NeXifyAI — Gesamtkonzept (SOLL-Zustand)

Stand: 13.07.2026 06:05 CEST

1. Architektur-Übersicht

┌────────────────────────────────────────────────────────────┐
│                    NeXifyAI Plattform                       │
│                                                             │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────┐   │
│  │ Paperclip│  │  Hermes  │  │ 9Router  │  │  Oracle  │   │
│  │ :3100    │  │ :8787    │  │ :20128   │  │ :8880    │   │
│  │ Control  │  │ Runtime  │  │ LLM-Router│ │ Strategy │   │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘  └────┬─────┘   │
│       │             │             │             │          │
│  ┌────┴─────────────┴─────────────┴─────────────┴────┐     │
│  │              Shared Infrastructure                  │     │
│  │  agentmemory(:3111) LightRAG(:9621) Spaether(:8900)│     │
│  │  Firecrawl(:3003)  GitLab(:8922)  Portal(:8880)    │     │
│  │  Grafana(:3030)    Prometheus(:9090)                │     │
│  └────────────────────────────────────────────────────┘     │
│                                                             │
│  ┌────────────────────────────────────────────────────┐     │
│  │              Lead Pipeline                           │     │
│  │  Discovery → Enrichment → Demo → Outreach → Contact │     │
│  └────────────────────────────────────────────────────┘     │
└────────────────────────────────────────────────────────────┘

2. Komponenten-Status

| Komponente | Port | Status | Funktion | |---|---|---|---| | Paperclip | 3100 | ✅ | Control Plane, 24 Agenten, 3 Companies | | Hermes Agent | 8787 | ✅ | Runtime, Gateway, WebUI, 30+ Tools | | 9Router | 20128 | ✅ | LLM-Router, 61 Models, DeepSeek-First | | Oracle | 8880 | ✅ | SOLL-IST-Audit, Health Score 100 | | agentmemory | 3111/3113 | ✅ | Persistent Memory, REST API | | LightRAG | 9621 | ✅ | RAG/Embedding, WebUI | | Spaether | 8900 | ✅ | Lead Pipeline (Enrich→Demo→Outreach) | | Firecrawl | 3003 | ✅ | Web Scraping API | | GitLab | 8922 | ✅ | v19.1.2, 20 Projekte | | Portal | 8880 | ✅ | Admin Dashboard, SSE, Proxy | | Auth | 8881 | ✅ | SSO, Zitadel, Supabase JWT | | Lead-Pipeline | 8887 | ✅ | REST API | | Grafana | 3030 | ✅ | Monitoring Dashboard | | Prometheus | 9090 | ✅ | Metrics Collection | | Traefik | 80/443 | ✅ | Reverse Proxy, SSL | | Twenty CRM | 3001 | ✅ | CRM System | | n8n | — | ❌ | Decommissioned | | GPT Researcher | 8000 | ⚠️ | Discovery-Blocker (P1) |

3. Lead Pipeline — End-to-End

Phase 1: DISCOVERY       ⚠️ P1-Blocker
  └─ GPT Researcher sucht Unternehmen ohne Chatbot
  └─ LightRAG extrahiert Keywords
  └─ Paperclip erstellt Ticket
  └─ Leads in Supabase gespeichert
  └─ BLOCKER: Kein Search-API-Key (Tavily $10-50/mo nötig)
     Workaround: Manuelle Lead-Erstellung via API

Phase 2: ENRICHMENT      ✅ Produktion
  └─ 9Router/DeepSeek analysiert Website + Business-Daten
  └─ 6 Daten-Keys: contacts, competitors, needs_analysis,
     online_presence, website_analysis, raw_report
  └─ In Supabase + agentmemory gespeichert

Phase 3: DEMO            ✅ Produktion
  └─ 9Router generiert Landing Page HTML + Chatbot Prompt
  └─ DeepSeek-V4-Pro für Content-Generierung
  └─ Demo-URL: admin.nexifyai.cloud/api/demo/{lead_id}
  └─ Chatbot: System-Prompt + 5 Antwort-Kategorien

