feat: quivr core chat history (#2824)

# Description

- Defined quivr-core `ChatHistory`
- `ChatHistory` can be iterated over in tuples of
`HumanMessage,AIMessage`
-  Brain appends to the chatHistory once response is received
- Brain holds a dict of chats and defines the default chat (TODO: define
a system of selecting the chats)
- Wrote test 
- Updated `QuivrQARAG` to use `ChatHistory` as input

backend/core/quivr_core/brain.py

@@ -6,8 +6,10 @@ from uuid import UUID, uuid4
 
 from langchain_core.documents import Document
 from langchain_core.embeddings import Embeddings
+from langchain_core.messages import AIMessage, HumanMessage
 from langchain_core.vectorstores import VectorStore
 
+from quivr_core.chat import ChatHistory
 from quivr_core.config import LLMEndpointConfig, RAGConfig
 from quivr_core.llm import LLMEndpoint
 from quivr_core.models import ParsedRAGResponse, SearchResult
@@ -108,13 +110,23 @@ class Brain:
         self.storage = storage
 
         # Chat history
-        self.chat_history: list[str] = []
+        self._chats = self._init_chats()
+        self.default_chat = list(self._chats.values())[0]
 
         # RAG dependencies:
         self.llm = llm
         self.vector_db = vector_db
         self.embedder = embedder
 
+    @property
+    def chat_history(self):
+        return self.default_chat
+
+    def _init_chats(self):
+        chat_id = uuid4()
+        default_chat = ChatHistory(chat_id=chat_id, brain_id=self.id)
+        return {chat_id: default_chat}
+
     @classmethod
     async def afrom_files(
         cls,
@@ -235,6 +247,9 @@ class Brain:
 
         return [SearchResult(chunk=d, score=s) for d, s in result]
 
+    def get_chat_history(self, chat_id: UUID):
+        return self._chats[chat_id]
+
     # TODO(@aminediro)
     def add_file(self) -> None:
         # add it to storage
@@ -242,7 +257,9 @@ class Brain:
         raise NotImplementedError
 
     def ask(
-        self, question: str, rag_config: RAGConfig | None = None
+        self,
+        question: str,
+        rag_config: RAGConfig | None = None,
     ) -> ParsedRAGResponse:
         llm = self.llm
 
@@ -257,9 +274,12 @@ class Brain:
             rag_config=rag_config, llm=llm, vector_store=self.vector_db
         )
 
-        # transformed_history = format_chat_history(history)
+        chat_history = self.default_chat
 
-        parsed_response = rag_pipeline.answer(question, [], [])
+        parsed_response = rag_pipeline.answer(question, chat_history, [])
+
+        chat_history.append(HumanMessage(content=question))
+        chat_history.append(AIMessage(content=parsed_response.answer))
 
         # Save answer to the chat history
         return parsed_response

backend/core/quivr_core/chat.py

@@ -0,0 +1,50 @@
+from datetime import datetime
+from typing import Any, Generator, Tuple
+from uuid import UUID, uuid4
+
+from langchain_core.messages import AIMessage, HumanMessage
+
+from quivr_core.models import ChatMessage
+
+
+class ChatHistory:
+    def __init__(self, chat_id: UUID, brain_id: UUID) -> None:
+        self.id = chat_id
+        self.brain_id = brain_id
+        # TODO(@aminediro): maybe use a deque() instead ?
+        self._msgs: list[ChatMessage] = []
+
+    def get_chat_history(self, newest_first: bool = False):
+        """Returns a ChatMessage list sorted by time
+
+        Returns:
+            list[ChatMessage]: list of chat messages
+        """
+        history = sorted(self._msgs, key=lambda msg: msg.message_time)
+        if newest_first:
+            return history[::-1]
+        return history
+
+    def __len__(self):
+        return len(self._msgs)
+
+    def append(
+        self, langchain_msg: AIMessage | HumanMessage, metadata: dict[str, Any] = {}
+    ):
+        chat_msg = ChatMessage(
+            chat_id=self.id,
+            message_id=uuid4(),
+            brain_id=self.brain_id,
+            msg=langchain_msg,
+            message_time=datetime.now(),
+            metadata=metadata,
+        )
+        self._msgs.append(chat_msg)
+
+    def iter_pairs(self) -> Generator[Tuple[HumanMessage, AIMessage], None, None]:
+        # Reverse the chat_history, newest first
+        it = iter(self.get_chat_history(newest_first=True))
+        for ai_message, human_message in zip(it, it):
+            assert isinstance(human_message.msg, HumanMessage)
+            assert isinstance(ai_message.msg, AIMessage)
+            yield (human_message.msg, ai_message.msg)

