Initial commit: breakpilot-core - Shared Infrastructure
Docker Compose with 24+ services: - PostgreSQL (PostGIS), Valkey, MinIO, Qdrant - Vault (PKI/TLS), Nginx (Reverse Proxy) - Backend Core API, Consent Service, Billing Service - RAG Service, Embedding Service - Gitea, Woodpecker CI/CD - Night Scheduler, Health Aggregator - Jitsi (Web/XMPP/JVB/Jicofo), Mailpit Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
286
docs-src/architecture/multi-agent.md
Normal file
286
docs-src/architecture/multi-agent.md
Normal file
@@ -0,0 +1,286 @@
|
||||
# Multi-Agent Architektur - Entwicklerdokumentation
|
||||
|
||||
**Status:** Implementiert
|
||||
**Modul:** `/agent-core/`
|
||||
|
||||
---
|
||||
|
||||
## 1. Übersicht
|
||||
|
||||
Die Multi-Agent-Architektur erweitert Breakpilot um ein verteiltes Agent-System basierend auf Mission Control Konzepten.
|
||||
|
||||
### Kernkomponenten
|
||||
|
||||
| Komponente | Pfad | Beschreibung |
|
||||
|------------|------|--------------|
|
||||
| Session Management | `/agent-core/sessions/` | Lifecycle & Recovery |
|
||||
| Shared Brain | `/agent-core/brain/` | Langzeit-Gedächtnis |
|
||||
| Orchestrator | `/agent-core/orchestrator/` | Koordination |
|
||||
| SOUL Files | `/agent-core/soul/` | Agent-Persönlichkeiten |
|
||||
|
||||
---
|
||||
|
||||
## 2. Agent-Typen
|
||||
|
||||
| Agent | Aufgabe | SOUL-Datei |
|
||||
|-------|---------|------------|
|
||||
| **TutorAgent** | Lernbegleitung, Fragen beantworten | `tutor-agent.soul.md` |
|
||||
| **GraderAgent** | Klausur-Korrektur, Bewertung | `grader-agent.soul.md` |
|
||||
| **QualityJudge** | BQAS Qualitätsprüfung | `quality-judge.soul.md` |
|
||||
| **AlertAgent** | Monitoring, Benachrichtigungen | `alert-agent.soul.md` |
|
||||
| **Orchestrator** | Task-Koordination | `orchestrator.soul.md` |
|
||||
|
||||
---
|
||||
|
||||
## 3. Wichtige Dateien
|
||||
|
||||
### Session Management
|
||||
```
|
||||
agent-core/sessions/
|
||||
├── session_manager.py # AgentSession, SessionManager, SessionState
|
||||
├── heartbeat.py # HeartbeatMonitor, HeartbeatClient
|
||||
└── checkpoint.py # CheckpointManager
|
||||
```
|
||||
|
||||
### Shared Brain
|
||||
```
|
||||
agent-core/brain/
|
||||
├── memory_store.py # MemoryStore, Memory (mit TTL)
|
||||
├── context_manager.py # ConversationContext, ContextManager
|
||||
└── knowledge_graph.py # KnowledgeGraph, Entity, Relationship
|
||||
```
|
||||
|
||||
### Orchestrator
|
||||
```
|
||||
agent-core/orchestrator/
|
||||
├── message_bus.py # MessageBus, AgentMessage, MessagePriority
|
||||
├── supervisor.py # AgentSupervisor, AgentInfo, AgentStatus
|
||||
└── task_router.py # TaskRouter, RoutingRule, RoutingResult
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 4. Datenbank-Schema
|
||||
|
||||
Die Migration befindet sich in:
|
||||
`/backend/migrations/add_agent_core_tables.sql`
|
||||
|
||||
### Tabellen
|
||||
|
||||
1. **agent_sessions** - Session-Daten mit Checkpoints
|
||||
2. **agent_memory** - Langzeit-Gedächtnis mit TTL
|
||||
