From 7e75ea80a6ee186856c0a4785dd200471012c177 Mon Sep 17 00:00:00 2001 From: Raven Date: Fri, 9 Aug 2024 04:04:30 -0400 Subject: [PATCH] Add detailed comments --- ai_log.js | 213 ++++++++++++++++++++++++++++++++---------------------- 1 file changed, 128 insertions(+), 85 deletions(-) diff --git a/ai_log.js b/ai_log.js index 3b11e0e..c7e0e42 100644 --- a/ai_log.js +++ b/ai_log.js @@ -1,12 +1,14 @@ -import express from 'express'; -import axios from 'axios'; -import bodyParser from 'body-parser'; -import cmd from 'cmd-promise'; -import cors from 'cors'; -import cheerio from 'cheerio'; -import 'dotenv/config'; -import llamaTokenizer from 'llama-tokenizer-js'; +// Import necessary modules for the application +import express from 'express'; // Express framework for building web server applications and handling HTTP requests and responses +import axios from 'axios'; // Axios is used to make HTTP requests to external APIs or services +import bodyParser from 'body-parser'; // Middleware for parsing incoming request bodies, specifically for handling JSON data +import cmd from 'cmd-promise'; // A module that allows execution of shell commands in a promise-based manner, making it easier to manage async operations +import cors from 'cors'; // Middleware to enable Cross-Origin Resource Sharing, allowing resources to be requested from another domain +import cheerio from 'cheerio'; // Cheerio is a server-side jQuery-like library for parsing and manipulating HTML content +import 'dotenv/config'; // Loads environment variables from a .env file into process.env, allowing secure storage of sensitive information +import llamaTokenizer from 'llama-tokenizer-js'; // A library for tokenizing text, which is crucial for managing the length of text inputs to the AI model +// Define a prompt that will guide the AI's behavior when analyzing NGINX logs for potential security issues const prompt = ` You are a security AI responsible for analyzing web traffic from NGINX logs and blocking malicious IPs. Your task is to review the logs for potential attacks and issues. If you identify a verified problem, include [ALERT] followed by a detailed description of the issue in your response. Ensure your message is formatted in Markdown compatible with Discord. @@ -28,209 +30,250 @@ You are a security AI responsible for analyzing web traffic from NGINX logs and - To ban an IP or flag it as a security risk, wrap it in a Markdown spoiler: ||IPHERE|| `; -const app = express(); -const port = 3001; +// Initialize the Express application and define the port on which the server will run +const app = express(); // Create an instance of an Express application +const port = 3001; // Define the port number for the server, 3001 is commonly used for development -app.use(cors()); // Enable CORS for all routes +// Middleware to enable CORS for all routes +app.use(cors()); // This allows the server to accept requests from any origin, useful for APIs that may be accessed by web applications from different domains -// Set a larger limit for the request body -app.use(bodyParser.json({ limit: '50mb' })); // Adjust the limit as needed +// Set a larger limit for the request body to handle large data payloads +app.use(bodyParser.json({ limit: '50mb' })); // The JSON body parser is configured with a 50MB limit, suitable for handling large JSON payloads -const TIMEOUT_DURATION = 100000; // Timeout duration in milliseconds (100 seconds) -const MAX_TOKENS = 8000; // Maximum tokens allowed by the model -const TOLERANCE = 100; // Tolerance to avoid exceeding token limit -let conversationHistory = {}; +// Define constants for the application, used to control various aspects of the server's behavior +const TIMEOUT_DURATION = 100000; // The maximum time (in milliseconds) the server will wait before timing out a request, set to 100 seconds +const MAX_TOKENS = 8000; // The maximum number of tokens (words and punctuation) allowed in a conversation, this limit helps manage API usage +const TOLERANCE = 100; // A buffer value used to prevent exceeding the MAX_TOKENS limit, ensuring the conversation stays within safe bounds +let conversationHistory = {}; // An object to store conversation history for each IP address, allowing the server to maintain context for each user -// Helper function to get current