backend/core/quivr_core/config.py

@@ -12,5 +12,6 @@ class LLMEndpointConfig(BaseModel):
 
 class RAGConfig(BaseModel):
     llm_config: LLMEndpointConfig = LLMEndpointConfig()
+    max_history: int = 10
     max_files: int = 20
     prompt: str | None = None

backend/core/quivr_core/models.py

@@ -3,6 +3,7 @@ from typing import Any
 from uuid import UUID
 
 from langchain_core.documents import Document
+from langchain_core.messages import AIMessage, HumanMessage
 from langchain_core.pydantic_v1 import BaseModel as BaseModelV1
 from langchain_core.pydantic_v1 import Field as FieldV1
 from pydantic import BaseModel
@@ -33,17 +34,13 @@ class cited_answer(BaseModelV1):
     )
 
 
-class GetChatHistoryOutput(BaseModel):
+class ChatMessage(BaseModelV1):
     chat_id: UUID
     message_id: UUID
-    user_message: str
+    brain_id: UUID
+    msg: AIMessage | HumanMessage
     message_time: datetime
-    assistant: str | None = None
-    prompt_title: str | None = None
-    brain_name: str | None = None
-    brain_id: UUID | None = None  # string because UUID is not JSON serializable
-    metadata: dict | None = None
-    thumbs: bool | None = None
+    metadata: dict[str, Any]
 
 
 class Source(BaseModel):

backend/core/quivr_core/quivr_rag.py

@@ -5,11 +5,13 @@ from typing import AsyncGenerator, Optional, Sequence
 from langchain.retrievers import ContextualCompressionRetriever
 from langchain_core.callbacks import Callbacks
 from langchain_core.documents import BaseDocumentCompressor, Document
+from langchain_core.messages import AIMessage, HumanMessage
 from langchain_core.messages.ai import AIMessageChunk
 from langchain_core.output_parsers import StrOutputParser
 from langchain_core.runnables import RunnableLambda, RunnablePassthrough
 from langchain_core.vectorstores import VectorStore
 
+from quivr_core.chat import ChatHistory
 from quivr_core.config import RAGConfig
 from quivr_core.llm import LLMEndpoint
 from quivr_core.models import (
@@ -59,7 +61,8 @@ class QuivrQARAG:
         return self.vector_store.as_retriever()
 
     def filter_history(
-        self, chat_history, max_history: int = 10, max_tokens: int = 2000
+        self,
+        chat_history: ChatHistory,
     ):
         """
         Filter out the chat history to only include the messages that are relevant to the current question
@@ -68,29 +71,23 @@ class QuivrQARAG:
         Returns a filtered chat_history with in priority: first max_tokens, then max_history where a Human message and an AI message count as one pair
         a token is 4 characters
         """
-        chat_history = chat_history[::-1]
         total_tokens = 0
         total_pairs = 0
-        filtered_chat_history = []
-        for i in range(0, len(chat_history), 2):
-            if i + 1 < len(chat_history):
-                human_message = chat_history[i]
-                ai_message = chat_history[i + 1]
-                message_tokens = (
-                    len(human_message.content) + len(ai_message.content)
-                ) // 4
-                if (
-                    total_tokens + message_tokens > max_tokens
-                    or total_pairs >= max_history
-                ):
-                    break
-                filtered_chat_history.append(human_message)
-                filtered_chat_history.append(ai_message)
-                total_tokens += message_tokens
-                total_pairs += 1
-        chat_history = filtered_chat_history[::-1]
+        filtered_chat_history: list[AIMessage | HumanMessage] = []
+        for human_message, ai_message in chat_history.iter_pairs():
+            # TODO: replace with tiktoken
+            message_tokens = (len(human_message.content) + len(ai_message.content)) // 4
+            if (
+                total_tokens + message_tokens > self.rag_config.llm_config.max_tokens
+                or total_pairs >= self.rag_config.max_history
+            ):
+                break
+            filtered_chat_history.append(human_message)
+            filtered_chat_history.append(ai_message)
+            total_tokens += message_tokens
+            total_pairs += 1
 
