Merge pull request #372 from BillSchumacher/redis-backend

Implement Local Cache and Redis Memory backend. 
Removes dependence on Pinecone

README.md

@@ -141,6 +141,39 @@ export CUSTOM_SEARCH_ENGINE_ID="YOUR_CUSTOM_SEARCH_ENGINE_ID"
 
 ```
 
+## Redis Setup
+
+Install docker desktop.
+
+Run:
+```
+docker run -d --name redis-stack-server -p 6379:6379 redis/redis-stack-server:latest
+```
+
+Set the following environment variables:
+```
+MEMORY_BACKEND=redis
+REDIS_HOST=localhost
+REDIS_PORT=6379
+REDIS_PASSWORD=
+```
+
+Note that this is not intended to be run facing the internet and is not secure, do not expose redis to the internet without a password or at all really.
+
+You can optionally set
+
+```
+WIPE_REDIS_ON_START=False
+```
+
+To persist memory stored in Redis.
+
+You can specify the memory index for redis using the following:
+
+````
+MEMORY_INDEX=whatever
+````
+
 ## 🌲 Pinecone API Key Setup
 
 Pinecone enable a vector based memory so a vast memory can be stored and only relevant memories

requirements.txt

@@ -12,4 +12,6 @@ docker
 duckduckgo-search
 google-api-python-client #(https://developers.google.com/custom-search/v1/overview)
 pinecone-client==2.2.1
+redis
+orjson
 Pillow

scripts/commands.py

@@ -1,6 +1,6 @@
 import browse
 import json
-from memory import PineconeMemory
+from memory import get_memory
 import datetime
 import agent_manager as agents
 import speak
@@ -53,10 +53,11 @@ def get_command(response):
 
 
 def execute_command(command_name, arguments):
-    memory = PineconeMemory()
+    memory = get_memory(cfg)
+
     try:
         if command_name == "google":
-            
+
             # Check if the Google API key is set and use the official search method
             # If the API key is not set or has only whitespaces, use the unofficial search method
             if cfg.google_api_key and (cfg.google_api_key.strip() if cfg.google_api_key else None):

scripts/config.py

@@ -1,11 +1,12 @@
+import abc
 import os
 import openai
 from dotenv import load_dotenv
 # Load environment variables from .env file
 load_dotenv()
 
 
-class Singleton(type):
+class Singleton(abc.ABCMeta, type):
     """
     Singleton metaclass for ensuring only one instance of a class.
     """
@@ -20,6 +21,10 @@ def __call__(cls, *args, **kwargs):
         return cls._instances[cls]
 
 
+class AbstractSingleton(abc.ABC, metaclass=Singleton):
+    pass
+
+
 class Config(metaclass=Singleton):
     """
     Configuration class to store the state of bools for different scripts access.
@@ -59,7 +64,14 @@ def __init__(self):
         # User agent headers to use when browsing web
         # Some websites might just completely deny request with an error code if no user agent was found.
         self.user_agent_header = {"User-Agent":"Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_4) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/83.0.4103.97 Safari/537.36"}
-
+        self.redis_host = os.getenv("REDIS_HOST", "localhost")
+        self.redis_port = os.getenv("REDIS_PORT", "6379")
+        self.redis_password = os.getenv("REDIS_PASSWORD", "")
+        self.wipe_redis_on_start = os.getenv("WIPE_REDIS_ON_START", "True") == 'True'
+        self.memory_index = os.getenv("MEMORY_INDEX", 'auto-gpt')
+        # Note that indexes must be created on db 0 in redis, this is not configureable.
+
+        self.memory_backend = os.getenv("MEMORY_BACKEND", 'local')
         # Initialize the OpenAI API client
         openai.api_key = self.openai_api_key
 

scripts/main.py

@@ -1,7 +1,7 @@
 import json
 import random
 import commands as cmd
-from memory import PineconeMemory
+from memory import get_memory
 import data
 import chat
 from colorama import Fore, Style
@@ -281,12 +281,9 @@ def parse_arguments():
 # Make a constant:
 user_input = "Determine which next command to use, and respond using the format specified above:"
 
-# raise an exception if pinecone_api_key or region is not provided
-if not cfg.pinecone_api_key or not cfg.pinecone_region: raise Exception("Please provide pinecone_api_key and pinecone_region")
 # Initialize memory and make sure it is empty.
 # this is particularly important for indexing and referencing pinecone memory
-memory = PineconeMemory()
-memory.clear()
+memory = get_memory(cfg, init=True)
 print('Using memory of type: ' + memory.__class__.__name__)
 
 # Interaction Loop

scripts/memory/__init__.py

@@ -0,0 +1,44 @@
+from memory.local import LocalCache
+try:
+    from memory.redismem import RedisMemory
+except ImportError:
+    print("Redis not installed. Skipping import.")
+    RedisMemory = None
+
+try:
+    from memory.pinecone import PineconeMemory
+except ImportError:
+    print("Pinecone not installed. Skipping import.")
+    PineconeMemory = None
+
+
+def get_memory(cfg, init=False):
+    memory = None
+    if cfg.memory_backend == "pinecone":
+        if not PineconeMemory:
+            print("Error: Pinecone is not installed. Please install pinecone"
+                  " to use Pinecone as a memory backend.")
+        else:
+            memory = PineconeMemory(cfg)
+            if init:
+                memory.clear()
+    elif cfg.memory_backend == "redis":
+        if not RedisMemory:
