[Improvement] Use SQLite for chat memory (#910)

Co-authored-by: Deven Patel <deven298@yahoo.com>

embedchain/client.py

@@ -5,7 +5,7 @@ import uuid
 
 import requests
 
-from embedchain.embedchain import CONFIG_DIR, CONFIG_FILE
+from embedchain.constants import CONFIG_DIR, CONFIG_FILE
 
 
 class Client:

embedchain/constants.py

@@ -0,0 +1,8 @@
+import os
+from pathlib import Path
+
+ABS_PATH = os.getcwd()
+HOME_DIR = str(Path.home())
+CONFIG_DIR = os.path.join(HOME_DIR, ".embedchain")
+CONFIG_FILE = os.path.join(CONFIG_DIR, "config.json")
+SQLITE_PATH = os.path.join(CONFIG_DIR, "embedchain.db")

embedchain/embedchain.py

@@ -1,9 +1,7 @@
 import hashlib
 import json
 import logging
-import os
 import sqlite3
-from pathlib import Path
 from typing import Any, Dict, List, Optional, Tuple, Union
 
 from dotenv import load_dotenv
@@ -12,6 +10,7 @@ from langchain.docstore.document import Document
 from embedchain.chunkers.base_chunker import BaseChunker
 from embedchain.config import AddConfig, BaseLlmConfig, ChunkerConfig
 from embedchain.config.apps.base_app_config import BaseAppConfig
+from embedchain.constants import SQLITE_PATH
 from embedchain.data_formatter import DataFormatter
 from embedchain.embedder.base import BaseEmbedder
 from embedchain.helper.json_serializable import JSONSerializable
@@ -25,12 +24,6 @@ from embedchain.vectordb.base import BaseVectorDB
 
 load_dotenv()
 
-ABS_PATH = os.getcwd()
-HOME_DIR = str(Path.home())
-CONFIG_DIR = os.path.join(HOME_DIR, ".embedchain")
-CONFIG_FILE = os.path.join(CONFIG_DIR, "config.json")
-SQLITE_PATH = os.path.join(CONFIG_DIR, "embedchain.db")
-
 
 class EmbedChain(JSONSerializable):
     def __init__(
@@ -602,6 +595,9 @@ class EmbedChain(JSONSerializable):
             input_query=input_query, contexts=contexts_data_for_llm_query, config=config, dry_run=dry_run
         )
 
+        # add conversation in memory
+        self.llm.add_history(self.config.id, input_query, answer)
+
         # Send anonymous telemetry
         self.telemetry.capture(event_name="chat", properties=self._telemetry_props)
 
@@ -645,5 +641,9 @@ class EmbedChain(JSONSerializable):
         self.db.reset()
         self.cursor.execute("DELETE FROM data_sources WHERE pipeline_id = ?", (self.config.id,))
         self.connection.commit()
+        self.clear_history()
         # Send anonymous telemetry
         self.telemetry.capture(event_name="reset", properties=self._telemetry_props)
+
+    def clear_history(self):
+        self.llm.memory.delete_chat_history(app_id=self.config.id)

embedchain/llm/base.py

@@ -1,14 +1,15 @@
 import logging
 from typing import Any, Dict, Generator, List, Optional
 
-from langchain.memory import ConversationBufferMemory
-from langchain.schema import BaseMessage
+from langchain.schema import BaseMessage as LCBaseMessage
 
 from embedchain.config import BaseLlmConfig
 from embedchain.config.llm.base import (DEFAULT_PROMPT,
                                         DEFAULT_PROMPT_WITH_HISTORY_TEMPLATE,
                                         DOCS_SITE_PROMPT_TEMPLATE)
 from embedchain.helper.json_serializable import JSONSerializable
+from embedchain.memory.base import ECChatMemory
+from embedchain.memory.message import ChatMessage
 
 
 class BaseLlm(JSONSerializable):
@@ -23,7 +24,7 @@ class BaseLlm(JSONSerializable):
         else:
             self.config = config
 
