github/AutoGPT
1
1
Fork 0
mirror of https://github.com/Significant-Gravitas/AutoGPT.git synced 2026-02-01 02:15:27 -05:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into security-and-robustness-improvements

Browse Source
This commit is contained in:
Slava Kurilyak (slavakurilyak.eth)
2023-04-09 14:56:09 -06:00
committed by GitHub
parent a4f130ff60 816dc14d82
commit fb36c14517
28 changed files with 884 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
scripts/ai_config.py
View File

@@ -1,6 +1,6 @@
import yaml
import data
import os
class AIConfig:
def __init__(self, ai_name="", ai_role="", ai_goals=[]):
@@ -9,7 +9,7 @@ class AIConfig:
self.ai_goals = ai_goals
# Soon this will go in a folder where it remembers more stuff about the run(s)
SAVE_FILE = "last_run_ai_settings.yaml"
SAVE_FILE = os.path.join(os.path.dirname(__file__), '..', 'ai_settings.yaml')
@classmethod
def load(cls, config_file=SAVE_FILE):

4
scripts/browse.py
View File

@@ -21,7 +21,7 @@ def sanitize_url(url):
# Function to make a request with a specified timeout and handle exceptions
def make_request(url, timeout=10):
try:
response = requests.get(url, timeout=timeout)
response = requests.get(url, headers=cfg.user_agent_header, timeout=timeout)
response.raise_for_status()
return response
except requests.exceptions.RequestException as e:
@@ -69,7 +69,7 @@ def format_hyperlinks(hyperlinks):
def scrape_links(url):
response = requests.get(url)
response = requests.get(url, headers=cfg.user_agent_header)
# Check if the response contains an HTTP error
if response.status_code >= 400:

42
scripts/chat.py
View File

@@ -23,6 +23,22 @@ def create_chat_message(role, content):
return {"role": role, "content": content}
def generate_context(prompt, relevant_memory, full_message_history, model):
current_context = [
create_chat_message(
"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
insertion_index = len(current_context)
# Count the currently used tokens
current_tokens_used = token_counter.count_message_tokens(current_context, model)
return next_message_to_add_index, current_tokens_used, insertion_index, current_context
# TODO: Change debug from hardcode to argument
def chat_with_ai(
@@ -41,7 +57,7 @@ def chat_with_ai(
prompt (str): The prompt explaining the rules to the AI.
user_input (str): The input from the user.
full_message_history (list): The list of all messages sent between the user and the AI.
permanent_memory (list): The list of items in the AI's permanent memory.
permanent_memory (Obj): The memory object containing the permanent memory.
token_limit (int): The maximum number of tokens allowed in the API call.
Returns:
@@ -53,18 +69,20 @@ def chat_with_ai(
print(f"Token limit: {token_limit}")
send_token_limit = token_limit - 1000
current_context = [
create_chat_message(
"system", prompt), create_chat_message(
"system", f"Permanent memory: {permanent_memory}")]
relevant_memory = permanent_memory.get_relevant(str(full_message_history[-5:]), 10)
# Add messages from the full message history until we reach the token limit
next_message_to_add_index = len(full_message_history) - 1
current_tokens_used = 0
insertion_index = len(current_context)
if debug:
print('Memory Stats: ', permanent_memory.get_stats())
next_message_to_add_index, current_tokens_used, insertion_index, current_context = generate_context(
prompt, relevant_memory, full_message_history, model)
while current_tokens_used > 2500:
# remove memories until we are under 2500 tokens
relevant_memory = relevant_memory[1:]
next_message_to_add_index, current_tokens_used, insertion_index, current_context = generate_context(
prompt, relevant_memory, full_message_history, model)
# Count the currently used tokens
current_tokens_used = token_counter.count_message_tokens(current_context, model)
current_tokens_used += token_counter.count_message_tokens([create_chat_message("user", user_input)], model) # Account for user input (appended later)
while next_message_to_add_index >= 0:
@@ -80,7 +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

