Add automatic web search for current events queries
Deploy to NAS / deploy (push) Successful in 1m8s
Deploy to NAS / deploy (push) Successful in 1m8s
- Detect keywords like 'latest', 'news', 'current', '2026' etc - Append :online suffix to model for web search + AI response - Retry logic: if web search fails, fall back to regular response - Increased timeout to 90s for web search queries - Fixed typo in fallback error handling
This commit is contained in:
+72
-25
@@ -1,11 +1,18 @@
|
||||
// ai.js - NanoGPT API wrapper for kekbot with fallback support
|
||||
// ai.js - NanoGPT API wrapper for kekbot with fallback and web search support
|
||||
// Copyright (C) 2025 Luis Bauza
|
||||
|
||||
import axios from 'axios';
|
||||
|
||||
const DEFAULT_MODEL = 'GLM-4.5-Air-Derestricted-Steam-ReExtract';
|
||||
const DEFAULT_MODEL = 'GLM-4.5-Air-Derestricted';
|
||||
const FALLBACK_MODEL = 'GLM-4.7-Flash';
|
||||
const FALLBACK_API_URL = 'https://api.z.ai/api/paas/v4/chat/completions';
|
||||
const FALLBACK_API_URL = 'https://api.z.ai/api/coding/paas/v4/chat/completions';
|
||||
|
||||
// Keywords that suggest the user wants current/web information
|
||||
const WEB_SEARCH_TRIGGERS = [
|
||||
'latest', 'news', 'current', 'today', 'yesterday', 'this week',
|
||||
'what is the', 'who is', 'what happened', 'search', 'find',
|
||||
'how do i', 'how to', 'what\'s the best', 'top ', '2024', '2025', '2026'
|
||||
];
|
||||
|
||||
const SASSY_SYSTEM_PROMPT = `You are kekbot, a sassy Discord chatter with a big personality.
|
||||
|
||||
@@ -19,7 +26,7 @@ CORE PERSONALITY:
|
||||
|
||||
OPINIONS:
|
||||
- You're a true gamer with strong opinions about games
|
||||
- You hate League of Legends and Riot Games with a passion - but only bring it up when asked
|
||||
- You hate League of Legends and Riot Games with a passion - - but only bring it up when asked
|
||||
- When asked about other games, respond normally without mentioning LoL
|
||||
- You have views on politics and aren't afraid to share them from a Democrat/Libertarian perspective
|
||||
|
||||
@@ -43,9 +50,9 @@ CONTEXT: You're chatting in a Discord server. Multiple people might be talking.
|
||||
const conversationHistory = new Map();
|
||||
const MAX_HISTORY_PER_CHANNEL = 10;
|
||||
|
||||
// Estimate token count (rough approximation: ~4 chars per token)
|
||||
function estimateTokens(text) {
|
||||
return Math.ceil(text.length / 4);
|
||||
function shouldUseWebSearch(prompt) {
|
||||
const lowerPrompt = prompt.toLowerCase();
|
||||
return WEB_SEARCH_TRIGGERS.some(trigger => lowerPrompt.includes(trigger));
|
||||
}
|
||||
|
||||
function buildMessages(history, prompt, mentionedUsername, systemPrompt) {
|
||||
@@ -54,28 +61,36 @@ function buildMessages(history, prompt, mentionedUsername, systemPrompt) {
|
||||
...history,
|
||||
{
|
||||
role: 'user',
|
||||
content: mentionedUsername ? `${mentionedUsername}: ${prompt}` : prompt,
|
||||
},
|
||||
content: mentionedUsername
|
||||
? `${mentionedUsername}: ${prompt}`
|
||||
: prompt
|
||||
}
|
||||
];
|
||||
return messages;
|
||||
}
|
||||
|
||||
async function callNanoGPT(messages, model) {
|
||||
function getModelWithSearch(model) {
|
||||
return `${model}:online`;
|
||||
}
|
||||
|
||||
async function callNanoGPT(messages, model, useWebSearch = false) {
|
||||
const modelToUse = useWebSearch ? getModelWithSearch(model) : model;
|
||||
|
||||
const response = await axios.post(
|
||||
'https://nano-gpt.com/api/v1/chat/completions',
|
||||
{
|
||||
model: model,
|
||||
model: modelToUse,
|
||||
messages: messages,
|
||||
temperature: 0.8,
|
||||
max_tokens: 500,
|
||||
max_tokens: 800,
|
||||
},
|
||||
{
|
||||
headers: {
|
||||
Authorization: `Bearer ${process.env.NANOGPT_API_KEY}`,