-        self.memory = ConversationBufferMemory()
+        self.memory = ECChatMemory()
         self.is_docs_site_instance = False
         self.online = False
         self.history: Any = None
@@ -44,11 +45,18 @@ class BaseLlm(JSONSerializable):
         """
         self.history = history
 
-    def update_history(self):
+    def update_history(self, app_id: str):
         """Update class history attribute with history in memory (for chat method)"""
-        chat_history = self.memory.load_memory_variables({})["history"]
+        chat_history = self.memory.get_recent_memories(app_id=app_id, num_rounds=10)
         if chat_history:
-            self.set_history(chat_history)
+            self.set_history([str(history) for history in chat_history])
+
+    def add_history(self, app_id: str, question: str, answer: str, metadata: Optional[Dict[str, Any]] = None):
+        chat_message = ChatMessage()
+        chat_message.add_user_message(question, metadata=metadata)
+        chat_message.add_ai_message(answer, metadata=metadata)
+        self.memory.add(app_id=app_id, chat_message=chat_message)
+        self.update_history(app_id=app_id)
 
     def generate_prompt(self, input_query: str, contexts: List[str], **kwargs: Dict[str, Any]) -> str:
         """
@@ -165,7 +173,6 @@ class BaseLlm(JSONSerializable):
         for chunk in answer:
             streamed_answer = streamed_answer + chunk
             yield chunk
-        self.memory.chat_memory.add_ai_message(streamed_answer)
         logging.info(f"Answer: {streamed_answer}")
 
     def query(self, input_query: str, contexts: List[str], config: BaseLlmConfig = None, dry_run=False):
@@ -257,8 +264,6 @@ class BaseLlm(JSONSerializable):
             if self.online:
                 k["web_search_result"] = self.access_search_and_get_results(input_query)
 
-            self.update_history()
-
             prompt = self.generate_prompt(input_query, contexts, **k)
             logging.info(f"Prompt: {prompt}")
 
@@ -267,16 +272,9 @@ class BaseLlm(JSONSerializable):
 
             answer = self.get_answer_from_llm(prompt)
 
-            self.memory.chat_memory.add_user_message(input_query)
-
             if isinstance(answer, str):
-                self.memory.chat_memory.add_ai_message(answer)
                 logging.info(f"Answer: {answer}")
 
-                # NOTE: Adding to history before and after. This could be seen as redundant.
-                # If we change it, we have to change the tests (no big deal).
-                self.update_history()
-
                 return answer
             else:
                 # this is a streamed response and needs to be handled differently.
@@ -287,7 +285,7 @@ class BaseLlm(JSONSerializable):
                 self.config: BaseLlmConfig = BaseLlmConfig.deserialize(prev_config)
 
     @staticmethod
-    def _get_messages(prompt: str, system_prompt: Optional[str] = None) -> List[BaseMessage]:
+    def _get_messages(prompt: str, system_prompt: Optional[str] = None) -> List[LCBaseMessage]:
         """
         Construct a list of langchain messages
 

