removes as_chain references to chain in docs and updates RAG tests an…

…d docs

django_ai_assistant/helpers/assistants.py

@@ -8,10 +8,6 @@
     DEFAULT_DOCUMENT_SEPARATOR,
 )
 from langchain.tools import StructuredTool
-from langchain_core.chat_history import (
-    BaseChatMessageHistory,
-    InMemoryChatMessageHistory,
-)
 from langchain_core.language_models import BaseChatModel
 from langchain_core.messages import (
     AIMessage,
@@ -81,8 +77,6 @@ class AIAssistant(abc.ABC):  # noqa: F821
     """Whether the assistant uses RAG (Retrieval-Augmented Generation) or not.\n
     Defaults to `False`.
     When True, the assistant will use a retriever to get documents to provide as context to the LLM.
-    For this to work, the `instructions` should contain a placeholder for the context,
-    which is `{context}` by default.
     Additionally, the assistant class should implement the `get_retriever` method to return
     the retriever to use."""
     _user: Any | None
@@ -257,58 +251,6 @@ def get_model_kwargs(self) -> dict[str, Any]:
         """
         return {}
 
-    def get_prompt_template(self) -> ChatPromptTemplate:
-        """Get the `ChatPromptTemplate` for the Langchain chain to use.\n
-        The system prompt comes from the `get_instructions` method.\n
-        The template includes placeholders for the instructions, chat `{history}`, user `{input}`,
-        and `{agent_scratchpad}`, all which are necessary for the chain to work properly.\n
-        The chat history is filled by the chain using the message history from `get_message_history`.\n
-        If the assistant uses RAG, the instructions should contain a placeholder
-        for the context, which is `{context}` by default, defined by the `get_context_placeholder` method.
-
-        Returns:
-            ChatPromptTemplate: The chat prompt template for the Langchain chain.
-        """
-        instructions = self.get_instructions()
-        context_placeholder = self.get_context_placeholder()
-        if self.has_rag and f"{context_placeholder}" not in instructions:
-            raise AIAssistantMisconfiguredError(
-                f"{self.__class__.__name__} has_rag=True"
-                f"but does not have a {{{context_placeholder}}} placeholder in instructions."
-            )
-
-        return ChatPromptTemplate.from_messages(
-            [
-                ("system", instructions),
-                MessagesPlaceholder(variable_name="history"),
-                ("human", "{input}"),
-                ("placeholder", "{agent_scratchpad}"),
-            ]
-        )
-
-    @with_cast_id
-    def get_message_history(self, thread_id: Any | None) -> BaseChatMessageHistory:
-        """Get the chat message history instance for the given `thread_id`.\n
-        The Langchain chain uses the return of this method to get the thread messages
-        for the assistant, filling the `history` placeholder in the `get_prompt_template`.\n
-
-        Args:
-            thread_id (Any | None): The thread ID for the chat message history.
-                If `None`, an in-memory chat message history is used.
-
-        Returns:
-            BaseChatMessageHistory: The chat message history instance for the given `thread_id`.
-        """
-
-        # DjangoChatMessageHistory must be here because Django may not be loaded yet elsewhere:
-        from django_ai_assistant.langchain.chat_message_histories import (
-            DjangoChatMessageHistory,
-        )
-
-        if thread_id is None:
-            return InMemoryChatMessageHistory()
-        return DjangoChatMessageHistory(thread_id)
-
     def get_llm(self) -> BaseChatModel:
         """Get the Langchain LLM instance for the assistant.
         By default, this uses the OpenAI implementation.\n
@@ -359,15 +301,6 @@ def get_document_prompt(self) -> PromptTemplate:
         """
         return DEFAULT_DOCUMENT_PROMPT
 