timestamp +// Helper function to get the current timestamp in a formatted string const getTimestamp = () => { - const now = new Date(); - const date = now.toLocaleDateString('en-US'); - const time = now.toLocaleTimeString('en-US'); - return `${date} [${time}]`; + const now = new Date(); // Get the current date and time + const date = now.toLocaleDateString('en-US'); // Format the date in the US locale + const time = now.toLocaleTimeString('en-US'); // Format the time in the US locale + return `${date} [${time}]`; // Return the formatted date and time as a string }; -// Middleware to track conversation history by CF-Connecting-IP +// Middleware to track conversation history based on the client's IP address app.use((req, res, next) => { + // Extract the client's IP address from various possible headers (CF-Connecting-IP, X-Forwarded-For, X-Real-IP) or fallback to req.ip const ip = req.headers['cf-connecting-ip'] || req.headers['x-forwarded-for'] || req.headers['x-real-ip'] || req.ip; - req.clientIp = ip; // Store the IP in a request property - console.log(`${getTimestamp()} [INFO] Incoming request from IP: ${req.clientIp}`); // Log the IP address + req.clientIp = ip; // Store the client's IP address in the request object for easy access later + // Log the incoming request along with the client's IP address and current timestamp + console.log(`${getTimestamp()} [INFO] Incoming request from IP: ${req.clientIp}`); + + // If this IP address has not been seen before, initialize a new conversation history for it if (!conversationHistory[req.clientIp]) { console.log(`${getTimestamp()} [INFO] Initializing conversation history for new IP: ${req.clientIp}`); + // Start the conversation with the predefined prompt that instructs the AI on how to analyze the logs conversationHistory[req.clientIp] = [ { role: 'system', content: prompt } ]; } - next(); + next(); // Move on to the next middleware or route handler }); +// Function to count the number of tokens in a conversation history using the llama tokenizer async function countLlamaTokens(messages) { - let totalTokens = 0; + let totalTokens = 0; // Initialize a counter for the total number of tokens for (const message of messages) { + // Only count tokens for user and assistant messages, not system messages if (message.role === 'user' || message.role === 'assistant') { - const encodedTokens = llamaTokenizer.encode(message.content); - totalTokens += encodedTokens.length; + const encodedTokens = llamaTokenizer.encode(message.content); // Tokenize the message content + totalTokens += encodedTokens.length; // Add the number of tokens in the current message to the total } } - return totalTokens; + return totalTokens; // Return the total number of tokens } +// Function to trim the conversation history to fit within the token limit async function trimConversationHistory(messages, maxLength, tolerance) { - let tokenLength = await countLlamaTokens(messages); + let tokenLength = await countLlamaTokens(messages); // Get the total number of tokens in the conversation + // Continue trimming the conversation history until it's within the allowed token limit while (tokenLength > maxLength - tolerance && messages.length > 1) { - messages.splice(1, 1); // Remove the oldest user/assistant message - tokenLength = await countLlamaTokens(messages); + messages.splice(1, 1); // Remove the oldest user/assistant message (the second item in the array) + tokenLength = await countLlamaTokens(messages); // Recalculate the total number of tokens after trimming console.log(`${getTimestamp()} [CLEANUP] Trimmed conversation history to ${tokenLength} tokens.`); } } -// Function to scrape web page +// Function to scrape a web page and extract its content async function scrapeWebPage(url) { console.log(`${getTimestamp()} [INFO] Starting to scrape URL: ${url}`); try { + // Perform an HTTP GET request to fetch the content of the specified URL const res = await axios.get(url); - const html = res.data; - const $ = cheerio.load(html); + const html = res.data; // Extract the HTML content from the response + const $ = cheerio.load(html); // Load the HTML into Cheerio for parsing and manipulation - // Extract page title, meta description and content - const pageTitle = $('head title').text().trim(); - const pageDescription = $('head