Add OpenAI Agents SDK and LangChain partner examples

CLAUDE.md

@@ -81,4 +81,5 @@ Both APIs are fully supported with sync/async variants and streaming support.
 
 - **Core**: `pydantic>=2`, `httpx>=0.27`, `twilio>=9.8.3`
 - **Server** (optional): `fastapi`, `uvicorn`, `python-multipart` — install with `pip install tac[server]`
-- **Dev**: `pytest`, `ruff`, `mypy`, `openai`, `openai-agents`
+- **Dev**: `pytest`, `ruff`, `mypy`
+- **Examples**: `openai`, `openai-agents`, `langchain-core`, `langchain-openai`, `boto3`, `strands-agents`, `python-dotenv`

README.md

@@ -169,7 +169,7 @@ For detailed architecture and advanced usage, see [CLAUDE.md](https://github.com
 
 **Examples & Guides:**
 - **[Getting Started Guide](https://github.com/twilio/twilio-agent-connect-python/tree/main/getting_started/)** - Setup wizard, examples, and comprehensive documentation
-- **[Partner SDK Examples](https://github.com/twilio/twilio-agent-connect-python/tree/main/getting_started/examples/partners/)** - Integration examples for OpenAI, AWS Bedrock Agent, AWS Bedrock AgentCore, and AWS Strands
+- **[Partner SDK Examples](https://github.com/twilio/twilio-agent-connect-python/tree/main/getting_started/examples/partners/)** - Integration examples for OpenAI (Chat Completions, Responses API, Agents SDK), LangChain, AWS Bedrock Agent, AWS Bedrock AgentCore, and AWS Strands
 - **[ConversationRelay-Only Mode](https://github.com/twilio/twilio-agent-connect-python/blob/main/getting_started/examples/features/relay_only.py)** - Get started with voice using just ConversationRelay
 - More examples coming soon
 

getting_started/README.md

@@ -47,6 +47,8 @@ Learn the core pattern for manually extracting and injecting TAC memory into **a
 Production-ready examples integrating TAC with partner SDKs:
 - **`openai_chat_completions.py`**: OpenAI Chat Completions API with automatic memory injection via `with_tac_memory()`
 - **`openai_responses_api.py`**: OpenAI Responses API with automatic memory injection
+- **`openai_agents.py`**: OpenAI Agents SDK integration
+- **`langchain.py`**: LangChain integration
 - **`aws_bedrock_agent.py`**: AWS Bedrock Agent integration
 - **`aws_bedrock_agentcore.py`**: AWS Bedrock AgentCore integration
 - **`aws_strands.py`**: AWS Strands agents integration
@@ -85,6 +87,8 @@ walks up from the script's directory, so it'll find
 uv run getting_started/examples/overview.py
 uv run getting_started/examples/partners/openai_chat_completions.py
 uv run getting_started/examples/partners/openai_responses_api.py
+uv run getting_started/examples/partners/openai_agents.py
+uv run getting_started/examples/partners/langchain.py
 uv run getting_started/examples/partners/aws_bedrock_agent.py
 uv run getting_started/examples/partners/aws_bedrock_agentcore.py
 uv run getting_started/examples/partners/aws_strands.py

getting_started/examples/.env.example

@@ -37,7 +37,7 @@ TWILIO_WHATSAPP_NUMBER=whatsapp:+1xxxxxxxxxx
 # Partner Examples - OpenAI
 # ============================================================================
 
-# Required for partners/openai_*.py examples
+# Required for partners/openai_*.py and partners/langchain.py examples
 OPENAI_API_KEY=sk-proj-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
 
