diff --git a/README.md b/README.md index 5dc477e3..ba80818d 100644 --- a/README.md +++ b/README.md @@ -57,7 +57,7 @@ Your support is greatly appreciated - 🗃️ File storage and summarization with GPT-3.5 ## 📋 Requirements -- [Python 3.7 or later](https://www.tutorialspoint.com/how-to-install-python-in-windows) +- [Python 3.8 or later](https://www.tutorialspoint.com/how-to-install-python-in-windows) - OpenAI API key - PINECONE API key @@ -141,6 +141,40 @@ 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 +``` +See https://hub.docker.com/r/redis/redis-stack-server for setting a password and additional configuration. + +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 diff --git a/requirements.txt b/requirements.txt index 7b104040..6a9ba643 100644 --- a/requirements.txt +++ b/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 diff --git a/scripts/chat.py b/scripts/chat.py index 8da074c6..a27fbfd7 100644 --- a/scripts/chat.py +++ b/scripts/chat.py @@ -26,8 +26,11 @@ def create_chat_message(role, content): def generate_context(prompt, relevant_memory, full_message_history, model): current_context = [ create_chat_message( - "system", prompt), create_chat_message( - "system", f"Permanent memory: {relevant_memory}")] + "system", prompt), + create_chat_message( + "system", f"The current time and date is {time.strftime('%c')}"), + create_chat_message( + "system", f"This reminds you of these events from your past:\n{relevant_memory}\n\n")] # Add messages from the full message history until we reach the token limit next_message_to_add_index = len(full_message_history) - 1 @@ -95,7 +98,7 @@ def chat_with_ai( # Count the currently used tokens current_tokens_used += tokens_to_add - + # Move to the next most recent message in the full message history next_message_to_add_index -= 1 diff --git a/scripts/commands.py b/scripts/commands.py index a45fb896..ba538395 100644 --- a/scripts/commands.py +++ b/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): @@ -108,7 +109,7 @@ def execute_command(command_name, arguments): elif command_name == "task_complete": shutdown() else: - return f"Unknown command {command_name}" + return f"Unknown command '{command_name}'. Please refer to the 'COMMANDS' list for availabe commands and only respond in the specified JSON format." # All errors, return "Error: + error message" except Exception as e: return "Error: " + str(e) diff --git a/scripts/config.py b/scripts/config.py index 959c3eb2..4d7adec1 100644 --- a/scripts/config.py +++ b/scripts/config.py @@ -1,3 +1,4 @@ +import abc import os import openai from dotenv import load_dotenv @@ -5,7 +6,7 @@ from dotenv import load_dotenv load_dotenv() -class Singleton(type): +class Singleton(abc.ABCMeta, type): """ Singleton metaclass for ensuring only one instance of a class. """ @@ -20,12 +21,17 @@ class Singleton(type): 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. """ def __init__(self): + self.debug = False self.continuous_mode = False self.speak_mode = False # TODO - make these models be self-contained, using langchain, so we can configure them once and call it good @@ -59,7 +65,14 @@ class Config(metaclass=Singleton): # 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 @@ -98,3 +111,6 @@ class Config(metaclass=Singleton): def set_pinecone_region(self, value: str): self.pinecone_region = value + + def set_debug_mode(self, value: bool): + self.debug = value diff --git a/scripts/data/prompt.txt b/scripts/data/prompt.txt index 363342c0..77a449de 100644 --- a/scripts/data/prompt.txt +++ b/scripts/data/prompt.txt @@ -18,7 +18,7 @@ COMMANDS: 12. Append to file: "append_to_file", args: "file": "", "text": "" 13. Delete file: "delete_file", args: "file": "" 14. Search Files: "search_files", args: "directory": "" -15. Evaluate Code: "evaluate_code", args: "code": "" +15. Evaluate Code: "evaluate_code", args: "code": "" 16. Get Improved Code: "improve_code", args: "suggestions": "", "code": "" 17. Write Tests: "write_tests", args: "code": "", "focus": "" 18. Execute Python File: "execute_python_file", args: "file": "" diff --git a/scripts/json_parser.py b/scripts/json_parser.py index 8ec9238b..c863ccdb 100644 --- a/scripts/json_parser.py +++ b/scripts/json_parser.py @@ -40,7 +40,7 @@ def fix_and_parse_json(json_str: str, try_to_fix_with_gpt: bool = True): if try_to_fix_with_gpt: print(f"Warning: Failed to parse AI output, attempting to fix.