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
timeout: 60000,
|
||||
},
|
||||
timeout: 90000, // Longer timeout for web search
|
||||
}
|
||||
);
|
||||
return response.data.choices[0].message.content;
|
||||
}
|
||||
@@ -95,27 +110,62 @@ async function callFallbackAPI(messages) {
|
||||
'Content-Type': 'application/json',
|
||||
},
|
||||
timeout: 60000,
|
||||
},
|
||||
}
|
||||
);
|
||||
return response.data.choices[0].message.content;
|
||||
}
|
||||
|
||||
async function callNanoGPTWithRetry(messages, model, useWebSearch = false, maxRetries = 2) {
|
||||
let lastError;
|
||||
|
||||
for (let attempt = 0; attempt <= maxRetries; attempt++) {
|
||||
try {
|
||||
return await callNanoGPT(messages, model, useWebSearch);
|
||||
} catch (error) {
|
||||
lastError = error;
|
||||
console.error(`NanoGPT attempt ${attempt + 1} failed:`, error.message);
|
||||
|
||||
// If it's a web search failure, try without web search
|
||||
if (useWebSearch && attempt === 0) {
|
||||
console.log('Web search failed, retrying without...');
|
||||
try {
|
||||
return await callNanoGPT(messages, model, false);
|
||||
} catch (retryError) {
|
||||
console.error('Retry without web search also failed:', retryError.message);
|
||||
}
|
||||
}
|
||||
|
||||
// Wait before retry (exponential backoff)
|
||||
if (attempt < maxRetries) {
|
||||
await new Promise(resolve => setTimeout(resolve, 1000 * (attempt + 1)));
|
||||
}
|
||||
}
|
||||
}
|
||||
throw lastError;
|
||||
}
|
||||
|
||||
export async function getAIResponse(prompt, channelId, mentionedUsername) {
|
||||
const history = conversationHistory.get(channelId) || [];
|
||||
const messages = buildMessages(history, prompt, mentionedUsername, SASSY_SYSTEM_PROMPT);
|
||||
const model = process.env.NANO_MODEL || DEFAULT_MODEL;
|
||||
|
||||
// Determine if we should use web search
|
||||
const useWebSearch = shouldUseWebSearch(prompt);
|
||||
if (useWebSearch) {
|
||||
console.log('Web search enabled for query:', prompt.substring(0, 50));
|
||||
}
|
||||
|
||||
// Try NanoGPT first
|
||||
if (process.env.NANOGPT_API_KEY) {
|
||||
try {
|
||||
console.log('Attempting NanoGPT...');
|
||||
const aiResponse = await callNanoGPT(messages, model);
|
||||
|
||||
console.log(useWebSearch ? 'Attempting NanoGPT with web search...' : 'Attempting NanoGPT...');
|
||||
const aiResponse = await callNanoGPTWithRetry(messages, model, useWebSearch);
|
||||
|
||||
// Update history on success
|
||||
const newHistory = [
|
||||
...history,
|
||||
{ role: 'user', content: prompt },
|
||||
{ role: 'assistant', content: aiResponse },
|
||||
{ role: 'assistant', content: aiResponse }
|
||||
];
|
||||
|
||||
if (newHistory.length > MAX_HISTORY_PER_CHANNEL) {
|
||||
@@ -134,12 +184,12 @@ export async function getAIResponse(prompt, channelId, mentionedUsername) {
|
||||
try {
|
||||
console.log('Attempting fallback (z.ai)...');
|
||||
const aiResponse = await callFallbackAPI(messages);
|
||||
|
||||
|
||||
// Update history on success
|
||||
const newHistory = [
|
||||
...history,
|
||||
{ role: 'user', content: prompt },
|
||||
{ role: 'assistant', content: aiResponse },
|
||||
{ role: 'assistant', content: aiResponse }
|
||||
];
|
||||
|
||||
if (newHistory.length > MAX_HISTORY_PER_CHANNEL) {
|
||||
@@ -149,10 +199,7 @@ export async function getAIResponse(prompt, channelId, mentionedUsername) {
|
||||
|
||||
return aiResponse;
|
||||
} catch (fallbackError) {
|
||||
console.error(
|
||||
'Fallback API also failed:',
|
||||
fallbackError.response?.data || fallbackError.message,
|
||||
);
|
||||
console.error('Fallback API also failed:', fallbackError.response?.data || fallbackError.message);
|
||||
throw new Error('All AI providers failed');
|
||||
}
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user