3. **agent_messages** - Audit-Trail für Inter-Agent Kommunikation
|
||||
|
||||
### Helper-Funktionen
|
||||
|
||||
```sql
|
||||
-- Abgelaufene Memories bereinigen
|
||||
SELECT cleanup_expired_agent_memory();
|
||||
|
||||
-- Inaktive Sessions bereinigen
|
||||
SELECT cleanup_stale_agent_sessions(48); -- 48 Stunden
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 5. Integration Voice-Service
|
||||
|
||||
Der `EnhancedTaskOrchestrator` erweitert den bestehenden `TaskOrchestrator`:
|
||||
|
||||
```python
|
||||
# voice-service/services/enhanced_task_orchestrator.py
|
||||
|
||||
from agent_core.sessions import SessionManager
|
||||
from agent_core.orchestrator import MessageBus
|
||||
|
||||
class EnhancedTaskOrchestrator(TaskOrchestrator):
|
||||
# Nutzt Session-Checkpoints für Recovery
|
||||
# Routet komplexe Tasks an spezialisierte Agents
|
||||
# Führt Quality-Checks via BQAS durch
|
||||
```
|
||||
|
||||
**Wichtig:** Der Enhanced Orchestrator ist abwärtskompatibel und kann parallel zum Original verwendet werden.
|
||||
|
||||
---
|
||||
|
||||
## 6. Integration BQAS
|
||||
|
||||
Der `QualityJudgeAgent` integriert BQAS mit dem Multi-Agent-System:
|
||||
|
||||
```python
|
||||
# voice-service/bqas/quality_judge_agent.py
|
||||
|
||||
from bqas.judge import LLMJudge
|
||||
from agent_core.orchestrator import MessageBus
|
||||
|
||||
class QualityJudgeAgent:
|
||||
# Wertet Responses in Echtzeit aus
|
||||
# Nutzt Memory für konsistente Bewertungen
|
||||
# Empfängt Evaluierungs-Requests via Message Bus
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 7. Code-Beispiele
|
||||
|
||||
### Session erstellen
|
||||
|
||||
```python
|
||||
from agent_core.sessions import SessionManager
|
||||
|
||||
manager = SessionManager(redis_client=redis, db_pool=pool)
|
||||
session = await manager.create_session(
|
||||
agent_type="tutor-agent",
|
||||
user_id="user-123"
|
||||
)
|
||||
```
|
||||
|
||||
### Memory speichern
|
||||
|
||||
```python
|
||||
from agent_core.brain import MemoryStore
|
||||
|
||||
store = MemoryStore(redis_client=redis, db_pool=pool)
|
||||
await store.remember(
|
||||
key="student:123:progress",
|
||||
value={"level": 5, "score": 85},
|
||||
agent_id="tutor-agent",
|
||||
ttl_days=30
|
||||
)
|
||||
```
|
||||
|
||||
### Nachricht senden
|
||||
|
||||
```python
|
||||
from agent_core.orchestrator import MessageBus, AgentMessage
|
||||
|
||||
bus = MessageBus(redis_client=redis)
|
||||
await bus.publish(AgentMessage(
|
||||
sender="orchestrator",
|
||||
receiver="grader-agent",
|
||||
message_type="grade_request",
|
||||
payload={"exam_id": "exam-1"}
|
||||
))
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 8. Tests ausführen
|
||||
|
||||
```bash
|
||||
# Alle Agent-Core Tests
|
||||
cd agent-core && pytest -v
|
||||
|
||||
# Mit Coverage-Report
|
||||
pytest --cov=. --cov-report=html
|
||||
|
||||
# Einzelne Module
|
||||
pytest tests/test_session_manager.py -v
|
||||
pytest tests/test_message_bus.py -v