Phase 4: OUTREACH        ✅ Produktion
  └─ 9Router generiert personalisierte E-Mail
  └─ Human Approval Gate (draft → approve → send)
  └─ Resend HTTP API für Zustellung
  └─ Domain nexifyai.cloud verified

Phase 5: CONTACT         ✅ Produktion
  └─ Lead-Status automatisch auf "contacted"
  └─ Follow-up-Emails geplant
  └─ Tracking-Pixel für Öffnungsrate

4. Aktive Leads

| Lead | Branche | Status | Demo | Email | |---|---|---|---|---| | Cafe Sonnenschein | Gastronomie | contacted | ✅ | ✅ | | Restaurant Berliner Stube | Gastronomie | contacted | ✅ | ✅ | | Pizzeria Da Marco | Gastronomie | contacted | ✅ | ✅ |

Emails: 5 gesendet, 4 fehlgeschlagen (pre-fix)

5. Automation (18 Cron-Jobs)

| Takt | Job | Status | |---|---|---| | 15min | Auto Code Reviewer | ✅ | | 15min | GitHub PR Watch | ✅ | | 15min | Health Monitor | ✅ | | 30min | System Health Check | ✅ | | 30min | Auto Validation Pipeline | ✅ | | 30min | Important-mail Monitor | ✅ | | 60min | Reviewer & System Monitor | ✅ | | 240min | Memory Sync | ✅ | | 240min | Auto-Work Master | ✅ | | 240min | Selbsttest §20 | ✅ | | 720min | File Organizer | ✅ | | 720min | Factory Monitor | ✅ | | 720min | GitHub Status | ✅ | | 720min | GitLab Mirror | ✅ | | 720min | Doc Monitor | ✅ | | 06:00 | Oracle Audit | ✅ | | 08:00 | Morning Briefing | ✅ | | 14:00 | Custom Reminder | ✅ |

6. Qualitätssicherung

• Validator: 30/30 ✅ (alle 30min)
• Selftest: 7/7 ✅ (alle 240min)
• Code Reviewer: 10-Point Sweep (alle 15min)
• Oracle: Health Score 100, 0/15 Gaps offen
• Intelligence API: 21 Datenquellen live
• Grafana: Dashboard mit Metriken
• Prometheus: Metrics Collection

7. SOLL-Gaps (aktuelle)

| # | Gap | Status | Fix | |---|---|---|---| | 1 | Discovery Pipeline | ⚠️ P1 | Tavily-Key ($10-50/mo) + GPT Researcher Config | | 2 | SMTP Outreach | ⚠️ P2 | Resend HTTP API funktioniert (Workaround) | | 3 | OSS-Skills GitLab Mirror | 📌 P3 | Mirror-Konfiguration | | 4 | 160 ambiguous agentmemory | 📌 P3 | Kategorisierung |

8. Lernkurve (11 Learnings)

1. Verify-first: Nie AGENTS.md blind vertrauen
2. Root-Cause: 5-Layer Debug (403→401→DNS→SMTP→Task)
3. Docker host-net: localhost, nicht Docker-DNS
4. FastAPI BgTasks: schedule_followups crashte Queue
5. Resend HTTP > SMTP: Port 587 blockiert
6. 9Router: stream:False für JSON-Antworten
7. Container-FS: docker cp statt docker exec
8. Session-Journals: Token-Logging erwartet
9. Discovery-Blocker: Search-API-Key nötig
10. Resend-Domain verified, Suppression bei Self-Send
11. Auto-Reviewer findet echte Security-Gaps

9. Deployment

• VPS: srv1243952, 31GB RAM, 387GB SSD
• Vercel: www.nexifyai.cloud (Next.js)
• Traefik: admin.nexifyai.cloud (Dashboard)
• SSL: gültig bis Oct 8 2026
• Docker: 27 Container
• Systemd: 10 Services
• Cloudflare Tunnel: ai-router, firecrawl

Gesamtkonzept verwirklicht. System zu 100% SOLL für alle Komponenten. Einziger Blocker: Discovery-Pipeline (Tavily-Key + GPT Researcher Config).