-    def get_context_placeholder(self) -> str:
-        """Get the RAG context placeholder to use in the prompt when `has_rag=True`.\n
-        Defaults to `"context"`. Override this method to use a different placeholder.
-
-        Returns:
-            str: the RAG context placeholder to use in the prompt.
-        """
-        return "context"
-
     def get_retriever(self) -> BaseRetriever:
         """Get the RAG retriever to use for fetching documents.\n
         Must be implemented by subclasses when `has_rag=True`.\n
@@ -455,15 +388,23 @@ def as_graph(self, thread_id: Any | None = None) -> Runnable[dict, dict]:
         Returns:
             the compiled graph
         """
+        # DjangoChatMessageHistory must be here because Django may not be loaded yet elsewhere.
+        # DjangoChatMessageHistory was used in the context of langchain, now that we are using
+        # langgraph this can be further simplified by just porting the add_messages logic.
+        from django_ai_assistant.langchain.chat_message_histories import (
+            DjangoChatMessageHistory,
+        )
+
+        message_history = DjangoChatMessageHistory(thread_id) if thread_id else None
+
         llm = self.get_llm()
         tools = self.get_tools()
         llm_with_tools = llm.bind_tools(tools) if tools else llm
-        message_history = self.get_message_history(thread_id)
 
         def custom_add_messages(left: list[BaseMessage], right: list[BaseMessage]):
             result = add_messages(left, right)
 
-            if thread_id:
+            if message_history:
                 messages_to_store = [
                     m
                     for m in result
@@ -504,7 +445,7 @@ def retriever(state: AgentState):
             }
 
         def history(state: AgentState):
-            history = message_history.messages if thread_id else []
+            history = message_history.messages if message_history else []
             return {"messages": [*history, HumanMessage(content=state["input"])]}
 
         def agent(state: AgentState):
@@ -551,23 +492,23 @@ def record_response(state: AgentState):
 
     @with_cast_id
     def invoke(self, *args: Any, thread_id: Any | None, **kwargs: Any) -> dict:
-        """Invoke the assistant Langchain chain with the given arguments and keyword arguments.\n
+        """Invoke the assistant Langchain graph with the given arguments and keyword arguments.\n
         This is the lower-level method to run the assistant.\n
-        The chain is created by the `as_graph` method.\n
+        The graph is created by the `as_graph` method.\n
 
         Args:
-            *args: Positional arguments to pass to the chain.
+            *args: Positional arguments to pass to the graph.
                 Make sure to include a `dict` like `{"input": "user message"}`.
             thread_id (Any | None): The thread ID for the chat message history.
                 If `None`, an in-memory chat message history is used.
-            **kwargs: Keyword arguments to pass to the chain.
+            **kwargs: Keyword arguments to pass to the graph.
 
         Returns:
-            dict: The output of the assistant chain,
+            dict: The output of the assistant graph,
                 structured like `{"output": "assistant response", "history": ...}`.
         """
-        chain = self.as_graph(thread_id)
-        return chain.invoke(*args, **kwargs)
+        graph = self.as_graph(thread_id)
+        return graph.invoke(*args, **kwargs)
 
     @with_cast_id
     def run(self, message: str, thread_id: Any | None = None, **kwargs: Any) -> str:
@@ -578,7 +519,7 @@ def run(self, message: str, thread_id: Any | None = None, **kwargs: Any) -> str:
             message (str): The user message to pass to the assistant.
             thread_id (Any | None): The thread ID for the chat message history.
                 If `None`, an in-memory chat message history is used.
-            **kwargs: Additional keyword arguments to pass to the chain.
+            **kwargs: Additional keyword arguments to pass to the graph.
 
         Returns:
             str: The assistant response to the user message.

django_ai_assistant/helpers/use_cases.py

@@ -120,7 +120,7 @@ def create_message(
         content (Any): Message content, usually a string
         request (HttpRequest | None): Current request, if any
     Returns:
-        dict: The output of the assistant chain,
+        dict: The output of the assistant,
             structured like `{"output": "assistant response", "history": ...}`
     Raises:
         AIUserNotAllowedError: If user is not allowed to create messages in the thread

docs/tutorial.md

@@ -1,6 +1,6 @@
 ---
 search:
-  boost: 2 
+    boost: 2
 ---
 
 # Tutorial
@@ -17,8 +17,8 @@ The tutorial below uses OpenAI's gpt-4o model, so make sure you have `OPENAI_API
 You can also use other models, keep reading to learn more. Just make sure their keys are properly set.
 
 !!! note
-    An easy way to set environment variables is to use a `.env` file in your project's root directory and use `python-dotenv` to load them.
-    Our [example project](https://github.com/vintasoftware/django-ai-assistant/tree/main/example#readme) uses this approach.
+An easy way to set environment variables is to use a `.env` file in your project's root directory and use `python-dotenv` to load them.
+Our [example project](https://github.com/vintasoftware/django-ai-assistant/tree/main/example#readme) uses this approach.
 