39
scripts/commands.py
View File

@@ -1,14 +1,15 @@
import browse
import json
import memory as mem
from memory import get_memory
import datetime
import agent_manager as agents
import speak
from config import Config
import ai_functions as ai
from file_operations import read_file, write_to_file, append_to_file, delete_file
from file_operations import read_file, write_to_file, append_to_file, delete_file, search_files
from execute_code import execute_python_file
from json_parser import fix_and_parse_json
from image_gen import generate_image
from duckduckgo_search import ddg
from googleapiclient.discovery import build
from googleapiclient.errors import HttpError
@@ -52,9 +53,11 @@ def get_command(response):
def execute_command(command_name, arguments):
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):
@@ -62,11 +65,7 @@ def execute_command(command_name, arguments):
else:
return google_search(arguments["input"])
elif command_name == "memory_add":
return commit_memory(arguments["string"])
elif command_name == "memory_del":
return delete_memory(arguments["key"])
elif command_name == "memory_ovr":
return overwrite_memory(arguments["key"], arguments["string"])
return memory.add(arguments["string"])
elif command_name == "start_agent":
return start_agent(
arguments["name"],
@@ -90,6 +89,8 @@ def execute_command(command_name, arguments):
return append_to_file(arguments["file"], arguments["text"])
elif command_name == "delete_file":
return delete_file(arguments["file"])
elif command_name == "search_files":
return search_files(arguments["directory"])
elif command_name == "browse_website":
return browse_website(arguments["url"], arguments["question"])
# TODO: Change these to take in a file rather than pasted code, if
@@ -103,10 +104,12 @@ def execute_command(command_name, arguments):
return ai.write_tests(arguments["code"], arguments.get("focus"))
elif command_name == "execute_python_file": # Add this command
return execute_python_file(arguments["file"])
elif command_name == "generate_image":
return generate_image(arguments["prompt"])
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)
@@ -218,7 +221,7 @@ def overwrite_memory(key, string):
elif isinstance(key, str):
_text = "Overwriting memory with key " + key + " and string " + string
# Overwrite the memory slot with the given string key and string
mem.string_key_memory[key] = string
mem.permanent_memory[key] = string
print(_text)
return _text
else:
@@ -256,18 +259,20 @@ def start_agent(name, task, prompt, model=cfg.fast_llm_model):
def message_agent(key, message):
global cfg
# Check if the key is a valid integer
if not is_valid_int(key):
return "Invalid key, cannot message agent."
agent_response = agents.message_agent(int(key), message)
if is_valid_int(key):
agent_response = agents.message_agent(int(key), message)
# Check if the key is a valid string
elif isinstance(key, str):
agent_response = agents.message_agent(key, message)
else:
return "Invalid key, must be an integer or a string."
# Speak response
if cfg.speak_mode:
speak.say_text(agent_response, 1)
return f"Agent {key} responded: {agent_response}"
return agent_response
def list_agents():

47
scripts/config.py
View File

@@ -1,10 +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.
"""
@@ -19,12 +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
@@ -34,11 +41,38 @@ class Config(metaclass=Singleton):
self.smart_token_limit = int(os.getenv("SMART_TOKEN_LIMIT", 8000))
self.openai_api_key = os.getenv("OPENAI_API_KEY")
self.use_azure = False
self.use_azure = os.getenv("USE_AZURE") == 'True'
if self.use_azure:
self.openai_api_base = os.getenv("OPENAI_API_BASE")
self.openai_api_version = os.getenv("OPENAI_API_VERSION")
self.openai_deployment_id = os.getenv("OPENAI_DEPLOYMENT_ID")
openai.api_type = "azure"
openai.api_base = self.openai_api_base
openai.api_version = self.openai_api_version
self.elevenlabs_api_key = os.getenv("ELEVENLABS_API_KEY")
self.google_api_key = os.getenv("GOOGLE_API_KEY")
self.custom_search_engine_id = os.getenv("CUSTOM_SEARCH_ENGINE_ID")
self.pinecone_api_key = os.getenv("PINECONE_API_KEY")
self.pinecone_region = os.getenv("PINECONE_ENV")
self.image_provider = os.getenv("IMAGE_PROVIDER")
self.huggingface_api_token = os.getenv("HUGGINGFACE_API_TOKEN")
# 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
@@ -70,4 +104,13 @@ class Config(metaclass=Singleton):
self.google_api_key = value
def set_custom_search_engine_id(self, value: str):
self.custom_search_engine_id = value
self.custom_search_engine_id = value
def set_pinecone_api_key(self, value: str):
self.pinecone_api_key = value
def set_pinecone_region(self, value: str):
self.pinecone_region = value
def set_debug_mode(self, value: bool):
self.debug = value

