From 9c0f2ba2bb3196211d162f45677e771a38becaa9 Mon Sep 17 00:00:00 2001 From: James Ketrenos Date: Wed, 19 Mar 2025 18:22:30 -0700 Subject: [PATCH] site fetching is working --- jupyter/stock.py | 235 +++++++++++++++++++++++++++++++---------------- jupyter/tools.py | 47 +++++++++- 2 files changed, 201 insertions(+), 81 deletions(-) diff --git a/jupyter/stock.py b/jupyter/stock.py index 08f8826..7a18842 100644 --- a/jupyter/stock.py +++ b/jupyter/stock.py @@ -31,6 +31,7 @@ try_import('yfinance', 'yfinance') try_import('dotenv', 'python-dotenv') try_import('geopy', 'geopy') try_import('hyphen', 'PyHyphen') +try_import('bs4', 'beautifulsoup4') from dotenv import load_dotenv from geopy.geocoders import Nominatim @@ -42,6 +43,7 @@ import pytz import requests import yfinance as yf from hyphen import hyphenator +from bs4 import BeautifulSoup # Local defined imports from tools import ( @@ -71,23 +73,20 @@ BOT_ADMIN="james" # Globals system_message = f""" You are a helpful information agent connected to the IRC network {IRC_SERVER}. Your name is {NICK}. +You have real time access to any website or URL the user asks about. +Messages from users are in the form "NICK: MESSAGE". The name before the colon (:) tells you which user asked about something. You are running { { 'model': MODEL_NAME, 'gpu': 'Intel Arc B580', 'cpu': 'Intel Core i9-14900KS', 'ram': '64G' } }. You were launched on {get_current_datetime()}. -You have real time access to current stock trading values, the current date and time, and current weather information for locations in the United States. +You have real time access to stock prices, the current date and time, and current weather information for locations in the United States. If you use any real time access, do not mention your knowledge cutoff. -Give short, courteous answers, no more than 2-3 sentences, keeping the answer less than about 100 characters. -If you have to cut the answer short, ask the user if they want more information and provide it if they say Yes. +Give short, courteous answers, no more than 2-3 sentences. Always be accurate. If you don't know the answer, say so. Do not make up details. - -You have tools to: -* get_current_datetime: Get current time and date. -* get_weather_by_location: Get-real time weather forecast. -* get_ticker_price: Get real-time value of a stock symbol. - -Those are the only tools available. +When you receive a response from summarize_site, you must: +1. Review the entire content returned by the second LLM +2. Provide the URL used to obtain the information. +3. Incorporate the information into your response as appropriate """ system_log = [{"role": "system", "content": system_message}] -history = [] tool_log = [] command_log = [] model = None @@ -103,12 +102,12 @@ def parse_args(): parser.add_argument("--irc-nickname", type=str, default=NICK, help=f"Bot nickname. default={NICK}") parser.add_argument("--irc-channel", type=str, default=CHANNEL, help=f"Channel to join. default={CHANNEL}") parser.add_argument("--irc-use-tls", type=bool, default=USE_TLS, help=f"Use TLS with --irc-server. default={USE_TLS}") + parser.add_argument("--irc-bot-admin", type=str, default=BOT_ADMIN, help=f"Nick that can send admin commands via IRC. default={BOT_ADMIN}") parser.add_argument("--ollama-server", type=str, default=OLLAMA_API_URL, help=f"Ollama API endpoint. default={OLLAMA_API_URL}") parser.add_argument("--ollama-model", type=str, default=MODEL_NAME, help=f"LLM model to use. default={MODEL_NAME}") parser.add_argument("--gradio-host", type=str, default=GRADIO_HOST, help=f"Host to launch gradio on. default={GRADIO_HOST} only if --gradio-enable is specified.") parser.add_argument("--gradio-port", type=str, default=GRADIO_PORT, help=f"Port to launch gradio on. default={GRADIO_PORT} only if --gradio-enable is specified.") parser.add_argument("--gradio-enable", action="store_true", default=GRADIO_ENABLE, help=f"If set to True, enable Gradio. default={GRADIO_ENABLE}") - parser.add_argument("--bot-admin", type=str, default=BOT_ADMIN, help=f"Nick that can send admin commands via IRC. default={BOT_ADMIN}") parser.add_argument('--level', type=str, choices=['DEBUG', 'INFO', 'WARNING', 'ERROR', 'CRITICAL'], default=LOG_LEVEL, help=f'Set