 ## What AI Assistants can do
 
@@ -90,8 +90,8 @@ AI: The weather in NYC is sunny with a temperature of 25°C.
 ```
 
 !!! note
-    State of the art models such as gpt-4o can process JSON well.
-    You can return a `json.dumps(api_output)` from a tool method and the model will be able to process it before responding the user.
+State of the art models such as gpt-4o can process JSON well.
+You can return a `json.dumps(api_output)` from a tool method and the model will be able to process it before responding the user.
 
 ### Tool parameters
 
@@ -131,7 +131,7 @@ class WeatherAIAssistant(AIAssistant):
 ```
 
 !!! note
-    It's important to provide a `description` for each field from `args_schema`. This improves the LLM's understanding of the tool's arguments.
+It's important to provide a `description` for each field from `args_schema`. This improves the LLM's understanding of the tool's arguments.
 
 ### Using Django logic in tools
 
@@ -173,8 +173,8 @@ class IssueManagementAIAssistant(AIAssistant):
 ```
 
 !!! warning
-    Make sure you only return to the LLM what the user can see, considering permissions and privacy.
-    Code the tools as if they were Django views.
+Make sure you only return to the LLM what the user can see, considering permissions and privacy.
+Code the tools as if they were Django views.
 
 ### Using pre-implemented tools
 
@@ -219,8 +219,8 @@ class MovieSearchAIAssistant(AIAssistant):
 ```
 
 !!! note
-    As of now, Django AI Assistant is powered by [LangChain](https://python.langchain.com/v0.2/docs/introduction/),
-    but previous knowledge on LangChain is NOT necessary to use this library, at least for the main use cases.
+As of now, Django AI Assistant is powered by [LangChain](https://python.langchain.com/v0.2/docs/introduction/),
+but previous knowledge on LangChain is NOT necessary to use this library, at least for the main use cases.
 
 ## Using an AI Assistant
 
@@ -274,7 +274,7 @@ urlpatterns = [
     path("ai-assistant/", include("django_ai_assistant.urls")),
     ...
 ]
-``` 
+```
 
 The built-in API supports retrieval of Assistants info, as well as CRUD for Threads and Messages.
 It has a OpenAPI schema that you can explore at `http://localhost:8000/ai-assistant/docs`, when running your project locally.
@@ -415,15 +415,13 @@ shows an example of a composed AI Assistant that's able to recommend movies and
 ### Retrieval Augmented Generation (RAG)
 
 You can use RAG in your AI Assistants. RAG means using a retriever to fetch chunks of textual data from a pre-existing DB to give
-context to the LLM. This context goes into the `{context}` placeholder in the `instructions` string, namely the system prompt.
-This means the LLM will have access to a context your retriever logic provides when generating the response,
+context to the LLM. This means the LLM will have access to a context your retriever logic provides when generating the response,
 thereby improving the quality of the response by avoiding generic or off-topic answers.
 
 For this to work, your must do the following in your AI Assistant:
 
-1. Add a `{context}` placeholder in the `instructions` string;
-2. Add `has_rag = True` as a class attribute;
-3. Override the `get_retriever` method to return a [Langchain Retriever](https://python.langchain.com/v0.2/docs/how_to/#retrievers).
+1. Add `has_rag = True` as a class attribute;
+2. Override the `get_retriever` method to return a [Langchain Retriever](https://python.langchain.com/v0.2/docs/how_to/#retrievers).
 
 For example:
 
@@ -436,10 +434,6 @@ class DocsAssistant(AIAssistant):
     instructions = (
         "You are an assistant for answering questions related to the provided context. "
         "Use the following pieces of retrieved context to answer the user's question. "
-        "\n\n"
-        "---START OF CONTEXT---\n"
-        "{context}"
-        "---END OF CONTEXT---\n"
     )
     model = "gpt-4o"
     has_rag = True

example/rag/ai_assistants.py

@@ -14,10 +14,6 @@ class DjangoDocsAssistant(AIAssistant):
         "Use the following pieces of retrieved context from Django's documentation to answer "
         "the user's question. If you don't know the answer, say that you don't know. "
         "Use three sentences maximum and keep the answer concise."
-        "\n\n"
-        "---START OF CONTEXT---\n"
-        "{context}"
-        "---END OF CONTEXT---\n"
     )
     model = "gpt-4o"
     has_rag = True