embedchain/memory/base.py

@@ -0,0 +1,112 @@
+import json
+import logging
+import sqlite3
+import uuid
+from typing import Any, Dict, List, Optional
+
+from embedchain.constants import SQLITE_PATH
+from embedchain.memory.message import ChatMessage
+from embedchain.memory.utils import merge_metadata_dict
+
+CHAT_MESSAGE_CREATE_TABLE_QUERY = """
+            CREATE TABLE IF NOT EXISTS chat_history (
+                app_id TEXT,
+                id TEXT,
+                question TEXT,
+                answer TEXT,
+                metadata TEXT,
+                created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+                PRIMARY KEY (id, app_id)
+            )
+            """
+
+
+class ECChatMemory:
+    def __init__(self) -> None:
+        with sqlite3.connect(SQLITE_PATH) as self.connection:
+            self.cursor = self.connection.cursor()
+
+            self.cursor.execute(CHAT_MESSAGE_CREATE_TABLE_QUERY)
+            self.connection.commit()
+
+    def add(self, app_id, chat_message: ChatMessage) -> Optional[str]:
+        memory_id = str(uuid.uuid4())
+        metadata_dict = merge_metadata_dict(chat_message.human_message.metadata, chat_message.ai_message.metadata)
+        if metadata_dict:
+            metadata = self._serialize_json(metadata_dict)
+        ADD_CHAT_MESSAGE_QUERY = """
+            INSERT INTO chat_history (app_id, id, question, answer, metadata)
+            VALUES (?, ?, ?, ?, ?)
+        """
+        self.cursor.execute(
+            ADD_CHAT_MESSAGE_QUERY,
+            (
+                app_id,
+                memory_id,
+                chat_message.human_message.content,
+                chat_message.ai_message.content,
+                metadata if metadata_dict else "{}",
+            ),
+        )
+        self.connection.commit()
+        logging.info(f"Added chat memory to db with id: {memory_id}")
+        return memory_id
+
+    def delete_chat_history(self, app_id: str):
+        DELETE_CHAT_HISTORY_QUERY = """
+            DELETE FROM chat_history WHERE app_id=?
+        """
+        self.cursor.execute(
+            DELETE_CHAT_HISTORY_QUERY,
+            (app_id,),
+        )
+        self.connection.commit()
+
+    def get_recent_memories(self, app_id, num_rounds=10) -> List[ChatMessage]:
+        """
+        Get the most recent num_rounds rounds of conversations
+        between human and AI, for a given app_id.
+        """
+
+        QUERY = """
+            SELECT * FROM chat_history
+            WHERE app_id=?
+            ORDER BY created_at DESC
+            LIMIT ?
+        """
+        self.cursor.execute(
+            QUERY,
+            (app_id, num_rounds),
+        )
+
+        results = self.cursor.fetchall()
+        history = []
+        for result in results:
+            app_id, id, question, answer, metadata, timestamp = result
+            metadata = self._deserialize_json(metadata=metadata)
+            memory = ChatMessage()
+            memory.add_user_message(question, metadata=metadata)
+            memory.add_ai_message(answer, metadata=metadata)
+            history.append(memory)
+        return history
+
+    def _serialize_json(self, metadata: Dict[str, Any]):
+        return json.dumps(metadata)
+
+    def _deserialize_json(self, metadata: str):
+        return json.loads(metadata)
+
+    def close_connection(self):
+        self.connection.close()
+
+    def count_history_messages(self, app_id: str):
+        QUERY = """
+        SELECT COUNT(*) FROM chat_history
+        WHERE app_id=?
+        """
+        self.cursor.execute(
+            QUERY,
+            (app_id,),
+        )
+        count = self.cursor.fetchone()[0]
+        return count

