diff --git a/packages/powermem-langchain/README.md b/packages/powermem-langchain/README.md
index 8cd1f28fd1..e26e346f4e 100644
--- a/packages/powermem-langchain/README.md
+++ b/packages/powermem-langchain/README.md
@@ -1,9 +1,9 @@
 # powermem-langchain
 
-This package is the LangChain middleware exercise for the VLDB 2026 summer
-school branch. It provides a package skeleton, a no-op middleware scaffold,
-contract tests, and a runnable OpenAI example. It does not provide a complete
-PowerMem middleware implementation.
+This package provides PowerMem long-term memory middleware for LangChain v1
+agents. It retrieves relevant memories before an agent run, injects them into
+model-visible context, and can save the final user/assistant interaction after
+the run.
 
 The goal is to integrate PowerMem as a long-term memory layer for LangChain v1
 agents. The provided scaffold is intentionally small; you may adjust it as your
@@ -13,7 +13,7 @@ This is useful beyond a single `create_agent` example: LangChain middleware is
 the extension point for controlling agent execution, and related projects such
 as Deep Agents also compose capabilities through middleware.
 
-## Task
+## Usage
 
 Implement:
 
@@ -44,7 +44,7 @@ agent = create_agent(
 )
 ```
 
-At minimum, the implementation should:
+The middleware:
 
 - Search PowerMem with the latest user message before the model call.
 - Add relevant memories to the model-visible context.
@@ -52,6 +52,13 @@ At minimum, the implementation should:
 - Skip write-back when `save_interactions=False`.
 - Use the explicit `user_id` constructor argument.
 - Keep the agent usable when PowerMem search fails.
+- Support both `invoke()` and `ainvoke()` agent execution.
+
+`user_id` is required and must be supplied explicitly. This keeps application
+identity separate from LangChain runtime or checkpoint state. Retrieved memory
+is added to the model request's system context without modifying the persisted
+conversation messages. Interaction write-back uses `infer=False`, preserving
+the final user and assistant messages as explicit PowerMem records.
 
 ## Tests
 
@@ -112,6 +119,5 @@ uv run --no-project \
     --user-id summer-school-demo
 ```
 
-With the placeholder middleware, the command can run but will not show the expected
-memory injection or write-back behavior. After implementation, the output should
-show seeded memories before the agent call and updated memories afterward.
+The output should show seeded memories before the agent call, an answer informed
+by those memories, and updated memories after the call.
diff --git a/packages/powermem-langchain/src/powermem_langchain/middleware.py b/packages/powermem-langchain/src/powermem_langchain/middleware.py
index 59c43c99f5..14807d0460 100644
--- a/packages/powermem-langchain/src/powermem_langchain/middleware.py
+++ b/packages/powermem-langchain/src/powermem_langchain/middleware.py
@@ -1,31 +1,38 @@
-"""LangChain middleware entry point for PowerMem.
-
-The VLDB 2026 summer school branch intentionally provides only the public entry
-point. Students are expected to replace this placeholder with a LangChain
-middleware implementation that satisfies the package contract tests.
-"""
+"""LangChain middleware that provides PowerMem-backed long-term memory."""
 
 from __future__ import annotations
 
+import inspect
+import logging
+from asyncio import to_thread
+from collections.abc import Awaitable, Callable, Mapping, Sequence
 from typing import Any, NotRequired, TypedDict
 
-from langchain.agents.middleware import AgentMiddleware, AgentState
+from langchain.agents.middleware import (
+    AgentMiddleware,
+    AgentState,
+    ModelRequest,
+    ModelResponse,
+)
+from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, SystemMessage
+
+logger = logging.getLogger(__name__)
 
 
 class PowerMemState(AgentState):
-    """State schema reserved for the PowerMem middleware implementation."""
+    """Agent state used to carry retrieved memory into model-call hooks."""
 
     powermem_context: NotRequired[str]
 
 
 class PowerMemStateUpdate(TypedDict):
-    """State update returned by memory-loading middleware hooks."""
+    """State update returned after loading relevant memories."""
 