\n If you see this warning frequently, it's likely that your prompt is confusing the AI. Try changing it up slightly.") # Now try to fix this up using the ai_functions - ai_fixed_json = fix_json(json_str, json_schema, False) + ai_fixed_json = fix_json(json_str, json_schema, cfg.debug) if ai_fixed_json != "failed": return json.loads(ai_fixed_json) else: diff --git a/scripts/main.py b/scripts/main.py index 4d68b450..d2cdd530 100644 --- a/scripts/main.py +++ b/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 @@ -123,7 +123,7 @@ def load_variables(config_file="config.yaml"): if ai_name == "": ai_name = "Entrepreneur-GPT" - if not ai_role: + if not ai_role: ai_role = input(f"{ai_name} is: ") if ai_role == "": ai_role = "an AI designed to autonomously develop and run businesses with the sole goal of increasing your net worth." @@ -140,7 +140,7 @@ def load_variables(config_file="config.yaml"): ai_goals.append(ai_goal) if len(ai_goals) == 0: ai_goals = ["Increase net worth", "Grow Twitter Account", "Develop and manage multiple businesses autonomously"] - + # Save variables to yaml file config = {"ai_name": ai_name, "ai_role": ai_role, "ai_goals": ai_goals} with open(config_file, "w") as file: @@ -166,22 +166,22 @@ def construct_prompt(): Fore.GREEN, f"Would you like me to return to being {config.ai_name}?", speak_text=True) - should_continue = input(f"""Continue with the last settings? + should_continue = input(f"""Continue with the last settings? Name: {config.ai_name} Role: {config.ai_role} -Goals: {config.ai_goals} +Goals: {config.ai_goals} Continue (y/n): """) if should_continue.lower() == "n": config = AIConfig() - if not config.ai_name: + if not config.ai_name: config = prompt_user() config.save() # Get rid of this global: global ai_name ai_name = config.ai_name - + full_prompt = config.construct_full_prompt() return full_prompt @@ -242,7 +242,7 @@ def parse_arguments(): global cfg cfg.set_continuous_mode(False) cfg.set_speak_mode(False) - + parser = argparse.ArgumentParser(description='Process arguments.') parser.add_argument('--continuous', action='store_true', help='Enable Continuous Mode') parser.add_argument('--speak', action='store_true', help='Enable Speak Mode') @@ -266,6 +266,10 @@ def parse_arguments(): print_to_console("GPT3.5 Only Mode: ", Fore.GREEN, "ENABLED") cfg.set_smart_llm_model(cfg.fast_llm_model) + if args.debug: + print_to_console("Debug Mode: ", Fore.GREEN, "ENABLED") + cfg.set_debug_mode(True) + # TODO: fill in llm values here @@ -281,12 +285,9 @@ next_action_count = 0 # 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 @@ -298,7 +299,7 @@ while True: user_input, full_message_history, memory, - cfg.fast_token_limit) # TODO: This hardcodes the model to use GPT3.5. Make this an argument + cfg.fast_token_limit, cfg.debug) # TODO: This hardcodes the model to use GPT3.5. Make this an argument # Print Assistant thoughts print_assistant_thoughts(assistant_reply) @@ -358,7 +359,7 @@ while True: f"COMMAND = {Fore.CYAN}{command_name}{Style.RESET_ALL} ARGUMENTS = {Fore.CYAN}{arguments}{Style.RESET_ALL}") # Execute command - if command_name is not None and command_name.lower() == "error": + if command_name is not None and command_name.lower().startswith( "error" ): result = f"Command {command_name} threw the following error: " + arguments elif command_name == "human_feedback": result = f"Human feedback: {user_input}" diff --git a/scripts/memory/__init__.py b/scripts/memory/__init__.py new file mode 100644 index 00000000..a441a46a --- /dev/null +++ b/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", +] diff --git a/scripts/memory/base.py b/scripts/memory/base.py new file mode 100644 index 00000000..d7ab7fcf --- /dev/null +++ b/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 diff --git a/scripts/memory/local.py b/scripts/memory/local.py new file mode 100644 index 00000000..8dc90021 --- /dev/null +++ b/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 diff --git a/scripts/memory.py b/scripts/memory/pinecone.py similarity index 80% rename