embedchain/memory/message.py

@@ -0,0 +1,72 @@
+import logging
+from typing import Any, Dict, Optional
+
+from embedchain.helper.json_serializable import JSONSerializable
+
+
+class BaseMessage(JSONSerializable):
+    """
+    The base abstract message class.
+
+    Messages are the inputs and outputs of Models.
+    """
+
+    # The string content of the message.
+    content: str
+
+    # The creator of the message. AI, Human, Bot etc.
+    by: str
+
+    # Any additional info.
+    metadata: Dict[str, Any]
+
+    def __init__(self, content: str, creator: str, metadata: Optional[Dict[str, Any]] = None) -> None:
+        super().__init__()
+        self.content = content
+        self.creator = creator
+        self.metadata = metadata
+
+    @property
+    def type(self) -> str:
+        """Type of the Message, used for serialization."""
+
+    @classmethod
+    def is_lc_serializable(cls) -> bool:
+        """Return whether this class is serializable."""
+        return True
+
+    def __str__(self) -> str:
+        return f"{self.creator}: {self.content}"
+
+
+class ChatMessage(JSONSerializable):
+    """
+    The base abstract chat message class.
+
+    Chat messages are the pair of (question, answer) conversation
+    between human and model.
+    """
+
+    human_message: Optional[BaseMessage] = None
+    ai_message: Optional[BaseMessage] = None
+
+    def add_user_message(self, message: str, metadata: Optional[dict] = None):
+        if self.human_message:
+            logging.info(
+                "Human message already exists in the chat message,\
+                overwritting it with new message."
+            )
+
+        self.human_message = BaseMessage(content=message, creator="human", metadata=metadata)
+
+    def add_ai_message(self, message: str, metadata: Optional[dict] = None):
+        if self.ai_message:
+            logging.info(
+                "AI message already exists in the chat message,\
+                overwritting it with new message."
+            )
+
+        self.ai_message = BaseMessage(content=message, creator="ai", metadata=metadata)
+
+    def __str__(self) -> str:
+        return f"{self.human_message} | {self.ai_message}"

embedchain/memory/utils.py

@@ -0,0 +1,35 @@
+from typing import Any, Dict, Optional
+
+
+def merge_metadata_dict(left: Optional[Dict[str, Any]], right: Optional[Dict[str, Any]]) -> Optional[Dict[str, Any]]:
+    """
+    Merge the metadatas of two BaseMessage types.
+
+    Args:
+        left (Dict[str, Any]): metadata of human message
+        right (Dict[str, Any]): metadata of ai message
+
+    Returns:
+        Dict[str, Any]: combined metadata dict with dedup
+        to be saved in db.
+    """
+    if not left and not right:
+        return None
+    elif not left:
+        return right
+    elif not right:
+        return left
+
+    merged = left.copy()
+    for k, v in right.items():
+        if k not in merged:
+            merged[k] = v
+        elif type(merged[k]) != type(v):
+            raise ValueError(f'additional_kwargs["{k}"] already exists in this message,' " but with a different type.")
+        elif isinstance(merged[k], str):
+            merged[k] += v
+        elif isinstance(merged[k], dict):
+            merged[k] = merge_metadata_dict(merged[k], v)
+        else:
+            raise ValueError(f"Additional kwargs key {k} already exists in this message.")
+    return merged

embedchain/pipeline.py

@@ -10,7 +10,8 @@ import yaml
 
 from embedchain import Client
 from embedchain.config import ChunkerConfig, PipelineConfig
-from embedchain.embedchain import CONFIG_DIR, EmbedChain
+from embedchain.constants import SQLITE_PATH
+from embedchain.embedchain import EmbedChain
 from embedchain.embedder.base import BaseEmbedder
 from embedchain.embedder.openai import OpenAIEmbedder
 from embedchain.factory import EmbedderFactory, LlmFactory, VectorDBFactory
@@ -22,8 +23,6 @@ from embedchain.utils import validate_yaml_config
 from embedchain.vectordb.base import BaseVectorDB
 from embedchain.vectordb.chroma import ChromaDB
 
-SQLITE_PATH = os.path.join(CONFIG_DIR, "embedchain.db")
-
 
 @register_deserializable
 class Pipeline(EmbedChain):

embedchain/utils.py

@@ -138,7 +138,8 @@ def detect_datatype(source: Any) -> DataType:
     formatted_source = format_source(str(source), 30)
 
     if url:
-        from langchain.document_loaders.youtube import ALLOWED_NETLOCK as YOUTUBE_ALLOWED_NETLOCS
+        from langchain.document_loaders.youtube import \
+            ALLOWED_NETLOCK as YOUTUBE_ALLOWED_NETLOCS
 
         if url.netloc in YOUTUBE_ALLOWED_NETLOCS:
             logging.debug(f"Source of `{formatted_source}` detected as `youtube_video`.")