     powermem_context: str
 
 
 class PowerMemMiddleware(AgentMiddleware[PowerMemState, Any, Any]):
-    """Placeholder for the summer school implementation."""
+    """Use a PowerMem client as a LangChain agent's long-term memory layer."""
 
     state_schema = PowerMemState
 
@@ -38,17 +45,225 @@ def __init__(
         save_interactions: bool = True,
         **kwargs: Any,
     ) -> None:
-        pass
+        if kwargs:
+            unexpected = ", ".join(sorted(kwargs))
+            raise TypeError(f"Unexpected PowerMemMiddleware arguments: {unexpected}")
+        if memory is None:
+            raise ValueError("memory must be a PowerMem client")
+        if not isinstance(user_id, str) or not user_id.strip():
+            raise ValueError("user_id must be a non-empty explicit user identity")
+        if isinstance(search_limit, bool) or not isinstance(search_limit, int):
+            raise TypeError("search_limit must be an integer")
+        if search_limit < 1:
+            raise ValueError("search_limit must be at least 1")
+
+        self.memory = memory
+        self.user_id = user_id
+        self.search_limit = search_limit
+        self.save_interactions = save_interactions
 
     def before_agent(self, state: PowerMemState, runtime) -> PowerMemStateUpdate | None:
-        pass
+        del runtime
+        query = _latest_message_text(state.get("messages", []), HumanMessage)
+        if not query:
+            return None
+
+        try:
+            result = self.memory.search(
+                query=query,
+                user_id=self.user_id,
+                limit=self.search_limit,
+            )
+            if inspect.isawaitable(result):
+                _close_awaitable(result)
+                raise TypeError(
+                    "The configured PowerMem client is asynchronous; use agent.ainvoke()"
+                )
+        except Exception:
+            logger.warning("PowerMem search failed; continuing without memory", exc_info=True)
+            return None
+
+        context = _format_search_result(result)
+        return {"powermem_context": context} if context else None
 
     async def abefore_agent(
         self,
         state: PowerMemState,
         runtime,
     ) -> PowerMemStateUpdate | None:
-        pass
+        del runtime
+        query = _latest_message_text(state.get("messages", []), HumanMessage)
+        if not query:
+            return None
+
+        try:
+            result = await _call_async_compatible(
+                self.memory.search,
+                query=query,
+                user_id=self.user_id,
+                limit=self.search_limit,
+            )
+        except Exception:
+            logger.warning("PowerMem search failed; continuing without memory", exc_info=True)
+            return None
+
+        context = _format_search_result(result)
+        return {"powermem_context": context} if context else None
+
+    def wrap_model_call(
+        self,
+        request: ModelRequest[Any],
+        handler: Callable[[ModelRequest[Any]], ModelResponse[Any]],
+    ) -> ModelResponse[Any]:
+        return handler(_inject_memory_context(request))
+
+    async def awrap_model_call(
+        self,
+        request: ModelRequest[Any],
+        handler: Callable[[ModelRequest[Any]], Awaitable[ModelResponse[Any]]],
+    ) -> ModelResponse[Any]:
+        return await handler(_inject_memory_context(request))
 