+            print("Error: Redis is not installed. Please install redis-py to"
+                  " use Redis as a memory backend.")
+        else:
+            memory = RedisMemory(cfg)
+
+    if memory is None:
+        memory = LocalCache(cfg)
+        if init:
+            memory.clear()
+    return memory
+
+
+__all__ = [
+    "get_memory",
+    "LocalCache",
+    "RedisMemory",
+    "PineconeMemory",
+]

scripts/memory/base.py

@@ -0,0 +1,31 @@
+"""Base class for memory providers."""
+import abc
+from config import AbstractSingleton
+import openai
+
+
+def get_ada_embedding(text):
+    text = text.replace("\n", " ")
+    return openai.Embedding.create(input=[text], model="text-embedding-ada-002")["data"][0]["embedding"]
+
+
+class MemoryProviderSingleton(AbstractSingleton):
+    @abc.abstractmethod
+    def add(self, data):
+        pass
+
+    @abc.abstractmethod
+    def get(self, data):
+        pass
+
+    @abc.abstractmethod
+    def clear(self):
+        pass
+
+    @abc.abstractmethod
+    def get_relevant(self, data, num_relevant=5):
+        pass
+
+    @abc.abstractmethod
+    def get_stats(self):
+        pass

scripts/memory/local.py

@@ -0,0 +1,114 @@
+import dataclasses
+import orjson
+from typing import Any, List, Optional
+import numpy as np
+import os
+from memory.base import MemoryProviderSingleton, get_ada_embedding
+
+
+EMBED_DIM = 1536
+SAVE_OPTIONS = orjson.OPT_SERIALIZE_NUMPY | orjson.OPT_SERIALIZE_DATACLASS
+
+
+def create_default_embeddings():
+    return np.zeros((0, EMBED_DIM)).astype(np.float32)
+
+
+@dataclasses.dataclass
+class CacheContent:
+    texts: List[str] = dataclasses.field(default_factory=list)
+    embeddings: np.ndarray = dataclasses.field(
+        default_factory=create_default_embeddings
+    )
+
+
+class LocalCache(MemoryProviderSingleton):
+
+    # on load, load our database
+    def __init__(self, cfg) -> None:
+        self.filename = f"{cfg.memory_index}.json"
+        if os.path.exists(self.filename):
+            with open(self.filename, 'rb') as f:
+                loaded = orjson.loads(f.read())
+                self.data = CacheContent(**loaded)
+        else:
+            self.data = CacheContent()
+
+    def add(self, text: str):
+        """
+        Add text to our list of texts, add embedding as row to our
+            embeddings-matrix
+
+        Args:
+            text: str
+
+        Returns: None
+        """
+        if 'Command Error:' in text:
+            return ""
+        self.data.texts.append(text)
+
+        embedding = get_ada_embedding(text)
+
+        vector = np.array(embedding).astype(np.float32)
+        vector = vector[np.newaxis, :]
+        self.data.embeddings = np.concatenate(
+            [
+                vector,
+                self.data.embeddings,
+            ],
+            axis=0,
+        )
+
+        with open(self.filename, 'wb') as f:
+            out = orjson.dumps(
+                self.data,
+                option=SAVE_OPTIONS
+            )
+            f.write(out)
+        return text
+
+    def clear(self) -> str:
+        """
+        Clears the redis server.
+
+        Returns: A message indicating that the memory has been cleared.
+        """
+        self.data = CacheContent()
+        return "Obliviated"
+
+    def get(self, data: str) -> Optional[List[Any]]:
+        """
+        Gets the data from the memory that is most relevant to the given data.
+
+        Args:
+            data: The data to compare to.
+
+        Returns: The most relevant data.
+        """
+        return self.get_relevant(data, 1)
+
+    def get_relevant(self, text: str, k: int) -> List[Any]:
+        """"
+        matrix-vector mult to find score-for-each-row-of-matrix
+         get indices for top-k winning scores
+         return texts for those indices
+        Args:
+            text: str
+            k: int
+
+        Returns: List[str]
+        """
+        embedding = get_ada_embedding(text)
+
+        scores = np.dot(self.data.embeddings, embedding)
+
+        top_k_indices = np.argsort(scores)[-k:][::-1]
+
+        return [self.data.texts[i] for i in top_k_indices]
+
+    def get_stats(self):
+        """
+        Returns: The stats of the local cache.
+        """
+        return len(self.data.texts), self.data.embeddings.shape

scripts/memory.py → scripts/memory/pinecone.py

@@ -1,21 +1,11 @@
-from config import Config, Singleton
-import pinecone
-import openai
-
-cfg = Config()
-
-
-def get_ada_embedding(text):
-    text = text.replace("\n", " ")
-    return openai.Embedding.create(input=[text], model="text-embedding-ada-002")["data"][0]["embedding"]
 
+import pinecone
 
-def get_text_from_embedding(embedding):
-    return openai.Embedding.retrieve(embedding, model="text-embedding-ada-002")["data"][0]["text"]
+from memory.base import MemoryProviderSingleton, get_ada_embedding
 
 
-class PineconeMemory(metaclass=Singleton):
-    def __init__(self):
+class PineconeMemory(MemoryProviderSingleton):
+    def __init__(self, cfg):
         pinecone_api_key = cfg.pinecone_api_key
         pinecone_region = cfg.pinecone_region
         pinecone.init(api_key=pinecone_api_key, environment=pinecone_region)