-        return chat_history
+        return filtered_chat_history[::-1]
 
     def build_chain(self, files: str):
         compression_retriever = ContextualCompressionRetriever(
@@ -146,7 +143,7 @@ class QuivrQARAG:
     def answer(
         self,
         question: str,
-        history: list[dict[str, str]],
+        history: ChatHistory,
         list_files: list[QuivrKnowledge],
         metadata: dict[str, str] = {},
     ) -> ParsedRAGResponse:
@@ -166,7 +163,7 @@ class QuivrQARAG:
     async def answer_astream(
         self,
         question: str,
-        history: list[dict[str, str]],
+        history: ChatHistory,
         list_files: list[QuivrKnowledge],
         metadata: dict[str, str] = {},
     ) -> AsyncGenerator[ParsedRAGChunkResponse, ParsedRAGChunkResponse]:

backend/core/quivr_core/utils.py

@@ -11,7 +11,7 @@ from langchain.schema import (
 from langchain_core.messages.ai import AIMessageChunk
 
 from quivr_core.models import (
-    GetChatHistoryOutput,
+    ChatMessage,
     ParsedRAGChunkResponse,
     ParsedRAGResponse,
     QuivrKnowledge,
@@ -44,7 +44,7 @@ def model_supports_function_calling(model_name: str):
 
 
 def format_chat_history(
-    history: List[GetChatHistoryOutput],
+    history: List[ChatMessage],
 ) -> List[Dict[str, str]]:
     """Format the chat history into a list of HumanMessage and AIMessage"""
     formatted_history = []

backend/core/tests/conftest.py

@@ -0,0 +1,8 @@
+import os
+
+import pytest
+
+
+@pytest.fixture(scope="session", autouse=True)
+def openai_api_key():
+    os.environ["OPENAI_API_KEY"] = "abcd"

backend/core/tests/test_brain.py

@@ -6,6 +6,7 @@ from langchain_core.embeddings import DeterministicFakeEmbedding, Embeddings
 from langchain_core.language_models import FakeListChatModel
 
 from quivr_core.brain import Brain
+from quivr_core.chat import ChatHistory
 from quivr_core.config import LLMEndpointConfig
 from quivr_core.llm import LLMEndpoint
 from quivr_core.storage.local_storage import TransparentStorage
@@ -25,7 +26,7 @@ def answers():
 
 
 @pytest.fixture(scope="function")
-def llm(answers: list[str]):
+def fake_llm(answers: list[str]):
     llm = FakeListChatModel(responses=answers)
     return LLMEndpoint(llm=llm, llm_config=LLMEndpointConfig(model="fake_model"))
 
@@ -41,23 +42,26 @@ def test_brain_empty_files():
         Brain.from_files(name="test_brain", file_paths=[])
 
 
-def test_brain_from_files_success(llm: LLMEndpoint, embedder, temp_data_file):
+def test_brain_from_files_success(fake_llm: LLMEndpoint, embedder, temp_data_file):
     brain = Brain.from_files(
-        name="test_brain", file_paths=[temp_data_file], embedder=embedder, llm=llm
+        name="test_brain", file_paths=[temp_data_file], embedder=embedder, llm=fake_llm
     )
     assert brain.name == "test_brain"
-    assert brain.chat_history == []
-    assert brain.llm == llm
+    assert len(brain.chat_history) == 0
+    assert brain.llm == fake_llm
     assert brain.vector_db.embeddings == embedder
+    assert isinstance(brain.default_chat, ChatHistory)
+    assert len(brain.default_chat) == 0
+
     # storage
     assert isinstance(brain.storage, TransparentStorage)
     assert len(brain.storage.get_files()) == 1
 