     def after_agent(self, state: PowerMemState, runtime) -> None:
-        pass
+        del runtime
+        if not self.save_interactions:
+            return
+
+        interaction = _final_interaction(state.get("messages", []))
+        if interaction is None:
+            return
+
+        result = self.memory.add(
+            messages=interaction,
+            user_id=self.user_id,
+            infer=False,
+        )
+        if inspect.isawaitable(result):
+            _close_awaitable(result)
+            raise TypeError(
+                "The configured PowerMem client is asynchronous; use agent.ainvoke()"
+            )
+
+    async def aafter_agent(self, state: PowerMemState, runtime) -> None:
+        del runtime
+        if not self.save_interactions:
+            return
+
+        interaction = _final_interaction(state.get("messages", []))
+        if interaction is None:
+            return
+
+        await _call_async_compatible(
+            self.memory.add,
+            messages=interaction,
+            user_id=self.user_id,
+            infer=False,
+        )
+
+
+async def _call_async_compatible(function: Callable[..., Any], **kwargs: Any) -> Any:
+    """Call native async methods directly and move sync SDK calls off the event loop."""
+
+    if inspect.iscoroutinefunction(function):
+        return await function(**kwargs)
+
+    result = await to_thread(function, **kwargs)
+    if inspect.isawaitable(result):
+        return await result
+    return result
+
+
+def _close_awaitable(awaitable: Any) -> None:
+    close = getattr(awaitable, "close", None)
+    if callable(close):
+        close()
+
+
+def _latest_message_text(
+    messages: Sequence[Any],
+    message_type: type[BaseMessage],
+) -> str | None:
+    for message in reversed(messages):
+        if isinstance(message, message_type):
+            text = _message_text(message)
+            if text:
+                return text
+    return None
+
+
+def _message_text(message: BaseMessage) -> str:
+    text = message.text
+    if isinstance(text, str):
+        return text.strip()
+    return str(text).strip()
+
+
+def _format_search_result(result: Any) -> str:
+    if not isinstance(result, Mapping):
+        return ""
+
+    raw_memories = result.get("results", [])
+    if not isinstance(raw_memories, Sequence) or isinstance(raw_memories, (str, bytes)):
+        return ""
+
+    memories: list[str] = []
+    for item in raw_memories:
+        if isinstance(item, Mapping):
+            value = item.get("memory") or item.get("content")
+        else:
+            value = item
+        if value is not None and (text := str(value).strip()):
+            memories.append(text)
+
+    if not memories:
+        return ""
+
+    lines = [
+        "Relevant long-term memories from PowerMem:",
+        "Use these only when they are relevant to the current request.",
+    ]
+    lines.extend(f"- {memory}" for memory in memories)
+    return "\n".join(lines)
+
+
+def _inject_memory_context(request: ModelRequest[Any]) -> ModelRequest[Any]:
+    context = request.state.get("powermem_context")
+    if not context:
+        return request
+
+    existing = request.system_message
+    if existing is None:
+        system_message = SystemMessage(content=context)
+    elif isinstance(existing.content, str):
+        separator = "\n\n" if existing.content else ""
+        system_message = existing.model_copy(
+            update={"content": f"{existing.content}{separator}{context}"}
+        )
+    else:
+        content = [*existing.content, {"type": "text", "text": context}]
+        system_message = existing.model_copy(update={"content": content})
+
+    return request.override(system_message=system_message)
+
+
+def _final_interaction(messages: Sequence[Any]) -> list[dict[str, str]] | None:
+    assistant_index: int | None = None
+    assistant_text: str | None = None
+    for index in range(len(messages) - 1, -1, -1):
+        message = messages[index]
+        if isinstance(message, AIMessage) and (text := _message_text(message)):
+            assistant_index = index
+            assistant_text = text
+            break
+
+    if assistant_index is None or assistant_text is None:
+        return None
+
+    user_text = _latest_message_text(messages[:assistant_index], HumanMessage)
+    if not user_text:
+        return None
+
+    return [
+        {"role": "user", "content": user_text},
+        {"role": "assistant", "content": assistant_text},
+    ]
diff --git a/packages/powermem-langchain/tests/test_middleware.py b/packages/powermem-langchain/tests/test_middleware.py
index 4f2fbbcfd9..07bbf0b6d2 100644
--- a/packages/powermem-langchain/tests/test_middleware.py
+++ b/packages/powermem-langchain/tests/test_middleware.py
@@ -7,7 +7,7 @@
 import pytest
 from langchain.agents import create_agent
 from langchain_core.language_models.chat_models import SimpleChatModel
-from langchain_core.messages import BaseMessage, HumanMessage
+from langchain_core.messages import BaseMessage, HumanMessage, SystemMessage
 from powermem import Memory
 from powermem_langchain import PowerMemMiddleware
 from pydantic import Field
@@ -38,6 +38,21 @@ def search(self, *args: Any, **kwargs: Any) -> dict[str, Any]:
         raise RuntimeError("search failed")
 
