🐝 Agent Swarm Intelligence

// 信息素机制示例 interface Pheromone { type: 'positive' | 'negative'; // 正向/负向信息素 strength: number; // 信息素强度 path: string[]; // 路径记录 timestamp: number; // 时间戳（用于衰减） } // Agent根据信息素做出决策 function makeDecision(agent: Agent, pheromones: Pheromone[]) { // 过滤有效信息素（衰减后） const validPheromones = pheromones .filter(p => calculateStrength(p) > threshold) .sort((a, b) => b.strength - a.strength); // 概率选择（强度越高概率越大） return weightedRandomSelect(validPheromones); }

// OpenClaw配置 - 群体智能模式 { "name": "content-swarm", "mode": "swarm", "agents": { "count": 10, "role": "content-writer", "interaction": { "type": "local", "range": 3, // 每个Agent只与最近的3个Agent通信 "rules": [ { "name": "avoid_duplicate", "condition": "topic_similarity > 0.8", "action": "diverge_topic" }, { "name": "align_style", "condition": "style_distance < 0.3", "action": "converge_style" } ] } }, "task": { "type": "generate_articles", "count": 50, "topic": "AI工具评测" } }

// 任务分配算法 class SwarmTaskAllocator { async allocate(tasks: Task[], agents: Agent[]) { // 1. 每个Agent发布"饥饿度"信息素 const hungerPheromones = agents.map(a => ({ agent: a, hunger: a.getLoad() / a.getCapacity(), position: a.getPosition() })); // 2. 任务向"饥饿"Agent聚集 for (const task of tasks) { const nearestHunger = hungerPheromones .sort((a, b) => { const distA = distance(task.position, a.position); const distB = distance(task.position, b.position); return (distA * a.hunger) - (distB * b.hunger); })[0]; await nearestHunger.agent.assign(task); } } }

// swarm-content-generator.ts import { OpenClaw, Agent, Task } from 'openclaw'; interface SwarmConfig { agentCount: number; interactionRange: number; pheromoneDecay: number; rules: SwarmRule[]; } interface SwarmRule { name: string; condition: (agent: Agent, neighbors: Agent[]) => boolean; action: (agent: Agent, neighbors: Agent[]) => Promise<void>; } class SwarmContentGenerator { private agents: Agent[] = []; private pheromoneMap: Map<string, Pheromone[]> = new Map(); constructor(private config: SwarmConfig) {} async initialize() { // 创建Agent群体 for (let i = 0; i < this.config.agentCount; i++) { const agent = await OpenClaw.createAgent({ role: 'content-writer', position: { x: Math.random(), y: Math.random() }, skills: ['writing', 'research', 'editing'] }); this.agents.push(agent); } } async generate(topic: string, count: number) { const results: Article[] = []; // 主循环：群体智能生成 while (results.length < count) { // 1. 每个Agent感知邻居 for (const agent of this.agents) { const neighbors = this.getNeighbors(agent); // 2. 应用群体规则 for (const rule of this.config.rules) { if (rule.condition(agent, neighbors)) { await rule.action(agent, neighbors); } } // 3. 生成内容 const article = await agent.generate({ topic, context: this.getPheromoneContext(agent), avoidDuplicate: true }); if (article) { results.push(article); // 4. 释放信息素 this.releasePheromone(agent, article); } } // 5. 信息素衰减 this.decayPheromones(); } return results; } private getNeighbors(agent: Agent): Agent[] { return this.agents .filter(a => a !== agent) .sort((a, b) => this.distance(agent, a) - this.distance(agent, b)) .slice(0, this.config.interactionRange); } private releasePheromone(agent: Agent, article: Article) { const pheromone: Pheromone = { type: 'positive', strength: article.quality, path: article.topics, timestamp: Date.now() }; const key = agent.id; if (!this.pheromoneMap.has(key)) { this.pheromoneMap.set(key, []); } this.pheromoneMap.get(key)!.push(pheromone); } } // 使用示例 const generator = new SwarmContentGenerator({ agentCount: 20, interactionRange: 3, pheromoneDecay: 0.1, rules: [ { name: 'avoid_duplicate', condition: (agent, neighbors) => { return neighbors.some(n => topicSimilarity(agent.currentTopic, n.currentTopic) > 0.8 ); }, action: async (agent) => { await agent.divergeTopic(); } } ] }); await generator.initialize(); const articles = await generator.generate('AI工具评测', 50);