

Claude models can use tools while leveraging reasoning capabilities. In this case, we showcase the native tool usage via the tools parameter in model_kwargs. And the tool_use field of the GeneratorOutput will show the final tool call.

# Define tools for function calling
calculator_tool = {
    "name": "calculator",
    "description": "Perform mathematical calculations",
    "input_schema": {
        "type": "object",
        "properties": {
            "expression": {
                "type": "string",
                "description": "Mathematical expression to evaluate",
            }
        },
        "required": ["expression"],
    },
}

database_tool = {
    "name": "database_query",
    "description": "Query product database",
    "input_schema": {
        "type": "object",
        "properties": {
            "query": {"type": "string", "description": "SQL query to execute"}
        },
        "required": ["query"],
    },
}

# Configure Claude with tools and reasoning
anthropic_llm = adal.Generator(
    model_client=adal.AnthropicAPIClient(),
    model_kwargs={
        "model": "claude-sonnet-4-20250514",
        "thinking": {"type": "enabled", "budget_tokens": 10000},
        "extra_headers": {"anthropic-beta": "interleaved-thinking-2025-05-14"},
        "tools": [calculator_tool, database_tool],
    },
)

prompt_kwargs = {
    "input_str": "What's the total revenue if we sold 150 units of product A at $50 each, and how does this compare to our average monthly revenue from the database?"
}

response = anthropic_llm(prompt_kwargs)
print(f"Anthropic: {response}")

Here is the output:

GeneratorOutput(id=None, data="I'll help you calculate the total revenue and compare it to your average monthly revenue. Let me start by calculating the revenue from the product A sales, then query the database for comparison.", thinking="The user is asking for two things:\n1. Calculate the total revenue from selling 150 units of product A at $50 each\n2. Compare this to the average monthly revenue from the database\n\nFor the first part, I can use the calculator tool to compute 150 × $50.\n\nFor the second part, I need to query the database to get information about monthly revenue. I'll need to construct a SQL query to get the average monthly revenue.\n\nLet me start with the calculation first, then query the database.", tool_use=Function(thought=None, name='calculator', args=[], kwargs={'expression': '150 * 50'}), error=None, usage=CompletionUsage(completion_tokens=213, prompt_tokens=704, total_tokens=917), raw_response="I'll help you calculate the total revenue and compare it to your average monthly revenue. Let me start by calculating the revenue from the product A sales, then query the database for comparison.", api_response=Message(id='msg_01QBwCzqGPLAwaxTDGChZsRN', content=[ThinkingBlock(signature='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', thinking="The user is asking for two things:\n1. Calculate the total revenue from selling 150 units of product A at $50 each\n2. Compare this to the average monthly revenue from the database\n\nFor the first part, I can use the calculator tool to compute 150 × $50.\n\nFor the second part, I need to query the database to get information about monthly revenue. I'll need to construct a SQL query to get the average monthly revenue.\n\nLet me start with the calculation first, then query the database.", type='thinking'), TextBlock(citations=None, text="I'll help you calculate the total revenue and compare it to your average monthly revenue. Let me start by calculating the revenue from the product A sales, then query the database for comparison.", type='text'), ToolUseBlock(id='toolu_01SwoVz9x4M24gceoybvUrqG', input={'expression': '150 * 50'}, name='calculator', type='tool_use')], model='claude-sonnet-4-20250514', role='assistant', stop_reason='tool_use', stop_sequence=None, type='message', usage=Usage(cache_creation_input_tokens=0, cache_read_input_tokens=0, input_tokens=704, output_tokens=213, server_tool_use=None, service_tier='standard')), metadata=None)

In particular, here is the tool_use field:

tool_use=Function(thought=None, name='calculator', args=[], kwargs={'expression': '150 * 50'})

As you can see, we have parsed the tool use response into a Function object.

ReAct Agent with Thinking Models¶

Instead of using the native tools, you can use the ReActAgent to use tools even easier. The output is directly parsed into a ReActAgentOutput object where you can find both the answer and the steps of the agent.

In this case, we are using a thinking model, so inside of ReActAgent, the manual CoT is disabled on the reasoning.

from adalflow.components.agent.react import ReActAgent

# Define thinking model configuration
thinking_model_kwargs = {
    "model": "claude-sonnet-4-20250514",
    "thinking": {"type": "enabled", "budget_tokens": 10000},
    "extra_headers": {"anthropic-beta": "interleaved-thinking-2025-05-14"},
}

# Define tools as Python functions
def calculator(expression: str, **kwargs) -> str:
    """Perform mathematical calculations"""
    return "7500"

def database_query(query: str, **kwargs) -> str:
    """Query product database"""
    return "5200"

# Create ReAct Agent with Anthropic thinking model
agent = ReActAgent(
    model_client=adal.AnthropicAPIClient(),
    model_kwargs=thinking_model_kwargs,
    tools=[calculator, database_query],
    use_cache=False,
    is_thinking_model=True,  # Enable thinking model support
)

prompt_kwargs = {
    "input_str": "What's the total revenue if we sold 150 units of product A at $50 each, and how does this compare to our average monthly revenue from the database?"
}

agent_response = agent(prompt_kwargs)
print(f"Agent: {agent_response}")

The ReAct Agent with thinking models provides several advantages:

Enhanced Reasoning: The agent can use Claude’s interleaved thinking to break down complex problems
Tool Integration: Seamless integration with Python functions as tools
Multi-step Planning: The agent can plan and execute multi-step reasoning processes
Automatic Tool Calling: Tools are automatically called based on the agent’s reasoning

Available Models¶

AdalFlow supports all current Anthropic Claude models. You can find the list of available models in the Anthropic API documentation.

Configuration Options¶

The AnthropicAPIClient supports various configuration options:

anthropic_client = adal.AnthropicAPIClient(
    api_key="your-api-key",  # Optional: uses ANTHROPIC_API_KEY env var
    base_url="https://api.anthropic.com",  # Optional: custom base URL
    timeout=60,  # Optional: request timeout in seconds
    max_retries=3,  # Optional: maximum retry attempts
)

Model-specific parameters can be passed via model_kwargs:

model_kwargs = {
    "model": "claude-sonnet-4-20250514",
    "max_tokens": 4096,
    "temperature": 0.7,
    "top_p": 0.9,
    "thinking": {"type": "enabled", "budget_tokens": 10000},
    "tools": [...],  # Function definitions
    "extra_headers": {"anthropic-beta": "interleaved-thinking-2025-05-14"},
}

API Reference