Replace keyword detection with proper tool calling for web search
Deploy to NAS / deploy (push) Successful in 1m11s
Deploy to NAS / deploy (push) Successful in 1m11s
- Define web_search function that AI can call when needed - AI decides autonomously when to search (no keyword matching) - Tool calling via NanoGPT API with function definitions - Format search results with sources for citations
This commit is contained in:
+133
-58
@@ -1,4 +1,4 @@
|
|||||||
// ai.js - NanoGPT API wrapper for kekbot with fallback and web search support
|
// ai.js - NanoGPT API wrapper for kekbot with fallback and web search tools
|
||||||
// Copyright (C) 2025 Luis Bauza
|
// Copyright (C) 2025 Luis Bauza
|
||||||
|
|
||||||
import axios from 'axios';
|
import axios from 'axios';
|
||||||
@@ -7,12 +7,24 @@ const DEFAULT_MODEL = 'GLM-4.5-Air-Derestricted';
|
|||||||
const FALLBACK_MODEL = 'GLM-4.7-Flash';
|
const FALLBACK_MODEL = 'GLM-4.7-Flash';
|
||||||
const FALLBACK_API_URL = 'https://api.z.ai/api/coding/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
|
// Tool definitions for function calling
|
||||||
const WEB_SEARCH_TRIGGERS = [
|
const WEB_SEARCH_TOOL = {
|
||||||
'latest', 'news', 'current', 'today', 'yesterday', 'this week',
|
type: 'function',
|
||||||
'what is the', 'who is', 'what happened', 'search', 'find',
|
function: {
|
||||||
'how do i', 'how to', 'what\'s the best', 'top ', '2024', '2025', '2026'
|
name: 'web_search',
|
||||||
];
|
description: 'Search the web for current information. Use this when you need up-to-date facts, news, or information that may have changed recently.',
|
||||||
|
parameters: {
|
||||||
|
type: 'object',
|
||||||
|
properties: {
|
||||||
|
query: {
|
||||||
|
type: 'string',
|
||||||
|
description: 'The search query to find relevant web results.'
|
||||||
|
}
|
||||||
|
},
|
||||||
|
required: ['query']
|
||||||
|
}
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
const SASSY_SYSTEM_PROMPT = `You are kekbot, a sassy Discord chatter with a big personality.
|
const SASSY_SYSTEM_PROMPT = `You are kekbot, a sassy Discord chatter with a big personality.
|
||||||
|
|
||||||
@@ -26,7 +38,7 @@ CORE PERSONALITY:
|
|||||||
|
|
||||||
OPINIONS:
|
OPINIONS:
|
||||||
- You're a true gamer with strong opinions about games
|
- 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
|
- 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
|
- You have views on politics and aren't afraid to share them from a Democrat/Libertarian perspective
|
||||||
|
|
||||||
@@ -43,6 +55,7 @@ RESPONSE RULES:
|
|||||||
- Don't use markdown formatting excessively - plain text with personality
|
- Don't use markdown formatting excessively - plain text with personality
|
||||||
- You can disagree with the user, you're not a yes-man
|
- You can disagree with the user, you're not a yes-man
|
||||||
- Stay in character always
|
- Stay in character always
|
||||||
|
- Use the web_search tool when you need current information
|
||||||
|
|
||||||
CONTEXT: You're chatting in a Discord server. Multiple people might be talking. Pay attention to who you're responding to.`;
|
CONTEXT: You're chatting in a Discord server. Multiple people might be talking. Pay attention to who you're responding to.`;
|
||||||
|
|
||||||
@@ -50,11 +63,6 @@ CONTEXT: You're chatting in a Discord server. Multiple people might be talking.
