# WEVAL Agents Architecture
## 13 Sovereign IAs | 4 Machines | 0€ provider cost

### Agent Topology
```
WEVIA Master (S204)
├── Director (autonomous cycles */30min)
├── L99 Brain (QA piloting, 18 crons)
├── WEDROID (backend diagnostic)
├── WEVCODE (code assistant)
├── WEVIA Life (chief of staff, email intel)
├── Ethica Agent (HCP pipeline 132K+)
├── Guardian (port protection */5min)
├── WEVADS IA (Arsenal intelligence, S95)
├── CodeRabbit (code review)
├── Blade IA (Windows OOB, v2.4)
├── DeerFlow (research agent)
└── OpenClaw (AI gateway, S151)
```

### Communication
- Master → Agents: PHP function calls + HTTP intents
- Master → S95: Sentinel exec (whitelist)
- Master → Windows: OOB chain (heartbeat → pending_exec → Invoke-Expression)
- Master → Providers: 7-tier cascade (Groq → HF → NVIDIA → Ollama → Cerebras → SambaNova → Mistral)

### Security
- All agents behind PHP session auth + HMAC 30j
- Public chatbot: SECURITE_PUBLIC guard (0 leak proven)
- Port protection: Guardian cron */5min, 18 ports monitored
- CrowdSec collaborative blocking
- chattr+i on critical files