Multica Multi-Agent Runtime

Multica workspace 多 Agent 协作长时间运行方案。

Overview

• Hybrid Runtime Architecture: Master Agent + Executor Pool + Shared State Store
• Collaboration Modes: Hierarchical / Sequential / Parallel
• Hybrid Persistence: Checkpoint + Event Log
• Hybrid Communication: Messages + Shared State

Architecture

┌─────────────────────────────────────────────────────────────┐
│                    Master Agent                              │
│  (任务分解、调度、状态管理、结果汇总)                          │
└─────────────────────────┬───────────────────────────────────┘
                          │
          ┌───────────────┼───────────────┐
          ▼               ▼               ▼
   ┌────────────┐  ┌────────────┐  ┌────────────┐
   │  Executor  │  │  Executor  │  │  Executor  │
   │   Pool     │  │   Pool     │  │   Pool     │
   │  (Planner) │  │ (Developer)│  │ (Reviewer) │
   └────────────┘  └────────────┘  └────────────┘
                          │
                          ▼
              ┌───────────────────────┐
              │   Shared State Store   │
              │  (Checkpoint + Event) │
              └───────────────────────┘

Project Structure

multica-multiagent-runtime/
├── src/
│   ├── master/
│   │   └── checkpoint_manager.py    # Master checkpoint management
│   ├── executor/
│   │   ├── __init__.py
│   │   ├── executor.py              # Executor base class
│   │   ├── pool.py                 # Executor pool management
│   │   └── protocol.py             # @mention protocol implementation
│   ├── checkpoint_manager/
│   │   ├── __init__.py
│   │   ├── file_store.py           # File-based checkpoint storage
│   │   ├── manager.py              # Checkpoint manager
│   │   ├── models.py               # Data models
│   │   └── store.py                # Store interface
│   └── shared-state/
│       ├── __init__.py
│       ├── base.py                  # Base shared state interface
│       ├── file_store.py           # File-based state storage
│       └── models.py                # State models
├── e2e-test-user-profile/           # E2E integration tests
│   ├── src/
│   │   ├── user_profile.py         # User profile feature
│   │   ├── checkpoint.py           # Checkpoint utilities
│   │   └── demo_checkpoint.py      # Checkpoint demo
│   └── tests/
│       └── test_user_profile.py    # Unit tests
├── tests/                           # Additional tests
├── docs/                            # Documentation
├── requirements.txt
└── README.md

Implementation Status

Phase 1: Core Infrastructure ✅

• Executor Agent Pool (src/executor/)
• Shared State Store (src/shared-state/)
• @mention Communication Protocol (src/executor/protocol.py)

Phase 2: Persistence ✅

• Checkpoint Manager (src/checkpoint_manager/, src/master/checkpoint_manager.py)
• Event Log mechanism
• State recovery flow

Phase 3: Error Handling ✅

• Error Classification
• Root Cause Analysis integration
• Recovery Strategies

Phase 4: Advanced Features 🔄

• Adaptive Executor scaling
• Intelligent task decomposition
• Performance monitoring

Executor Agent Pool

Agent	职责	场景
Planner Agent	任务分解与规划	复杂任务拆分
Developer Agent	代码实现	功能开发
Reviewer Agent	代码审查	PR/代码审查
SRE Agent	基础设施	部署/监控
Researcher Agent	调研分析	技术调研
Debugger Agent	调试排错	bug定位

Communication Protocol

@mention 触发条件：

• 分派任务给 Executor → @Executor
• 分派架构设计 → @Architect
• 分派开发任务 → @Developer
• 分派审查任务 → @Code Reviewer
• Executor 完成需验收 → @原分派者

Error Recovery

1. Root Cause Analysis First - 根因分析优先
2. Error Classification - Retryable / Recoverable / Fatal
3. Recovery Strategies - Retry with backoff / Rollback to checkpoint / Escalate

License

MIT