the logging level. default={LOG_LEVEL}') return parser.parse_args() @@ -205,27 +204,31 @@ def split_paragraph_with_hyphenation(text, line_length=80, language='en_US'): return result_lines # %% -def handle_tool_calls(message): +async def handle_tool_calls(message): response = [] tools_used = [] for tool_call in message['tool_calls']: arguments = tool_call['function']['arguments'] tool = tool_call['function']['name'] - if tool == 'get_ticker_price': - ticker = arguments.get('ticker') - if not ticker: + match tool: + case 'get_ticker_price': + ticker = arguments.get('ticker') + if not ticker: + ret = None + else: + ret = get_ticker_price(ticker) + tools_used.append(tool) + case 'summarize_site': + ret = await summarize_site(arguments.get('url'), arguments.get('question', 'what is the summary of this content?')) + tools_used.append(tool) + case 'get_current_datetime': + ret = get_current_datetime(arguments.get('timezone')) + tools_used.append(tool) + case 'get_weather_by_location': + ret = get_weather_by_location(arguments.get('city'), arguments.get('state')) + tools_used.append(tool) + case _: ret = None - else: - ret = get_ticker_price(ticker) - tools_used.append(tool) - elif tool == 'get_current_datetime': - ret = get_current_datetime(arguments.get('timezone')) - tools_used.append(tool) - elif tool == 'get_weather_by_location': - ret = get_weather_by_location(arguments.get('city'), arguments.get('state')) - tools_used.append(tool) - else: - ret = None response.append({ "role": "tool", "content": str(ret), @@ -237,6 +240,76 @@ def handle_tool_calls(message): return response, tools_used # %% +def total_json_length(dict_array): + total = 0 + for item in dict_array: + # Convert dictionary to minimized JSON string + json_string = json.dumps(item, separators=(',', ':')) + total += len(json_string) + return total + +async def summarize_site(url, question): + """ + Fetches content from a URL, extracts the text, and uses Ollama to summarize it. + + Args: + url (str): The URL of the website to summarize + + Returns: + str: A summary of the website content + """ + global model, client + try: + # Fetch the webpage + headers = { + 'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36' + } + logging.info(f"Fetching {url}") + response = requests.get(url, headers=headers, timeout=10) + response.raise_for_status() + logging.info(f"{url} returned. Processing...") + # Parse the HTML + soup = BeautifulSoup(response.text, 'html.parser') + + # Remove script and style elements + for script in soup(["script", "style"]): + script.extract() + + # Get text content + text = soup.get_text(separator=' ', strip=True) + + # Clean up text (remove extra whitespace) + lines = (line.strip() for line in text.splitlines()) + chunks = (phrase.strip() for line in lines for phrase in line.split(" ")) + text = ' '.join(chunk for chunk in chunks if chunk) + + # Limit text length if needed (Ollama may have token limits) + max_chars = 100000 + if len(text) > max_chars: + text = text[:max_chars] + "..." + + # Create Ollama client + logging.info(f"Requesting summary of: {text}") + + # Generate summary using Ollama + prompt = f"CONTENTS:\n\n{text}\n\n{question}" + response = client.generate(model=model, + system="You are given the contents of {url}. Answer the question about the contents", + prompt=prompt) + + logging.info(response['response']) + + return { + 'source': 'summarizer-llm', + 'content': response['response'], + 'metadata': get_current_datetime() + } + + except requests.exceptions.RequestException as e: + return f"Error fetching the URL: {str(e)}" + except Exception as e: + return f"Error processing the website content: {str(e)}" + async def chat(history, is_irc=False): global client, model, irc_bot, system_log, tool_log if not client: @@ -250,6 +323,8 @@ async def chat(history, is_irc=False): tools_used = [] if 'tool_calls' in response['message']: message = response['message'] + tool_result, tools_used = await handle_tool_calls(message) + # Convert Message object to a proper dictionary format message_dict = { 'role': message.get('role', 'assistant'), @@ -262,9 +337,11 @@ async def chat(history, is_irc=False): for tc in message['tool_calls'] ] - tool_result, tools_used = handle_tool_calls(message) messages.append(message_dict) # Add properly formatted dict instead of Message object - messages.append(tool_result) + if isinstance(tool_result, list): + messages.extend(tool_result) + else: + messages.append(tool_result) try: response = client.chat(model=model, messages=messages) except Exception: @@ -334,7 +411,7 @@ class DynamicIRCBot(pydle.Client): max_lines = 10 irc_lines = [] for line in message.splitlines(): - lines = split_paragraph_with_hyphenation(line, line_length=450) + lines = split_paragraph_with_hyphenation(line, line_length=300) irc_lines.extend(lines) # Send the first 'max_lines' non-empty lines @@ -384,9 +461,8 @@ class DynamicIRCBot(pydle.Client): user = None content = message - # If this message is not directed to the bot - if not user or user != self.nickname: + if target != self.nickname and (not user or user != self.nickname): logging.info(f"Message not directed to {self.nickname}") # Add this message to the history either to the current 'user' context or create # add a new message @@ -403,7 +479,7 @@ class DynamicIRCBot(pydle.Client): return matches = re.match(r"^!([^\s]+)\s*(.*)?$", content) - if not matches or (self.bot_admin and source != self.bot_admin and source != self.nickname): + if not matches: logging.info(f"Non-command directed message to {self.nickname}: Invoking chat...") # Add this message to the history either to the current 'user' context or create # add a new message @@ -425,7 +501,7 @@ class DynamicIRCBot(pydle.Client): command = matches.group(1) arguments = matches.group(2).strip() logging.info(f"Command directed to {self.nickname}: command={command}, arguments={arguments}") - + is_admin = source == self.nickname or source == self.bot_admin match command: case "help": response = f"info, context, reset, system [prompt], server [address], join channel" @@ -434,14 +510,19 @@ class DynamicIRCBot(pydle.Client): response = str(self.system_info) case "context": - if len(self.history) > 1: - response = '"' + '","'.join(self.history[-1]['content'].split('\n')) + '"' - else: - response = "" + system_log_size = total_json_length(system_log) + history_size = total_json_length(self.history) + tools_size = total_json_length(tools) + total_size = system_log_size + history_size + tools_size + response = f"\nsystem prompt: {system_log_size}" + response += f"\nhistory: {history_size} in {len(self.history)} entries." + response += f"\ntools: {tools_size} in {len(tools)} tools." + response += f"\ntotal context: {total_size}" + response += f"\ntotal tool calls: {len(tool_log)}" case "reset": system_log = [{"role": "system", "content": system_message}] - history = [] + self.history = [] tool_log = [] command_log = [] response = 'All contexts reset' @@ -455,31 +536,37 @@ class DynamicIRCBot(pydle.Client): response = " ".join(lines) case "server": - server = arguments.split(" ", 1) - if server[0] == "": - server = IRC_SERVER + if not is_admin: + response = "You need to be admin to use this command." else: - server = server[0] - try: - await self.connect(server, 6667, tls=False) - response="Connected to {server}" - except Exception: - response = f"Unable to connect to {server}" - logging.exception({ "error": f"Unable to process message {content}"}) + server = arguments.split(" ", 1) + if server[0] == "": + server = IRC_SERVER + else: + server = server[0] + try: + await self.connect(server, 6667, tls=False) + response="Connected to {server}" + except Exception: + response = f"Unable to connect to {server}" + logging.exception({ "error": f"Unable to process message {content}"}) case "join": - channel = arguments.strip() - if channel == "" or re.match(r"\s", channel): - response = "Usage: !join CHANNEL" + if not is_admin: + response = "You need to be admin to use this command." else: - if not re.match(r"^#", channel): - channel = f"#{channel}" - if self.channel and self.channel != channel: - await self.part(channel) - if channel: - await self.bot.join(channel) - self.channel = channel - response = f"Joined {channel}." + channel = arguments.strip() + if channel == "" or re.match(r"\s", channel): + response = "Usage: !join CHANNEL" + else: + if not re.match(r"^#", channel): + channel = f"#{channel}" + if self.channel and