from scripts/memory.py rename to scripts/memory/pinecone.py index 0d265a31..8e1eaa57 100644 --- a/scripts/memory.py +++ b/scripts/memory/pinecone.py @@ -1,21 +1,11 @@ -from config import Config, Singleton + import pinecone -import openai -cfg = Config() +from memory.base import MemoryProviderSingleton, get_ada_embedding -def get_ada_embedding(text): - text = text.replace("\n", " ") - return openai.Embedding.create(input=[text], model="text-embedding-ada-002")["data"][0]["embedding"] - - -def get_text_from_embedding(embedding): - return openai.Embedding.retrieve(embedding, model="text-embedding-ada-002")["data"][0]["text"] - - -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) diff --git a/scripts/memory/redismem.py b/scripts/memory/redismem.py new file mode 100644 index 00000000..2082fe58 --- /dev/null +++ b/scripts/memory/redismem.py @@ -0,0 +1,143 @@ +"""Redis memory provider.""" +from typing import Any, List, Optional +import redis +from redis.commands.search.field import VectorField, TextField +from redis.commands.search.query import Query +from redis.commands.search.indexDefinition import IndexDefinition, IndexType +import numpy as np + +from memory.base import MemoryProviderSingleton, get_ada_embedding + + +SCHEMA = [ + TextField("data"), + VectorField( + "embedding", + "HNSW", + { + "TYPE": "FLOAT32", + "DIM": 1536, + "DISTANCE_METRIC": "COSINE" + } + ), +] + + +class RedisMemory(MemoryProviderSingleton): + def __init__(self, cfg): + """ + Initializes the Redis memory provider. + + Args: + cfg: The config object. + + Returns: None + """ + redis_host = cfg.redis_host + redis_port = cfg.redis_port + redis_password = cfg.redis_password + self.dimension = 1536 + self.redis = redis.Redis( + host=redis_host, + port=redis_port, + password=redis_password, + db=0 # Cannot be changed + ) + self.cfg = cfg + if cfg.wipe_redis_on_start: + self.redis.flushall() + try: + self.redis.ft(f"{cfg.memory_index}").create_index( + fields=SCHEMA, + definition=IndexDefinition( + prefix=[f"{cfg.memory_index}:"], + index_type=IndexType.HASH + ) + ) + except Exception as e: + print("Error creating Redis search index: ", e) + existing_vec_num = self.redis.get(f'{cfg.memory_index}-vec_num') + self.vec_num = int(existing_vec_num.decode('utf-8')) if\ + existing_vec_num else 0 + + def add(self, data: str) -> str: + """ + Adds a data point to the memory. + + Args: + data: The data to add. + + Returns: Message indicating that the data has been added. + """ + if 'Command Error:' in data: + return "" + vector = get_ada_embedding(data) + vector = np.array(vector).astype(np.float32).tobytes() + data_dict = { + b"data": data, + "embedding": vector + } + pipe = self.redis.pipeline() + pipe.hset(f"{self.cfg.memory_index}:{self.vec_num}", mapping=data_dict) + _text = f"Inserting data into memory at index: {self.vec_num}:\n"\ + f"data: {data}" + self.vec_num += 1 + pipe.set(f'{self.cfg.memory_index}-vec_num', self.vec_num) + pipe.execute() + return _text + + 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 clear(self) -> str: + """ + Clears the redis server. + + Returns: A message indicating that the memory has been cleared. + """ + self.redis.flushall() + return "Obliviated" + + def get_relevant( + self, + data: str, + num_relevant: int = 5 + ) -> Optional[List[Any]]: + """ + Returns all the data in the memory that is relevant to the given data. + Args: + data: The data to compare to. + num_relevant: The number of relevant data to return. + + Returns: A list of the most relevant data. + """ + query_embedding = get_ada_embedding(data) + base_query = f"*=>[KNN {num_relevant} @embedding $vector AS vector_score]" + query = Query(base_query).return_fields( + "data", + "vector_score" + ).sort_by("vector_score").dialect(2) + query_vector = np.array(query_embedding).astype(np.float32).tobytes() + + try: + results = self.redis.ft(f"{self.cfg.memory_index}").search( + query, query_params={"vector": query_vector} + ) + except Exception as e: + print("Error calling Redis search: ", e) + return None + return [result.data for result in results.docs] + + def get_stats(self): + """ + Returns: The stats of the memory index. + """ + return self.redis.ft(f"{self.cfg.memory_index}").info()