11
scripts/data.py
View File

@@ -1,15 +1,14 @@
import os
from pathlib import Path
SRC_DIR = Path(__file__).parent
def load_prompt():
try:
# get directory of this file:
file_dir = Path(os.path.dirname(os.path.realpath(__file__)))
data_dir = file_dir / "data"
prompt_file = data_dir / "prompt.txt"
# Load the promt from data/prompt.txt
with open(SRC_DIR/ "data/prompt.txt", "r") as prompt_file:
file_dir = Path(__file__).parent
prompt_file_path = file_dir / "data" / "prompt.txt"
# Load the prompt from data/prompt.txt
with open(prompt_file_path, "r") as prompt_file:
prompt = prompt_file.read()
return prompt

36
scripts/data/prompt.txt
View File

@@ -1,17 +1,15 @@
CONSTRAINTS:
1. ~4000 word limit for memory. Your memory is short, so immediately save important information to long term memory and code to files.
2. No user assistance
3. Exclusively use the commands listed in double quotes e.g. "command name"
1. ~4000 word limit for short term memory. Your short term memory is short, so immediately save important information to files.
2. If you are unsure how you previously did something or want to recall past events, thinking about similar events will help you remember.
3. No user assistance
4. Exclusively use the commands listed in double quotes e.g. "command name"
COMMANDS:
1. Google Search: "google", args: "input": "<search>"
2. Memory Add: "memory_add", args: "string": "<string>"
3. Memory Delete: "memory_del", args: "key": "<key>"
4. Memory Overwrite: "memory_ovr", args: "key": "<key>", "string": "<string>"
5. Browse Website: "browse_website", args: "url": "<url>", "question": "<what_you_want_to_find_on_website>"
6. Start GPT Agent: "start_agent", args: "name": <name>, "task": "<short_task_desc>", "prompt": "<prompt>"
6. Start GPT Agent: "start_agent", args: "name": "<name>", "task": "<short_task_desc>", "prompt": "<prompt>"
7. Message GPT Agent: "message_agent", args: "key": "<key>", "message": "<message>"
8. List GPT Agents: "list_agents", args: ""
9. Delete GPT Agent: "delete_agent", args: "key": "<key>"
@@ -19,11 +17,13 @@ COMMANDS:
11. Read file: "read_file", args: "file": "<file>"
12. Append to file: "append_to_file", args: "file": "<file>", "text": "<text>"
13. Delete file: "delete_file", args: "file": "<file>"
14. Evaluate Code: "evaluate_code", args: "code": "<full _code_string>"
15. Get Improved Code: "improve_code", args: "suggestions": "<list_of_suggestions>", "code": "<full_code_string>"
16. Write Tests: "write_tests", args: "code": "<full_code_string>", "focus": "<list_of_focus_areas>"
17. Execute Python File: "execute_python_file", args: "file": "<file>"
18. Task Complete (Shutdown): "task_complete", args: "reason": "<reason>"
14. Search Files: "search_files", args: "directory": "<directory>"
15. Evaluate Code: "evaluate_code", args: "code": "<full_code_string>"
16. Get Improved Code: "improve_code", args: "suggestions": "<list_of_suggestions>", "code": "<full_code_string>"
17. Write Tests: "write_tests", args: "code": "<full_code_string>", "focus": "<list_of_focus_areas>"
18. Execute Python File: "execute_python_file", args: "file": "<file>"
19. Task Complete (Shutdown): "task_complete", args: "reason": "<reason>"
20. Generate Image: "generate_image", args: "prompt": "<prompt>"
RESOURCES:
@@ -43,12 +43,6 @@ You should only respond in JSON format as described below
RESPONSE FORMAT:
{
"command": {
"name": "command name",
"args":{
"arg name": "value"
}
},
"thoughts":
{
"text": "thought",
@@ -56,6 +50,12 @@ RESPONSE FORMAT:
"plan": "- short bulleted\n- list that conveys\n- long-term plan",
"criticism": "constructive self-criticism",
"speak": "thoughts summary to say to user"
},
"command": {
"name": "command name",
"args":{
"arg name": "value"
}
}
}

17
scripts/file_operations.py
View File

@@ -58,3 +58,20 @@ def delete_file(filename):
return "File deleted successfully."
except Exception as e:
return "Error: " + str(e)
def search_files(directory):
found_files = []
if directory == "" or directory == "/":
search_directory = working_directory
else:
search_directory = safe_join(working_directory, directory)
for root, _, files in os.walk(search_directory):
for file in files:
if file.startswith('.'):
continue
relative_path = os.path.relpath(os.path.join(root, file), working_directory)
found_files.append(relative_path)
return found_files