meta[name="description"]').attr('content'); - const pageContent = $('body').text().trim(); + // Extract specific elements from the HTML: the page title, meta description, and body content + const pageTitle = $('head title').text().trim(); // Get the text of the tag + const pageDescription = $('head meta[name="description"]').attr('content'); // Get the content of the meta description + const pageContent = $('body').text().trim(); // Get all text content within the <body> tag - // Construct response message with page details - let response = `Title: ${pageTitle}\n`; + // Construct a response message with the extracted details + let response = `Title: ${pageTitle}\n`; // Start with the page title if (pageDescription) { - response += `Description: ${pageDescription}\n`; + response += `Description: ${pageDescription}\n`; // Add the meta description if it exists } if (pageContent) { - const MAX_CONTENT_LENGTH = process.env.MAX_CONTENT_LENGTH || 2000; + const MAX_CONTENT_LENGTH = process.env.MAX_CONTENT_LENGTH || 2000; // Set a maximum length for the content + // Clean the page content to remove unnecessary whitespace and special characters let plainTextContent = $('<div>').html(pageContent).text().trim().replace(/[\r\n\t]+/g, ' '); + // Define a regular expression pattern to identify code-like content const codePattern = /\/\/|\/\*|\*\/|\{|\}|\[|\]|\bfunction\b|\bclass\b|\b0x[0-9A-Fa-f]+\b|\b0b[01]+\b/; - const isCode = codePattern.test(plainTextContent); + const isCode = codePattern.test(plainTextContent); // Check if the content resembles code if (isCode) { - plainTextContent = plainTextContent.replace(codePattern, ''); + plainTextContent = plainTextContent.replace(codePattern, ''); // Remove code-like patterns if detected } + // Further clean the content by removing text within parentheses plainTextContent = plainTextContent.replace(/ *\([^)]*\) */g, ''); + // If the content is too long, truncate it and add an ellipsis if (plainTextContent.length > MAX_CONTENT_LENGTH) { plainTextContent = plainTextContent.substring(0, MAX_CONTENT_LENGTH) + '...'; } - response += `Content: ${plainTextContent.trim()}`; + response += `Content: ${plainTextContent.trim()}`; // Add the cleaned and possibly truncated content to the response } - response += `\nURL: ${url}`; + response += `\nURL: ${url}`; // Include the original URL in the response console.log(`${getTimestamp()} [INFO] Successfully scraped URL: ${url}`); - return response; + return response; // Return the constructed response } catch (err) { + // If the scraping process fails, log an error with details and return null console.error(`${getTimestamp()} [ERROR] Failed to scrape URL: ${url}`, err); return null; } } +// Function to process incoming requests, handle AI interactions, and return a response async function processRequest(req, res) { - const startTime = Date.now(); // Start time tracking - const ip = req.clientIp; - console.log(`${getTimestamp()} [INFO] Handling chat request from IP: ${ip}`); // Log the IP address + const startTime = Date.now(); // Record the start time of the request processing for performance tracking + const ip = req.clientIp; // Retrieve the client's IP address from the request object + console.log(`${getTimestamp()} [INFO] Handling chat request from IP: ${ip}`); // Log the request details + // Set a timeout for the request processing, ensuring it doesn't hang indefinitely const timeout = setTimeout(() => { console.error(`${getTimestamp()} [ERROR] Request timed out for IP: ${ip}`); - res.status(408).json({ message: "Request timed out" }); + res.status(408).json({ message: "Request timed out" }); // Send a timeout response if the processing takes too long }, TIMEOUT_DURATION); try { - let userMessage = req.body.message; + let userMessage = req.body.message; // Extract the user's message from the request body console.log(`${getTimestamp()} [INFO] Received user message: ${userMessage}`); - userMessage = req.body.message + `\nDate/Time:${getTimestamp()}`; + userMessage = req.body.message + `\nDate/Time:${getTimestamp()}`; // Append the current date and time to the user's message + // Initialize conversation history if it doesn't exist for the IP if (!conversationHistory[ip]) { console.log(`${getTimestamp()} [INFO] Initializing conversation history for new IP: ${ip}`); - conversationHistory[ip] = [{ role: 'system', content: prompt }]; + conversationHistory[ip] = [{ role: 'system', content: prompt }]; // Start the conversation with the predefined prompt } + // Add the user's message to the conversation history for the IP conversationHistory[ip].push({ role: 'user', content: userMessage }); - // Trim conversation history if it exceeds the token limit + // Trim the conversation history if it exceeds the token limit await trimConversationHistory(conversationHistory[ip], MAX_TOKENS, TOLERANCE); + // Split the user's message into individual log lines const logLines = userMessage.split('\n'); + // Define a regex pattern to identify lines containing client IP addresses const clientIpRegex = /\[Client (\d{1,3}\.){3}\d{1,3}\]/; + // Filter the log lines to only include those with valid client IP addresses const filteredLogLines = logLines.filter(line => clientIpRegex.test(line)); + // If no valid IP addresses are found in the log lines, send a response indicating this if (filteredLogLines.length === 0) { console.log(`${getTimestamp()} [INFO] No valid client IP addresses found in the log.`); res.json({ message: "No valid client IP addresses found in the log." }); return; } + // Join the filtered log lines back into a single string for processing const filteredMessage = filteredLogLines.join('\n'); + // Send the request to the llama API for processing and response generation console.log(`${getTimestamp()} [INFO] Sending request to llama API for response`); const response = await axios.post('http://127.0.0.1:8003/v1/chat/completions', { - model: 'gpt-3.5-turbo', - messages: [...conversationHistory[ip], { role: 'user', content: filteredMessage }] + model: 'gpt-3.5-turbo', // Specify the AI model to use + messages: [...conversationHistory[ip], { role: 'user', content: filteredMessage }] // Include the conversation history and the filtered message }); + // Extract the AI's response from the API's response data const assistantMessage = response.data.choices[0].message; + // Add the AI's response to the conversation history conversationHistory[ip].push(assistantMessage); + // Log the AI's response and additional details like the finish reason and token usage console.log(`${getTimestamp()} [INFO] Received response from llama API: ${assistantMessage.content}`); console.log(`${getTimestamp()} [DEBUG] Finish Reason: ${response.data.choices[0].finish_reason}`); console.log(`${getTimestamp()} [STATS] Usage: prompt_tokens=${response.data.usage.prompt_tokens}, completion_tokens=${response.data.usage.completion_tokens}, total_tokens=${response.data.usage.total_tokens}`); - clearTimeout(timeout); - res.json(assistantMessage); + clearTimeout(timeout); // Clear the timeout to prevent it from triggering + res.json(assistantMessage); // Send the AI's response back to the client } catch (error) { + // If an error occurs during request processing, log the error and send a 500 response console.error(`${getTimestamp()} [ERROR] An error occurred while handling chat request`, error); - clearTimeout(timeout); - res.status(500).json({ message: "An error occurred", error: error.message }); + clearTimeout(timeout); // Clear the timeout to prevent it from triggering + res.status(500).json({ message: "An error occurred", error: error.message }); // Send an error response } finally { - const endTime = Date.now(); // End time tracking - const processingTime = ((endTime - startTime) / 1000).toFixed(2); // Calculate processing time in seconds - console.log(`${getTimestamp()} [STATS] Processing Time: ${processingTime} seconds`); + // Record the end time and calculate the total processing time in seconds + const endTime = Date.now(); + const processingTime = ((endTime - startTime) / 1000).toFixed(2); // Convert milliseconds to seconds + console.log(`${getTimestamp()} [STATS] Processing Time: ${processingTime} seconds`); // Log the processing time console.log(`${getTimestamp()} [INFO] Finished processing chat request for IP: ${ip}`); } } +// Route to handle incoming chat requests, trim the message content, and process the request app.post('/api/v1/chat', async (req, res) => { // Trim the incoming message to fit within token limits - const messageContent = req.body.message; - const encodedTokens = llamaTokenizer.encode(messageContent); - const