|
|||||||
const conversationHistory = new Map();
|
const conversationHistory = new Map();
|
||||||
const MAX_HISTORY_PER_CHANNEL = 10;
|
const MAX_HISTORY_PER_CHANNEL = 10;
|
||||||
|
|
||||||
function shouldUseWebSearch(prompt) {
|
|
||||||
const lowerPrompt = prompt.toLowerCase();
|
|
||||||
return WEB_SEARCH_TRIGGERS.some(trigger => lowerPrompt.includes(trigger));
|
|
||||||
}
|
|
||||||
|
|
||||||
function buildMessages(history, prompt, mentionedUsername, systemPrompt) {
|
function buildMessages(history, prompt, mentionedUsername, systemPrompt) {
|
||||||
const messages = [
|
const messages = [
|
||||||
{ role: 'system', content: systemPrompt },
|
{ role: 'system', content: systemPrompt },
|
||||||
@@ -69,17 +77,101 @@ function buildMessages(history, prompt, mentionedUsername, systemPrompt) {
|
|||||||
return messages;
|
return messages;
|
||||||
}
|
}
|
||||||
|
|
||||||
function getModelWithSearch(model) {
|
async function performWebSearch(query) {
|
||||||
return `${model}:online`;
|
const response = await axios.post(
|
||||||
|
'https://nano-gpt.com/api/web',
|
||||||
|
{
|
||||||
|
query: query,
|
||||||
|
provider: 'linkup',
|
||||||
|
depth: 'standard',
|
||||||
|
outputType: 'sourcedAnswer'
|
||||||
|
},
|
||||||
|
{
|
||||||
|
headers: {
|
||||||
|
Authorization: `Bearer ${process.env.NANOGPT_API_KEY}`,
|
||||||
|
'Content-Type': 'application/json',
|
||||||
|
},
|
||||||
|
timeout: 60000,
|
||||||
|
}
|
||||||
|
);
|
||||||
|
|
||||||
|
const data = response.data.data;
|
||||||
|
const metadata = response.data.metadata;
|
||||||
|
|
||||||
|
// Format the response nicely
|
||||||
|
let formattedResult = '';
|
||||||
|
|
||||||
|
if (typeof data === 'string') {
|
||||||
|
formattedResult = data;
|
||||||
|
} else if (data.answer) {
|
||||||
|
formattedResult = data.answer;
|
||||||
|
if (data.sources && data.sources.length > 0) {
|
||||||
|
formattedResult += '\n\nSources:\n';
|
||||||
|
data.sources.forEach((source, i) => {
|
||||||
|
formattedResult += `${i + 1}. ${source.name || source.url}\n`;
|
||||||
|
});
|
||||||
|
}
|
||||||
|
} else if (data.results) {
|
||||||
|
formattedResult = 'Search results:\n';
|
||||||
|
data.results.slice(0, 5).forEach((result, i) => {
|
||||||
|
formattedResult += `${i + 1}. ${result.title || result.name}\n`;
|
||||||
|
if (result.url) formattedResult += ` ${result.url}\n`;
|
||||||
|
if (result.content) formattedResult += ` ${result.content.substring(0, 200)}...\n`;
|
||||||
|
});
|
||||||
}
|
}
|
||||||
|
|
||||||
async function callNanoGPT(messages, model, useWebSearch = false) {
|
return formattedResult;
|
||||||
const modelToUse = useWebSearch ? getModelWithSearch(model) : model;
|
}
|
||||||
|
|
||||||
|
async function callNanoGPTWithTools(messages, model, maxRetries = 2) {
|
||||||
|
let lastError;
|
||||||
|
|
||||||
|
for (let attempt = 0; attempt <= maxRetries; attempt++) {
|
||||||
|
try {
|
||||||
const response = await axios.post(
|
const response = await axios.post(
|
||||||
'https://nano-gpt.com/api/v1/chat/completions',
|
'https://nano-gpt.com/api/v1/chat/completions',
|
||||||
{
|
{
|
||||||
model: modelToUse,
|
model: model,
|
||||||
|
messages: messages,
|
||||||
|
tools: [WEB_SEARCH_TOOL],
|
||||||
|
temperature: 0.8,
|
||||||
|
max_tokens: 800,
|
||||||
|
},
|
||||||
|
{
|
||||||
|
headers: {
|
||||||
|
Authorization: `Bearer ${process.env.NANOGPT_API_KEY}`,
|
||||||
|
'Content-Type': 'application/json',
|
||||||
|
},
|
||||||
|
timeout: 90000,
|
||||||
|
}
|
||||||
|
);
|
||||||
|
|
||||||
|
const choice = response.data.choices[0];
|
||||||
|
const message = choice.message;
|
||||||
|
|
||||||
|
