Architecture
Hermès Agent 系统架构文档
Overview
┌─────────────────────────────────────────────────────────────┐
│ Hermès Agent │
├─────────────────────────────────────────────────────────────┤
│ ┌───────────┐ ┌───────────┐ ┌───────────┐ │
│ │ Trend │ │ Review │ │ Skills │ │
│ │ Radar │ │ Agent │ │ Manager │ │
│ └───────────┘ └───────────┘ └───────────┘ │
├─────────────────────────────────────────────────────────────┤
│ Core Engine │
│ ┌───────────┐ ┌───────────┐ ┌───────────┐ │
│ │ Memory │ │ Skills │ │ Tools │ │
│ │ System │ │ Library │ │ Executor │ │
│ └───────────┘ └───────────┘ └───────────┘ │
├─────────────────────────────────────────────────────────────┤
│ External Services │
│ ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ │
│ │ GH │ │ TG │ │HF Hub│ │ Trend│ │Cloud │ │
│ │ │ │ │ │ │ │Radar │ │flare │ │
│ └──────┘ └──────┘ └──────┘ └──────┘ └──────┘ │
└─────────────────────────────────────────────────────────────┘
Components
1. Agent Core
- Session Management: Handles conversation state
- Context Compaction: Long context optimization
- Tool Orchestration: Coordinates tool execution
2. Memory System
| Level | Purpose | Persistence |
|---|---|---|
| L0 | Working context | Session |
| L1 | Scene memory | Short-term |
| L2 | Episodic memory | Medium-term |
| L3 | Semantic memory | Long-term |
3. Skills Engine
- Dynamic skill loading
- Context-aware skill matching
- Skill chaining for complex tasks
4. External Integrations
- GitHub: Repo management, PR workflow
- Telegram: Notifications, commands
- HuggingFace: Model Hub, datasets
- TrendRadar: News aggregation
Data Flow
User Input → Context Builder → Agent Core → Skills/Tools → Response
↑ ↓
└────────── Memory System ←─────────┘
Security
- Environment variable isolation
- API key encryption
- GitHub token scopes
- Branch protection rules
Last Updated: 2026-05-17