57
scripts/image_gen.py Normal file
View File

@@ -0,0 +1,57 @@
import requests
import io
import os.path
from PIL import Image
from config import Config
import uuid
import openai
from base64 import b64decode
cfg = Config()
working_directory = "auto_gpt_workspace"
def generate_image(prompt):
filename = str(uuid.uuid4()) + ".jpg"
# DALL-E
if cfg.image_provider == 'dalle':
openai.api_key = cfg.openai_api_key
response = openai.Image.create(
prompt=prompt,
n=1,
size="256x256",
response_format="b64_json",
)
print("Image Generated for prompt:" + prompt)
image_data = b64decode(response["data"][0]["b64_json"])
with open(working_directory + "/" + filename, mode="wb") as png:
png.write(image_data)
return "Saved to disk:" + filename
# STABLE DIFFUSION
elif cfg.image_provider == 'sd':
API_URL = "https://api-inference.huggingface.co/models/CompVis/stable-diffusion-v1-4"
headers = {"Authorization": "Bearer " + cfg.huggingface_api_token}
response = requests.post(API_URL, headers=headers, json={
"inputs": prompt,
})
image = Image.open(io.BytesIO(response.content))
print("Image Generated for prompt:" + prompt)
image.save(os.path.join(working_directory, filename))
return "Saved to disk:" + filename
else:
return "No Image Provider Set"

8
scripts/json_parser.py
View File

@@ -24,6 +24,7 @@ def fix_and_parse_json(json_str: str, try_to_fix_with_gpt: bool = True):
"""
try:
json_str = json_str.replace('\t', '')
return json.loads(json_str)
except Exception as e:
# Let's do something manually - sometimes GPT responds with something BEFORE the braces:
@@ -39,7 +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:
@@ -51,7 +52,7 @@ def fix_and_parse_json(json_str: str, try_to_fix_with_gpt: bool = True):
def fix_json(json_str: str, schema: str, debug=False) -> str:
# Try to fix the JSON using gpt:
function_string = "def fix_json(json_str: str, schema:str=None) -> str:"
args = [json_str, schema]
args = [f"'''{json_str}'''", f"'''{schema}'''"]
description_string = """Fixes the provided JSON string to make it parseable and fully complient with the provided schema.\n If an object or field specifed in the schema isn't contained within the correct JSON, it is ommited.\n This function is brilliant at guessing when the format is incorrect."""
# If it doesn't already start with a "`", add one:
@@ -67,7 +68,8 @@ def fix_json(json_str: str, schema: str, debug=False) -> str:
print(f"Fixed JSON: {result_string}")
print("----------- END OF FIX ATTEMPT ----------------")
try:
return json.loads(result_string)
json.loads(result_string) # just check the validity
return result_string
except:
# Get the call stack:
# import traceback

21
scripts/llm_utils.py
View File

@@ -6,11 +6,20 @@ openai.api_key = cfg.openai_api_key
# Overly simple abstraction until we create something better
def create_chat_completion(messages, model=None, temperature=None, max_tokens=None)->str:
response = openai.ChatCompletion.create(
model=model,
messages=messages,
temperature=temperature,
max_tokens=max_tokens
)
if cfg.use_azure:
response = openai.ChatCompletion.create(
deployment_id=cfg.openai_deployment_id,
model=model,
messages=messages,
temperature=temperature,
max_tokens=max_tokens
)
else:
response = openai.ChatCompletion.create(
model=model,
messages=messages,
temperature=temperature,
max_tokens=max_tokens
)
return response.choices[0].message["content"]

