MultiAgentSwarm

Name: MultiAgentSwarm
Author: huanglynch

huanglynch February 23, 2026

1 copies 0 downloads

Self-evolving ReAct Multi-Agent Swarm | Balanced模式省60-80% token | 全程思考可见 + 后台自我进化 | 支持Ollama/DeepSeek本地部署 | 🇨🇳 中文版

Agent Definition

<image-card alt="Stars" src="https://img.shields.io/github/stars/huanglynch/MultiAgentSwarm?style=social" ></image-card>
<image-card alt="License" src="https://img.shields.io/github/license/huanglynch/MultiAgentSwarm" ></image-card>

# MultiAgentSwarm v4.1.0 (ReAct Visualization Edition)
**Self-Adaptive Digital Team**  
**A fully visible, self-evolving ReAct Digital Team** that intelligently decides *when to be fast* and *when to go deep*.

---

**MultiAgentSwarm v4.1.0** 不再是简单的“多个LLM并行聊天”，而是一个**完全可视化、智能决策、自进化 ReAct 数字团队** —— 它完美还原经典 ReAct 架构，同时注入群体智能、动态规划、分层终身记忆与生产级交付能力。

<p align="center">
  <strong>
    <a href="#english-version">🇬🇧 English Version</a>
      |  
    <a href="#chinese-version">🇨🇳 中文版</a>
  </strong>
</p>

<p align="center">
  <img src="images/architecture-diagram.png" alt="MultiAgentSwarm v3.2.0 Architecture" width="95%" style="border-radius: 12px; box-shadow: 0 10px 30px rgba(0,0,0,0.3);">
</p>

---

## English Version <a id="english-version"></a>

**MultiAgentSwarm v4.1.0**  
**A fully visible, self-evolving ReAct Digital Team** that intelligently decides *when to be fast* and *when to go deep*.

### Core Philosophy (First Principles)

We rejected the “use everything on every task” anti-pattern.  
Instead, we built a swarm that **knows itself** — automatically choosing the right level of intelligence for each request.

This is **the only swarm that**:
- **Intelligently routes** every task: Simple / Medium / Balanced / Complex
- **Forces visible ReAct thinking** in every single agent response
- Delivers **Balanced mode** as the default sweet spot (3 agents, planning + tools, best quality/speed ratio)
- Runs **Tree-of-Thoughts** with 3 concurrent exploration branches
- Spawns **Dynamic Agents** on-the-fly with true parallel subtask execution (Hierarchical Supervisor)
- Maintains **lifelong hierarchical memory** (PrimalMemory + Vector + Knowledge Graph) with automatic decay & distillation
- Performs **all self-evolution completely in the backgr

Comments

More Agents

View all

hybrid-model-workflow

HAL 分层混合模型工作流 — 强模型(Claude)负责理解/拆解/验收，低成本模型(DeepSeek)负责检索/提取/清洗。Hermes Agent skill。

ph4ble

Dynamic-Review-Agent

An LLM agent fine-tuned on DeepSeek for spaced repetition, dynamically integrating knowledge points based on the Ebbinghaus forgetting curve.

1838177

StellarOS-Watch

基于 STM32F103 构建的端到端 AI 智能手表生态。自研“零重定位”原生机器码动态加载引擎与页面栈式 UI 框架；集成生产级 OTA 回滚保护机制与高带宽（921600 baud）串口协议栈。通过 Node.js 中继实现 DeepSeek AI 语义控制及 ASRPRO 语音全双工交互，是一个集成了分布式计算、现代存储管理与 AI Agent 的嵌入式全栈工程。

chenshuang888