self.channel != channel: + await self.part(channel) + if channel: + await self.bot.join(channel) + self.channel = channel + response = f"Joined {channel}." case _: response = f"Unrecognized command: {command}" @@ -539,7 +626,7 @@ async def create_ui(): ) with gr.Row(scale=0): clear = gr.Button("Clear") - timer = gr.Timer(1) + refresh = gr.Button("Sync with IRC") async def do_entry(message): if not irc_bot: @@ -557,23 +644,15 @@ async def create_ui(): return irc_bot.history, system_log, tool_log, command_log def update_log(history): + if not irc_bot: + return gr.skip() # This function updates the log after the chatbot responds - return system_log + history, tool_log, command_log + return system_log + irc_bot.history, tool_log, command_log - def check_history(): - global last_history_len, last_command_len - if not irc_bot or last_history_len == len(irc_bot.history): - history = gr.skip() - else: - history = irc_bot.history - last_history_len = len(irc_bot.history) - if last_command_len == len(command_log): - commands = gr.skip() - else: - commands = command_log - last_command_len = len(command_log) - - return history, commands + def get_history(): + if not irc_bot: + return gr.skip() + return irc_bot.history, system_log + irc_bot.history, tool_log, command_log entry.submit( do_entry, @@ -585,7 +664,7 @@ async def create_ui(): outputs=[chat_history, tool_history, command_history] ) - timer.tick(check_history, inputs=None, outputs=[chatbot, command_history]) + refresh.click(get_history, inputs=None, outputs=[chatbot, chat_history, tool_history, command_history]) clear.click(do_clear, inputs=None, outputs=[chatbot, chat_history, tool_history, command_history], queue=False) @@ -606,7 +685,7 @@ async def main(): client = ollama.Client(host=args.ollama_server) model = args.ollama_model - irc_bot = DynamicIRCBot(args.irc_nickname, args.irc_channel, args.bot_admin, args) + irc_bot = DynamicIRCBot(args.irc_nickname, args.irc_channel, args.irc_bot_admin, args) await irc_bot.connect(args.irc_server, args.irc_port, tls=args.irc_use_tls) if args.gradio_enable: diff --git a/jupyter/tools.py b/jupyter/tools.py index aa72b55..95ed853 100644 --- a/jupyter/tools.py +++ b/jupyter/tools.py @@ -69,6 +69,9 @@ def get_weather_by_location(city, state, country="USA"): # Step 3: Get the forecast data from the grid endpoint forecast = get_forecast(grid_endpoint) + if not forecast['location']: + forecast['location'] = location + return forecast def get_coordinates(location): @@ -210,7 +213,7 @@ def get_ticker_price(ticker_symbols): # Create a Ticker object try: ticker = yf.Ticker(ticker_symbol) - + print(ticker) # Get the latest market data ticker_data = ticker.history(period="1d") @@ -268,7 +271,7 @@ def get_current_datetime(timezone="America/Los_Angeles"): # %% -tools = [{ +tools = [ { "type": "function", "function": { "name": "get_ticker_price", @@ -285,6 +288,44 @@ tools = [{ "additionalProperties": False } } +}, { + "type": "function", + "function": { + "name": "summarize_site", + "description": "Requests a second LLM agent to download the requested site and answer a question about the site. For example if the user says 'What are the top headlines on cnn.com?' you would use summarize_site to get the answer.", + "parameters": { + "type": "object", + "properties": { + "url": { + "type": "string", + "description": "The website URL to download and process", + }, + "question": { + "type": "string", + "description": "The question to ask the second LLM about the content", + }, + }, + "required": ["url", "question"], + "additionalProperties": False + }, + "returns": { + "type": "object", + "properties": { + "source": { + "type": "string", + "description": "Identifier for the source LLM" + }, + "content": { + "type": "string", + "description": "The complete response from the second LLM" + }, + "metadata": { + "type": "object", + "description": "Additional information about the response" + } + } + } + } }, { "type": "function", "function": { @@ -324,4 +365,4 @@ tools = [{ } }] -__all__ = [ 'tools', 'get_current_datetime', 'get_weather_by_location', 'get_ticker_price' ] \ No newline at end of file +__all__ = [ 'tools', 'get_current_datetime', 'get_weather_by_location', 'get_ticker_price' ]