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 9. Deployment-Schritte
|
||||
|
||||
### 1. Migration ausführen
|
||||
|
||||
```bash
|
||||
psql -h localhost -U breakpilot -d breakpilot \
|
||||
-f backend/migrations/add_agent_core_tables.sql
|
||||
```
|
||||
|
||||
### 2. Voice-Service aktualisieren
|
||||
|
||||
```bash
|
||||
# Sync zu Server
|
||||
rsync -avz --exclude 'node_modules' --exclude '.git' \
|
||||
/path/to/breakpilot-pwa/ server:/path/to/breakpilot-pwa/
|
||||
|
||||
# Container neu bauen
|
||||
docker compose build --no-cache voice-service
|
||||
|
||||
# Starten
|
||||
docker compose up -d voice-service
|
||||
```
|
||||
|
||||
### 3. Verifizieren
|
||||
|
||||
```bash
|
||||
# Session-Tabelle prüfen
|
||||
psql -c "SELECT COUNT(*) FROM agent_sessions;"
|
||||
|
||||
# Memory-Tabelle prüfen
|
||||
psql -c "SELECT COUNT(*) FROM agent_memory;"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 10. Monitoring
|
||||
|
||||
### Metriken
|
||||
|
||||
| Metrik | Beschreibung |
|
||||
|--------|--------------|
|
||||
| `agent_session_count` | Anzahl aktiver Sessions |
|
||||
| `agent_heartbeat_delay_ms` | Zeit seit letztem Heartbeat |
|
||||
| `agent_message_latency_ms` | Nachrichtenlatenz |
|
||||
| `agent_memory_count` | Gespeicherte Memories |
|
||||
| `agent_routing_success_rate` | Erfolgreiche Routings |
|
||||
|
||||
### Health-Check-Endpunkte
|
||||
|
||||
```
|
||||
GET /api/v1/agents/health # Supervisor Status
|
||||
GET /api/v1/agents/sessions # Aktive Sessions
|
||||
GET /api/v1/agents/memory/stats # Memory-Statistiken
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 11. Troubleshooting
|
||||
|
||||
### Problem: Session nicht gefunden
|
||||
|
||||
1. Prüfen ob Valkey läuft: `redis-cli ping`
|
||||
2. Session-Timeout prüfen (default 24h)
|
||||
3. Heartbeat-Status checken
|
||||
|
||||
### Problem: Message Bus Timeout
|
||||
|
||||
1. Redis Pub/Sub Status prüfen
|
||||
2. Ziel-Agent registriert?
|
||||
3. Timeout erhöhen (default 30s)
|
||||
|
||||
### Problem: Memory nicht gefunden
|
||||
|
||||
1. Namespace korrekt?
|
||||
2. TTL abgelaufen?
|
||||
3. Cleanup-Job gelaufen?
|
||||
|
||||
---
|
||||
|
||||
## 12. Erweiterungen
|
||||
|
||||
### Neuen Agent hinzufügen
|
||||
|
||||
1. SOUL-Datei erstellen in `/agent-core/soul/`
|
||||
2. Routing-Regel in `task_router.py` hinzufügen
|
||||
3. Handler beim Supervisor registrieren
|
||||
4. Tests schreiben
|
||||
|
||||
### Neuen Memory-Typ hinzufügen
|
||||
|
||||
1. Key-Schema definieren (z.B. `student:*:progress`)
|
||||
2. TTL festlegen
|
||||
3. Access-Pattern dokumentieren
|
||||
|
||||
---
|
||||
|
||||
## 13. Referenzen
|
||||
|
||||
- **Agent-Core README:** `/agent-core/README.md`
|
||||
- **Migration:** `/backend/migrations/add_agent_core_tables.sql`
|
||||
- **Voice-Service Integration:** `/voice-service/services/enhanced_task_orchestrator.py`
|
||||
- **BQAS Integration:** `/voice-service/bqas/quality_judge_agent.py`
|
||||
- **Tests:** `/agent-core/tests/`
|
||||
Reference in New Issue
Block a user