// Check if AI wants to call a tool
|
||||||
|
if (message.tool_calls && message.tool_calls.length > 0) {
|
||||||
|
for (const toolCall of message.tool_calls) {
|
||||||
|
if (toolCall.function.name === 'web_search') {
|
||||||
|
const args = JSON.parse(toolCall.function.arguments);
|
||||||
|
console.log('AI requested web search for:', args.query);
|
||||||
|
|
||||||
|
// Perform the search
|
||||||
|
const searchResult = await performWebSearch(args.query);
|
||||||
|
|
||||||
|
// Add tool result to messages and continue
|
||||||
|
messages.push(message);
|
||||||
|
messages.push({
|
||||||
|
role: 'tool',
|
||||||
|
tool_call_id: toolCall.id,
|
||||||
|
content: searchResult
|
||||||
|
});
|
||||||
|
|
||||||
|
// Get final response with search results
|
||||||
|
const finalResponse = await axios.post(
|
||||||
|
'https://nano-gpt.com/api/v1/chat/completions',
|
||||||
|
{
|
||||||
|
model: model,
|
||||||
messages: messages,
|
messages: messages,
|
||||||
temperature: 0.8,
|
temperature: 0.8,
|
||||||
max_tokens: 800,
|
max_tokens: 800,
|
||||||
@@ -89,10 +181,28 @@ async function callNanoGPT(messages, model, useWebSearch = false) {
|
|||||||
Authorization: `Bearer ${process.env.NANOGPT_API_KEY}`,
|
Authorization: `Bearer ${process.env.NANOGPT_API_KEY}`,
|
||||||
'Content-Type': 'application/json',
|
'Content-Type': 'application/json',
|
||||||
},
|
},
|
||||||
timeout: 90000, // Longer timeout for web search
|
timeout: 90000,
|
||||||
}
|
}
|
||||||
);
|
);
|
||||||
return response.data.choices[0].message.content;
|
|
||||||
|
return finalResponse.data.choices[0].message.content;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
// No tool calls, return direct response
|
||||||
|
return message.content;
|
||||||
|
|
||||||
|
} catch (error) {
|
||||||
|
lastError = error;
|
||||||
|
console.error(`NanoGPT attempt ${attempt + 1} failed:`, error.message);
|
||||||
|
|
||||||
|
if (attempt < maxRetries) {
|
||||||
|
await new Promise(resolve => setTimeout(resolve, 1000 * (attempt + 1)));
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
throw lastError;
|
||||||
}
|
}
|
||||||
|
|
||||||
async function callFallbackAPI(messages) {
|
async function callFallbackAPI(messages) {
|
||||||
@@ -115,51 +225,16 @@ async function callFallbackAPI(messages) {
|
|||||||
return response.data.choices[0].message.content;
|
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) {
|
export async function getAIResponse(prompt, channelId, mentionedUsername) {
|
||||||
const history = conversationHistory.get(channelId) || [];
|
const history = conversationHistory.get(channelId) || [];
|
||||||
const messages = buildMessages(history, prompt, mentionedUsername, SASSY_SYSTEM_PROMPT);
|
const messages = buildMessages(history, prompt, mentionedUsername, SASSY_SYSTEM_PROMPT);
|
||||||
const model = process.env.NANO_MODEL || DEFAULT_MODEL;
|
const model = process.env.NANO_MODEL || DEFAULT_MODEL;
|
||||||
|
|
||||||
// Determine if we should use web search
|
// Try NanoGPT first (with tool calling)
|
||||||
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) {
|
if (process.env.NANOGPT_API_KEY) {
|
||||||
try {
|
try {
|
||||||
console.log(useWebSearch ? 'Attempting NanoGPT with web search...' : 'Attempting NanoGPT...');
|
console.log('Attempting NanoGPT with tool calling...');
|
||||||
const aiResponse = await callNanoGPTWithRetry(messages, model, useWebSearch);
|
const aiResponse = await callNanoGPTWithTools(messages, model);
|
||||||
|
|
||||||
// Update history on success
|
// Update history on success
|
||||||
const newHistory = [
|
const newHistory = [
|
||||||
|
|||||||
Reference in New Issue
Block a user