 
+class RecordingMemory:
+    def __init__(self, results: list[dict[str, Any]] | None = None) -> None:
+        self.results = results or []
+        self.search_calls: list[dict[str, Any]] = []
+        self.add_calls: list[dict[str, Any]] = []
+
+    def search(self, **kwargs: Any) -> dict[str, Any]:
+        self.search_calls.append(kwargs)
+        return {"results": self.results}
+
+    def add(self, **kwargs: Any) -> dict[str, Any]:
+        self.add_calls.append(kwargs)
+        return {"results": []}
+
+
 def _message_text(messages: list[BaseMessage]) -> str:
     return "\n".join(str(message.content) for message in messages)
 
@@ -162,6 +177,67 @@ def test_search_failure_is_fail_open_by_default():
     assert result["messages"][-1].content == "Agent still runs"
 
 
+def test_uses_latest_user_message_and_explicit_user_id():
+    memory = RecordingMemory()
+    model = CapturingChatModel(responses=["ok"])
+    agent = create_agent(
+        model=model,
+        tools=[],
+        middleware=[
+            PowerMemMiddleware(
+                memory=memory,
+                user_id="explicit-user",
+                search_limit=3,
+            )
+        ],
+    )
+
+    agent.invoke(
+        {
+            "messages": [
+                HumanMessage(content="Older question"),
+                HumanMessage(content="Newest question"),
+            ]
+        }
+    )
+
+    assert memory.search_calls == [
+        {"query": "Newest question", "user_id": "explicit-user", "limit": 3}
+    ]
+    assert memory.add_calls[0]["user_id"] == "explicit-user"
+
+
+def test_preserves_existing_system_prompt_when_injecting_memory():
+    memory = RecordingMemory([{"memory": "PowerMem context"}])
+    model = CapturingChatModel(responses=["ok"])
+    agent = create_agent(
+        model=model,
+        tools=[],
+        system_prompt="Original system prompt",
+        middleware=[
+            PowerMemMiddleware(
+                memory=memory,
+                user_id="alice",
+                save_interactions=False,
+            )
+        ],
+    )
+
+    agent.invoke({"messages": [HumanMessage(content="Question")]})
+
+    system_messages = [
+        message for message in model.calls[0] if isinstance(message, SystemMessage)
+    ]
+    assert len(system_messages) == 1
+    assert "Original system prompt" in str(system_messages[0].content)
+    assert "PowerMem context" in str(system_messages[0].content)
+
+
+def test_rejects_missing_user_id():
+    with pytest.raises(ValueError, match="user_id"):
+        PowerMemMiddleware(memory=RecordingMemory())
+
+
 @pytest.mark.asyncio
 async def test_async_agent_uses_powermem_memory(tmp_path: Path):
     memory = _sqlite_memory(tmp_path)
@@ -183,3 +259,27 @@ async def test_async_agent_uses_powermem_memory(tmp_path: Path):
     await agent.ainvoke({"messages": [HumanMessage(content="Use my async profile.")]})
 
     assert "User prefers async examples." in _message_text(model.calls[0])
+
+
+@pytest.mark.asyncio
+async def test_async_agent_persists_interaction():
+    memory = RecordingMemory()
+    model = CapturingChatModel(responses=["Async response"])
+    agent = create_agent(
+        model=model,
+        tools=[],
+        middleware=[PowerMemMiddleware(memory=memory, user_id="async-user")],
+    )
+
+    await agent.ainvoke({"messages": [HumanMessage(content="Async request")]})
+
+    assert memory.add_calls == [
+        {
+            "messages": [
+                {"role": "user", "content": "Async request"},
+                {"role": "assistant", "content": "Async response"},
+            ],
+            "user_id": "async-user",
+            "infer": False,
+        }
+    ]