139 lines
3.9 KiB
TypeScript
139 lines
3.9 KiB
TypeScript
import { modelConfigs, shedulerAICharacter } from '../../src/config/aiCharacters';
|
||
import OpenAI from 'openai';
|
||
|
||
interface AICharacter {
|
||
id: string;
|
||
name: string;
|
||
tags?: string[];
|
||
}
|
||
|
||
interface MessageHistory {
|
||
role: string;
|
||
content: string;
|
||
name: string;
|
||
}
|
||
|
||
export async function onRequestPost({ env, request }) {
|
||
console.log('scheduler');
|
||
try {
|
||
const { message, history, availableAIs } = await request.json();
|
||
const selectedAIs = await scheduleAIResponses(message, history, availableAIs, env);
|
||
|
||
return Response.json({
|
||
selectedAIs: selectedAIs
|
||
});
|
||
} catch (error) {
|
||
console.error(error);
|
||
return Response.json(
|
||
{ error: error.message },
|
||
{ status: 500 }
|
||
);
|
||
}
|
||
}
|
||
|
||
async function analyzeMessageWithAI(message: string, allTags: string[], env: any, history: MessageHistory[] = []): Promise<string[]> {
|
||
const shedulerAI = shedulerAICharacter(message, allTags);
|
||
const modelConfig = modelConfigs.find(config => config.model === shedulerAI.model);
|
||
const apiKey = env[modelConfig.apiKey];
|
||
if (!apiKey) {
|
||
throw new Error(`${modelConfig.model} 的API密钥未配置`);
|
||
}
|
||
const openai = new OpenAI({
|
||
apiKey: apiKey,
|
||
baseURL: modelConfig.baseURL,
|
||
});
|
||
|
||
const prompt = shedulerAI.custom_prompt;
|
||
|
||
try {
|
||
const completion = await openai.chat.completions.create({
|
||
model: shedulerAI.model,
|
||
messages: [
|
||
{ role: "user", content: prompt },
|
||
...history.slice(-10), // 添加历史消息
|
||
{ role: "user", content: message }
|
||
],
|
||
});
|
||
|
||
const matchedTags = completion.choices[0].message.content?.split(',').map(tag => tag.trim()) || [];
|
||
return matchedTags;
|
||
} catch (error) {
|
||
console.error('AI分析失败:', error.message);
|
||
return [];
|
||
}
|
||
}
|
||
|
||
async function scheduleAIResponses(
|
||
message: string,
|
||
history: MessageHistory[],
|
||
availableAIs: AICharacter[],
|
||
env: any
|
||
): Promise<string[]> {
|
||
// 1. 收集所有可用的标签
|
||
const allTags = new Set<string>();
|
||
availableAIs.forEach(ai => {
|
||
ai.tags?.forEach(tag => allTags.add(tag));
|
||
});
|
||
|
||
// 2. 使用AI模型分析消息并匹配标签
|
||
const matchedTags = await analyzeMessageWithAI(message, Array.from(allTags), env, history);
|
||
console.log('matchedTags', matchedTags, allTags);
|
||
//如果含有"文字游戏"标签，则需要全员参与
|
||
if (matchedTags.includes("文字游戏")) {
|
||
return availableAIs.map(ai => ai.id);
|
||
}
|
||
// 3. 计算每个AI的匹配分数
|
||
const aiScores = new Map<string, number>();
|
||
const messageLC = message.toLowerCase();
|
||
|
||
for (const ai of availableAIs) {
|
||
if (!ai.tags) continue;
|
||
|
||
let score = 0;
|
||
// 标签匹配分数
|
||
matchedTags.forEach(tag => {
|
||
if (ai.tags?.includes(tag)) {
|
||
score += 2; // 每个匹配的标签得2分
|
||
}
|
||
});
|
||
|
||
// 直接提到AI名字额外加分
|
||
if (messageLC.includes(ai.name.toLowerCase())) {
|
||
score += 5;
|
||
}
|
||
|
||
// 历史对话相关性加分
|
||
const recentHistory = history.slice(-5); // 只看最近5条消息
|
||
recentHistory.forEach(hist => {
|
||
if (hist.name === ai.name && hist.content.length > 0) {
|
||
score += 1; // 最近有参与对话的AI加分
|
||
}
|
||
});
|
||
|
||
if (score > 0) {
|
||
aiScores.set(ai.id, score);
|
||
}
|
||
}
|
||
|
||
// 4. 根据分数排序选择AI
|
||
const sortedAIs = Array.from(aiScores.entries())
|
||
.sort((a, b) => b[1] - a[1])
|
||
.map(([id]) => id);
|
||
|
||
// 5. 如果没有匹配到任何AI，随机选择1-2个
|
||
if (sortedAIs.length === 0) {
|
||
console.log('没有匹配到任何AI，随机选择1-2个');
|
||
const maxResponders = Math.min(2, availableAIs.length);
|
||
const numResponders = Math.floor(Math.random() * maxResponders) + 1;
|
||
|
||
const shuffledAIs = [...availableAIs]
|
||
.sort(() => Math.random() - 0.5)
|
||
.slice(0, numResponders);
|
||
|
||
return shuffledAIs.map(ai => ai.id);
|
||
}
|
||
|
||
// 6. 限制最大回复数量
|
||
const MAX_RESPONDERS = 3;
|
||
return sortedAIs.slice(0, MAX_RESPONDERS);
|
||
}
|