 # ============================================================================

getting_started/examples/partners/langchain.py

@@ -0,0 +1,130 @@
+"""
+Example: Using LangChain with TAC
+
+Demonstrates how to connect Twilio Agent Connect with LangChain
+for voice and SMS channels.
+
+Prerequisites:
+    pip install langchain-core langchain-openai
+
+Environment Variables:
+    OPENAI_API_KEY - Your OpenAI API key
+"""
+
+from dotenv import load_dotenv
+from langchain_core.messages import AIMessage, HumanMessage
+from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
+from langchain_openai import ChatOpenAI
+
+from tac import TAC, TACConfig
+from tac.adapters import MemoryPromptBuilder
+from tac.channels.sms import SMSChannel, SMSChannelConfig
+from tac.channels.voice import VoiceChannel, VoiceChannelConfig
+from tac.core.logging import get_logger
+from tac.models.session import ConversationSession
+from tac.models.tac import TACMemoryResponse
+from tac.server import TACFastAPIServer
+
+load_dotenv()
+
+logger = get_logger(__name__)
+
+# Initialize TAC with configuration from environment variables
+tac = TAC(config=TACConfig.from_env())
+
+# Create channel handlers for Voice and SMS
+voice_channel = VoiceChannel(tac, config=VoiceChannelConfig(memory_mode="once"))
+sms_channel = SMSChannel(tac, config=SMSChannelConfig(memory_mode="always"))
+
+# Create LangChain LLM
+llm = ChatOpenAI(model="gpt-5.4-mini", temperature=0)
+
+# Create prompt template with conversation history support
+prompt_template = ChatPromptTemplate.from_messages(
+    [
+        (
+            "system",
+            "You are a customer service agent speaking with a user over voice or SMS. "
+            "Keep responses short and conversational — a sentence or two. "
+            "Do not use markdown, asterisks, bullets, or emojis; your words will be "
+            "spoken aloud or sent as plain text.",
+        ),
+        MessagesPlaceholder(variable_name="chat_history"),
+        ("user", "{user_message}"),
+    ]
+)
+
+# Create chain using LCEL (pipe operator)
+chain = prompt_template | llm
+
+# Store conversation history per conversation
+conversation_history: dict[str, list[HumanMessage | AIMessage]] = {}
+
+
+async def handle_message_ready(
+    user_message: str,
+    context: ConversationSession,
+    memory_response: TACMemoryResponse | None,
+) -> str:
+    """
+    Callback invoked when a message is ready to be processed.
+
+    This example uses LangChain with LCEL to process the message
+    and maintains conversation history per conversation.
+
+    Args:
+        user_message: The customer's message text
+        context: Session data (conversation_id, channel, profile, etc.)
+        memory_response: Optional retrieved memories (observations, summaries, communications)
+
+    Returns:
+        Response string to send to the channel
+    """
+    conv_id = context.conversation_id
+
+    try:
+        # Initialize conversation history for new conversations
+        if conv_id not in conversation_history:
+            conversation_history[conv_id] = []
+
+        # Compose user message with memory context (user message first, then memory)
+        prompt = MemoryPromptBuilder.compose(
+            system_prompt=user_message,
+            memory_response=memory_response,
+            context=context,
+        )
+
+        # Invoke the chain with conversation history and composed prompt
+        response = await chain.ainvoke(
+            {
+                "chat_history": conversation_history[conv_id],
+                "user_message": prompt,
+            }
+        )
+
+        # Extract text content from the response
+        llm_response = response.content if hasattr(response, "content") else str(response)
+
+        # Update conversation history
+        conversation_history[conv_id].append(HumanMessage(content=user_message))
+        conversation_history[conv_id].append(AIMessage(content=llm_response))
+
+        return llm_response
+
+    except Exception as e:
+        logger.error("Error processing message", conversation_id=conv_id, error=str(e))
+        return "Sorry, I encountered an error processing your message."
+
+
+# Register the message handler callback
+tac.on_message_ready(handle_message_ready)
+
+if __name__ == "__main__":
+    # TACFastAPIServer creates a FastAPI app with all required endpoints:
+    # - /twiml: Voice call webhook (returns TwiML with ConversationRelay)
+    # - /ws: WebSocket endpoint for Voice channel
+    # - /webhook: Conversation webhook for all channels
+    server = TACFastAPIServer(
+        tac=tac, voice_channel=voice_channel, messaging_channels=[sms_channel]
+    )
+    server.start()