39
scripts/main.py
View File

@@ -1,7 +1,7 @@
import json
import random
import commands as cmd
import memory as mem
from memory import get_memory
import data
import chat
from colorama import Fore, Style
@@ -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
@@ -277,9 +281,15 @@ prompt = construct_prompt()
# Initialize variables
full_message_history = []
result = None
next_action_count = 0
# Make a constant:
user_input = "Determine which next command to use, and respond using the format specified above:"
# Initialize memory and make sure it is empty.
# this is particularly important for indexing and referencing pinecone memory
memory = get_memory(cfg, init=True)
print('Using memory of type: ' + memory.__class__.__name__)
# Interaction Loop
while True:
# Send message to AI, get response
@@ -288,10 +298,9 @@ while True:
prompt,
user_input,
full_message_history,
mem.permanent_memory,
cfg.fast_token_limit) # TODO: This hardcodes the model to use GPT3.5. Make this an argument
memory,
cfg.fast_token_limit, cfg.debug) # TODO: This hardcodes the model to use GPT3.5. Make this an argument
# print("assistant reply: "+assistant_reply)
# Print Assistant thoughts
print_assistant_thoughts(assistant_reply)
@@ -301,7 +310,7 @@ while True:
except Exception as e:
print_to_console("Error: \n", Fore.RED, str(e))
if not cfg.continuous_mode:
if not cfg.continuous_mode and next_action_count == 0:
### GET USER AUTHORIZATION TO EXECUTE COMMAND ###
# Get key press: Prompt the user to press enter to continue or escape
# to exit
@@ -311,13 +320,21 @@ while True:
Fore.CYAN,
f"COMMAND = {Fore.CYAN}{command_name}{Style.RESET_ALL} ARGUMENTS = {Fore.CYAN}{arguments}{Style.RESET_ALL}")
print(
f"Enter 'y' to authorise command or 'n' to exit program, or enter feedback for {ai_name}...",
f"Enter 'y' to authorise command, 'y -N' to run N continuous commands, 'n' to exit program, or enter feedback for {ai_name}...",
flush=True)
while True:
console_input = input(Fore.MAGENTA + "Input:" + Style.RESET_ALL)
if console_input.lower() == "y":
user_input = "GENERATE NEXT COMMAND JSON"
break
elif console_input.lower().startswith("y -"):
try:
next_action_count = abs(int(console_input.split(" ")[1]))
user_input = "GENERATE NEXT COMMAND JSON"
except ValueError:
print("Invalid input format. Please enter 'y -n' where n is the number of continuous tasks.")
continue
break
elif console_input.lower() == "n":
user_input = "EXIT"
break
@@ -342,12 +359,20 @@ while True:
f"COMMAND = {Fore.CYAN}{command_name}{Style.RESET_ALL} ARGUMENTS = {Fore.CYAN}{arguments}{Style.RESET_ALL}")
# Execute command
if command_name.lower() == "error":
if 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}"
else:
result = f"Command {command_name} returned: {cmd.execute_command(command_name, arguments)}"
if next_action_count > 0:
next_action_count -= 1
memory_to_add = f"Assistant Reply: {assistant_reply} " \
f"\nResult: {result} " \
f"\nHuman Feedback: {user_input} "
memory.add(memory_to_add)
# Check if there's a result from the command append it to the message
# history

1
scripts/memory.py
View File

@@ -1 +0,0 @@
permanent_memory = []

44
scripts/memory/__init__.py Normal file
View File

@@ -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",
]

31
scripts/memory/base.py Normal file
View File

@@ -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

114
scripts/memory/local.py Normal file
View File

@@ -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

51
scripts/memory/pinecone.py Normal file
View File

@@ -0,0 +1,51 @@
import pinecone
from memory.base import MemoryProviderSingleton, get_ada_embedding
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)
dimension = 1536
metric = "cosine"
pod_type = "p1"
table_name = "auto-gpt"
# this assumes we don't start with memory.
# for now this works.
# we'll need a more complicated and robust system if we want to start with memory.
self.vec_num = 0
if table_name not in pinecone.list_indexes():
pinecone.create_index(table_name, dimension=dimension, metric=metric, pod_type=pod_type)
self.index = pinecone.Index(table_name)
def add(self, data):
vector = get_ada_embedding(data)
# no metadata here. We may wish to change that long term.
resp = self.index.upsert([(str(self.vec_num), vector, {"raw_text": data})])
_text = f"Inserting data into memory at index: {self.vec_num}:\n data: {data}"
self.vec_num += 1
return _text
def get(self, data):
return self.get_relevant(data, 1)
def clear(self):
self.index.delete(deleteAll=True)
return "Obliviated"
def get_relevant(self, data, num_relevant=5):
"""
Returns all the data in the memory that is relevant to the given data.
:param data: The data to compare to.
:param num_relevant: The number of relevant data to return. Defaults to 5
"""
query_embedding = get_ada_embedding(data)
results = self.index.query(query_embedding, top_k=num_relevant, include_metadata=True)
sorted_results = sorted(results.matches, key=lambda x: x.score)
return [str(item['metadata']["raw_text"]) for item in sorted_results]
def get_stats(self):
return self.index.describe_index_stats()

143
scripts/memory/redismem.py Normal file
View File

@@ -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()

2
scripts/speak.py
View File

@@ -42,7 +42,7 @@ def say_text(text, voice_index=0):
if not cfg.elevenlabs_api_key:
gtts_speech(text)
else:
success = eleven_labs_speech(text)
success = eleven_labs_speech(text, voice_index)
if not success:
gtts_speech(text)