LangChain Adapter

# LLM providers (mirrors langchain_openai / langchain_anthropic)
from macaw_adapters.langchain.openai import ChatOpenAI
from macaw_adapters.langchain.anthropic import ChatAnthropic

# Memory classes (mirrors langchain.memory)
from macaw_adapters.langchain.memory import (
    ConversationBufferMemory,
    ConversationBufferWindowMemory,
    ConversationSummaryMemory
)

# Agent functions (mirrors langchain.agents)
from macaw_adapters.langchain.agents import (
    create_react_agent,
    create_openai_functions_agent,
    AgentExecutor
)

# Tool wrappers
from macaw_adapters.langchain.tools import SecureToolWrapper, wrap_tools

# Callback handler for audit logging
from macaw_adapters.langchain.callbacks import MACAWCallbackHandler

from macaw_adapters.langchain import ChatOpenAI

llm = ChatOpenAI(
    model="gpt-4o",
    temperature=0.7,
    max_tokens=1024,
    api_key="sk-...",         # Or use OPENAI_API_KEY env var
    base_url=None,            # Custom API endpoint
    organization=None,        # OpenAI organization ID
    timeout=None,             # Request timeout
    max_retries=2,            # Retry attempts
    macaw_client=None         # Optional pre-configured MACAWClient
)

# All standard LangChain methods work:
response = llm.invoke("Hello!")
async_response = await llm.ainvoke("Hello!")
for chunk in llm.stream("Tell me a story"):
    print(chunk.content, end="")
responses = llm.batch(["Q1", "Q2", "Q3"])

from macaw_adapters.langchain import ChatOpenAI, MACAWCallbackHandler

# Create LLM (has its own MACAWClient)
llm = ChatOpenAI(model="gpt-4o")

# Create callback handler from LLM's MACAW client
handler = MACAWCallbackHandler.from_client(llm._macaw)

# Or create with a new client
handler = MACAWCallbackHandler.create(app_name="my-app")

# Use with any LangChain component
chain = LLMChain(llm=llm, callbacks=[handler])

# Events logged:
# • on_llm_start / on_llm_end / on_llm_error
# • on_tool_start / on_tool_end / on_tool_error
# • on_chain_start / on_chain_end / on_chain_error
# • on_agent_action / on_agent_finish

from macaw_adapters.langchain import (
    ChatOpenAI,
    MACAWCallbackHandler,
    create_react_agent,
    AgentExecutor
)
from langchain.tools import Tool

# Secure LLM
llm = ChatOpenAI(model="gpt-4o")

# Define tools
tools = [
    Tool(name="calculator", func=lambda x: eval(x), description="Math"),
    Tool(name="search", func=search_web, description="Web search"),
]

# Create agent with MAPL security policy
agent = create_react_agent(
    llm, tools, prompt,
    security_policy={
        "resources": ["tool:calculator", "tool:search"],
        "denied_resources": ["tool:admin"],
        "constraints": {
            "denied_parameters": {
                "tool:*": {"input": ["*password*", "*secret*"]}
            }
        }
    }
)

executor = AgentExecutor(
    agent=agent,
    tools=tools,
    security_policy={...}  # Same MAPL format
)

result = executor.invoke({"input": "What's 2+2?"})

from macaw_adapters.langchain import (
    ChatOpenAI,
    ConversationBufferMemory,
    ConversationBufferWindowMemory,
    ConversationSummaryMemory
)

# Each session_id gets isolated memory
alice_memory = ConversationBufferMemory(session_id="alice")
bob_memory = ConversationBufferMemory(session_id="bob")

# Save conversation turn
alice_memory.save_context(
    {"input": "Hello"},
    {"output": "Hi there!"}
)

# Load history
history = alice_memory.load_memory_variables({})

# Window memory (last k exchanges)
window_memory = ConversationBufferWindowMemory(k=5, session_id="user-123")

# Summary memory (uses LLM to summarize)
summary_memory = ConversationSummaryMemory(llm=ChatOpenAI(), session_id="user-123")

from macaw_adapters.langchain import SecureToolWrapper, wrap_tools
from langchain.tools import Tool

# Wrap individual tools
calculator = Tool(name="Calculator", func=calc, description="Math")
secure_calc = SecureToolWrapper(calculator)

# Or wrap all tools at once
tools = [calculator, search_tool, database_tool]
secure_tools = wrap_tools(tools)

# Use wrapped tools in agents
agent = create_react_agent(llm, secure_tools, prompt)

from macaw_adapters.langchain import cleanup

# Clean up all MACAW resources when done
cleanup()