 
 @pytest.mark.asyncio
-async def test_brain_ask_txt(llm: LLMEndpoint, embedder, temp_data_file, answers):
+async def test_brain_ask_txt(fake_llm: LLMEndpoint, embedder, temp_data_file, answers):
     brain = await Brain.afrom_files(
-        name="test_brain", file_paths=[temp_data_file], embedder=embedder, llm=llm
+        name="test_brain", file_paths=[temp_data_file], embedder=embedder, llm=fake_llm
     )
     answer = brain.ask("question")
 
@@ -68,7 +72,7 @@ async def test_brain_ask_txt(llm: LLMEndpoint, embedder, temp_data_file, answers
 
 
 @pytest.mark.asyncio
-async def test_brain_from_langchain_docs(llm, embedder):
+async def test_brain_from_langchain_docs(embedder):
     chunk = Document("content_1", metadata={"id": uuid4()})
     brain = await Brain.afrom_langchain_documents(
         name="test", langchain_documents=[chunk], embedder=embedder
@@ -92,3 +96,17 @@ async def test_brain_search(
     assert len(result) == 1
     assert result[0].chunk == chunk
     assert result[0].score == 0
+
+
+@pytest.mark.asyncio
+async def test_brain_get_history(
+    fake_llm: LLMEndpoint, embedder, temp_data_file, answers
+):
+    brain = await Brain.afrom_files(
+        name="test_brain", file_paths=[temp_data_file], embedder=embedder, llm=fake_llm
+    )
+
+    brain.ask("question")
+    brain.ask("question")
+
+    assert len(brain.default_chat) == 4

backend/core/tests/test_chat_history.py

@@ -0,0 +1,77 @@
+from time import sleep
+from uuid import uuid4
+
+import pytest
+from langchain_core.messages import AIMessage, HumanMessage
+
+from quivr_core.chat import ChatHistory
+
+
+@pytest.fixture
+def ai_message():
+    return AIMessage("ai message")
+
+
+@pytest.fixture
+def human_message():
+    return HumanMessage("human message")
+
+
+def test_chat_history_constructor():
+    brain_id, chat_id = uuid4(), uuid4()
+    chat_history = ChatHistory(brain_id=brain_id, chat_id=chat_id)
+
+    assert chat_history.brain_id == brain_id
+    assert chat_history.id == chat_id
+    assert len(chat_history._msgs) == 0
+
+
+def test_chat_history_append(ai_message: AIMessage, human_message: HumanMessage):
+    chat_history = ChatHistory(uuid4(), uuid4())
+    chat_history.append(ai_message)
+
+    assert len(chat_history) == 1
+    chat_history.append(human_message)
+    assert len(chat_history) == 2
+
+
+def test_chat_history_get_history(ai_message: AIMessage, human_message: HumanMessage):
+    chat_history = ChatHistory(uuid4(), uuid4())
+    chat_history.append(ai_message)
+    chat_history.append(human_message)
+    chat_history.append(ai_message)
+    sleep(0.01)
+    chat_history.append(human_message)
+
+    msgs = chat_history.get_chat_history()
+
+    assert len(msgs) == 4
+    assert msgs[-1].message_time > msgs[0].message_time
+    assert isinstance(msgs[0].msg, AIMessage)
+    assert isinstance(msgs[1].msg, HumanMessage)
+
+    msgs = chat_history.get_chat_history(newest_first=True)
+    assert msgs[-1].message_time < msgs[0].message_time
+
+
+def test_chat_history_iter_pairs_invalid(
+    ai_message: AIMessage, human_message: HumanMessage
+):
+    with pytest.raises(AssertionError):
+        chat_history = ChatHistory(uuid4(), uuid4())
+        chat_history.append(ai_message)
+        chat_history.append(ai_message)
+        next(chat_history.iter_pairs())
+
+
+def test_chat_history_iter_pais(ai_message: AIMessage, human_message: HumanMessage):
+    chat_history = ChatHistory(uuid4(), uuid4())
+
+    chat_history.append(human_message)
+    chat_history.append(ai_message)
+    chat_history.append(human_message)
+    chat_history.append(ai_message)
+
+    result = list(chat_history.iter_pairs())
+
+    assert result == [(human_message, ai_message), (human_message, ai_message)]