getting_started/examples/partners/openai_agents.py

@@ -0,0 +1,102 @@
+"""
+Example: Using OpenAI Agents SDK with TAC
+
+Demonstrates how to connect Twilio Agent Connect with the OpenAI Agents SDK
+for voice and SMS channels.
+
+Prerequisites:
+    pip install openai-agents
+
+Environment Variables:
+    OPENAI_API_KEY - Your OpenAI API key (optional, uses default client if not set)
+"""
+
+from agents import Agent, Runner
+from dotenv import load_dotenv
+
+from tac import TAC, TACConfig
+from tac.adapters import MemoryPromptBuilder
+from tac.channels.sms import SMSChannel, SMSChannelConfig
+from tac.channels.voice import VoiceChannel, VoiceChannelConfig
+from tac.core.logging import get_logger
+from tac.models.session import ConversationSession
+from tac.models.tac import TACMemoryResponse
+from tac.server import TACFastAPIServer
+
+load_dotenv()
+
+logger = get_logger(__name__)
+
+# Initialize TAC with configuration from environment variables
+tac = TAC(config=TACConfig.from_env())
+
+# Create channel handlers for Voice and SMS
+voice_channel = VoiceChannel(tac, config=VoiceChannelConfig(memory_mode="once"))
+sms_channel = SMSChannel(tac, config=SMSChannelConfig(memory_mode="always"))
+
+# Create OpenAI Agent with instructions
+agent = Agent(
+    name="Customer Service Agent",
+    model="gpt-5.4-mini",
+    instructions=(
+        "You are a customer service agent speaking with a user over voice or SMS. "
+        "Keep responses short and conversational — a sentence or two. "
+        "Do not use markdown, asterisks, bullets, or emojis; your words will be "
+        "spoken aloud or sent as plain text."
+    ),
+)
+
+
+async def handle_message_ready(
+    user_message: str,
+    context: ConversationSession,
+    memory_response: TACMemoryResponse | None,
+) -> str:
+    """
+    Callback invoked when a message is ready to be processed.
+
+    This example uses the OpenAI Agents SDK to run a stateless agent.
+
+    Args:
+        user_message: The customer's message text
+        context: Session data (conversation_id, channel, profile, etc.)
+        memory_response: Optional retrieved memories (observations, summaries, communications)
+
+    Returns:
+        Response string to send to the channel
+    """
+    conv_id = context.conversation_id
+
+    try:
+        # Compose user message with memory context (user message first, then memory)
+        prompt = MemoryPromptBuilder.compose(
+            system_prompt=user_message,
+            memory_response=memory_response,
+            context=context,
+        )
+
+        # Run the agent with the composed prompt
+        result = await Runner.run(agent, prompt)
+
+        # Extract the agent's response
+        llm_response = result.final_output or "No response from agent."
+
+        return llm_response
+
+    except Exception as e:
+        logger.error("Error processing message", conversation_id=conv_id, error=str(e))
+        return "Sorry, I encountered an error processing your message."
+
+
+# Register the message handler callback
+tac.on_message_ready(handle_message_ready)
+
+if __name__ == "__main__":
+    # TACFastAPIServer creates a FastAPI app with all required endpoints:
+    # - /twiml: Voice call webhook (returns TwiML with ConversationRelay)
+    # - /ws: WebSocket endpoint for Voice channel
+    # - /webhook: Conversation webhook for all channels
+    server = TACFastAPIServer(
+        tac=tac, voice_channel=voice_channel, messaging_channels=[sms_channel]
+    )
+    server.start()

pyproject.toml

@@ -53,6 +53,9 @@ examples = [
     "boto3>=1.34.0",
     "boto3-stubs[bedrock-agent-runtime,bedrock-agentcore]>=1.34.0",
     "strands-agents>=1.30.0",
+    # LangChain examples
+    "langchain-core>=1.3.0",
+    "langchain-openai>=1.2.0",
     # Common
     "python-dotenv>=1.0.0,<2",
     "fastapi>=0.115.0,<1",