MAX_MESSAGE_TOKENS = MAX_TOKENS - (await countLlamaTokens([{ role: 'system', content: prompt }])) - TOLERANCE; + const messageContent = req.body.message; // Get the user's message from the request body + const encodedTokens = llamaTokenizer.encode(messageContent); // Tokenize the message to determine its length in tokens + const MAX_MESSAGE_TOKENS = MAX_TOKENS - (await countLlamaTokens([{ role: 'system', content: prompt }])) - TOLERANCE; // Calculate the maximum allowed tokens for the user's message + // If the message exceeds the allowed token limit, trim it to fit let trimmedMessageContent = messageContent; if (encodedTokens.length > MAX_MESSAGE_TOKENS) { - trimmedMessageContent = llamaTokenizer.decode(encodedTokens.slice(0, MAX_MESSAGE_TOKENS)); + trimmedMessageContent = llamaTokenizer.decode(encodedTokens.slice(0, MAX_MESSAGE_TOKENS)); // Truncate the message and decode it back to a string } + // Process the trimmed message and send the response await processRequest({ ...req, body: { message: trimmedMessageContent } }, res); }); +// Route to fetch the conversation history for a specific IP address app.get('/api/v1/conversation-history', (req, res) => { - const ip = req.clientIp; - console.log(`${getTimestamp()} [INFO] Fetching conversation history for IP: ${ip}`); // Log the IP address - res.json(conversationHistory[ip]); + const ip = req.clientIp; // Get the client's IP address from the request object + console.log(`${getTimestamp()} [INFO] Fetching conversation history for IP: ${ip}`); // Log the request details + res.json(conversationHistory[ip]); // Send the conversation history for the IP as a JSON response }); +// Route to restart the core AI service via Docker, typically used to refresh the model or resolve issues app.post('/api/v1/restart-core', (req, res) => { - console.log(`${getTimestamp()} [INFO] Restarting core service`); - cmd(`docker restart llama-gpu-server`).then(out => { - console.log(`${getTimestamp()} [INFO] Core service restarted`); - res.json(out.stdout); - }).catch(err => { - console.error(`${getTimestamp()} [ERROR] Failed to restart core service`, err); - res.status(500).json({ message: "An error occurred while restarting the core service", error: err.message }); + console.log(`${getTimestamp()} [INFO] Restarting core service`); // Log the restart action + cmd(`docker restart llama-gpu-server`).then(out => { // Execute a shell command to restart the Docker container running the AI model + console.log(`${getTimestamp()} [INFO] Core service restarted`); // Log the successful restart + res.json(out.stdout); // Send the output of the restart command back to the client + }).catch(err => { // Handle any errors that occur during the restart + console.error(`${getTimestamp()} [ERROR] Failed to restart core service`, err); // Log the error + res.status(500).json({ message: "An error occurred while restarting the core service", error: err.message }); // Send an error response }); }); +// Route to reset the conversation history for a specific IP address, effectively starting a new session app.post('/api/v1/reset-conversation', (req, res) => { - const ip = req.clientIp; - console.log(`${getTimestamp()} [INFO] Resetting conversation history for IP: ${ip}`); // Log the IP address + const ip = req.clientIp; // Get the client's IP address from the request object + console.log(`${getTimestamp()} [INFO] Resetting conversation history for IP: ${ip}`); // Log the reset action + // Reset the conversation history to its initial state for the given IP conversationHistory[ip] = [ { role: 'system', content: prompt } ]; - console.log(`${getTimestamp()} [INFO] Conversation history reset for IP: ${ip}`); - res.json({ message: "Conversation history reset for IP: " + ip }); + console.log(`${getTimestamp()} [INFO] Conversation history reset for IP: ${ip}`); // Log the successful reset + res.json({ message: "Conversation history reset for IP: " + ip }); // Send a confirmation message back to the client }); +// Start the Express server on the defined port, making the API available for requests app.listen(port, () => { - console.log(`${getTimestamp()} [INFO] Server running at http://localhost:${port}`); + console.log(`${getTimestamp()} [INFO] Server running at http://localhost:${port}`); // Log the server startup and its URL });