docs: STATE.md + index.md aktualisiert — Ist-Zustand 21.03.2026
Routing, Tools, CT 121 Status, Deploy-Workflow dokumentiert. CT 121 Open Deep Research als gestoppt markiert.
This commit is contained in:
parent
8e72cf1180
commit
5a93736a60
1 changed files with 85 additions and 26 deletions
|
|
@ -1,34 +1,93 @@
|
|||
# Hausmeister Bot - STATE
|
||||
**Stand:** 21.03.2026
|
||||
**Status:** Saubere Local-First Architektur mit Sonar-Websuche
|
||||
**Status:** Produktiv, sauber, Local-First Architektur
|
||||
|
||||
## Architektur (3 Pfade)
|
||||
---
|
||||
|
||||
| Pfad | Modell | Endpoint | Zweck |
|
||||
|------|--------|----------|-------|
|
||||
| Text + Tools | qwen3:30b-a3b | Ollama lokal (RTX 3090) | Alle Homelab-Tools |
|
||||
| Vision | qwen3-vl:32b | Ollama lokal (RTX 3090) | Bilderkennung, OCR |
|
||||
| Websuche | perplexity/sonar | OpenRouter | Preise, News, Recherche |
|
||||
| Deep Research | CT 121 LangGraph | Direkt-API | Tiefenrecherche (explizit) |
|
||||
| Fallback | qwen2.5:14b | Ollama lokal | Bei Timeout |
|
||||
## LLM-Routing (4 Pfade)
|
||||
|
||||
## Routing (_route_model)
|
||||
- Web-Keywords (preis, recherche, news, etc.) -> Sonar via OpenRouter
|
||||
- Deep Research / Tiefenrecherche -> CT 121 direkt
|
||||
- Alles andere -> qwen3:30b-a3b lokal
|
||||
| Pfad | Trigger | Modell | Endpoint | Kosten |
|
||||
|------|---------|--------|----------|--------|
|
||||
| Lokal Text | Standard | qwen3:30b-a3b | Ollama RTX 3090 | $0.00 |
|
||||
| Lokal Vision | Bild/Foto | qwen3-vl:32b | Ollama RTX 3090 | $0.00 |
|
||||
| Online Suche | preis, gold, news, recherche, ... | perplexity/sonar | OpenRouter | ~$0.005/Anfrage |
|
||||
| Tiefensuche | tiefensuche, tiefenrecherche, ... | perplexity/sonar-deep-research | OpenRouter | ~$0.05-0.20/Anfrage |
|
||||
| Fallback | Ollama Timeout | qwen2.5:14b | Ollama RTX 3090 | $0.00 |
|
||||
|
||||
## Features
|
||||
- /no_think fuer Ollama-Modelle (schnellere Antworten)
|
||||
- Timeout-Fallback auf qwen2.5:14b
|
||||
- Passthrough-Tools (Grafana-Daten direkt durchreichen)
|
||||
- Memory-System + Session-History
|
||||
- 19 Tool-Module (auto-discovery via tool_loader)
|
||||
### Local-Override (immer lokal, auch bei Kosten-Keywords)
|
||||
api kosten, guthaben, openrouter, container, status, fehler, logs,
|
||||
feed, backup, memory, mail, seafile, forgejo, grafana, savetv, wordpress, matomo, tailscale
|
||||
|
||||
## Was funktioniert
|
||||
- Lokale KI steuert alle Homelab-Dienste (RSS, Proxmox, Loki, etc.)
|
||||
- Websuche laeuft ueber Perplexity Sonar (kein Tool-Calling, ein API-Call)
|
||||
- Vision lokal via qwen3-vl:32b
|
||||
- Deep Research via CT 121
|
||||
### Tiefensuche Trigger-Woerter
|
||||
tiefensuche, tiefenrecherche, tiefe suche, detailrecherche,
|
||||
ausfuehrliche recherche, vollstaendige recherche, recherchiere genau,
|
||||
analysiere genau, deep research
|
||||
|
||||
## Git-Stand
|
||||
Committed und nach Forgejo gepusht. Auto-Sync laeuft.
|
||||
---
|
||||
|
||||
## Tools (40 gesamt, auto-discovery via tool_loader)
|
||||
|
||||
get_api_costs, get_all_containers, get_container_detail, get_container_logs,
|
||||
get_errors, count_errors, get_silent_hosts, get_feed_stats,
|
||||
get_forgejo_status, create_issue, close_issue, get_grafana_status,
|
||||
get_temperaturen, get_energie, get_heizung, get_mail_summary,
|
||||
get_mail_count, search_mail, get_mails_by_period, get_smart_mail_digest,
|
||||
get_matomo_analytics, get_matomo_trend, memory_read, memory_suggest,
|
||||
session_search, session_summary, get_backup_status, get_server_metrics,
|
||||
get_server_warnings, get_savetv_status, get_savetv_tipps,
|
||||
get_savetv_archive_filme, savetv_record, savetv_download,
|
||||
get_seafile_status, get_service_directory, get_tailscale_status,
|
||||
web_search, get_wordpress_stats, deep_research
|
||||
|
||||
---
|
||||
|
||||
## CT 121 (deep-research)
|
||||
|
||||
| Service | Status | Anmerkung |
|
||||
|---------|--------|-----------|
|
||||
| SearXNG (Docker) | aktiv | genutzt von web_search Tool |
|
||||
| Open Deep Research | gestoppt + deaktiviert | ersetzt durch Sonar Deep Research |
|
||||
|
||||
Grund: Open Deep Research lieferte schlechte Qualitaet bei Preis- und
|
||||
Regionalanfragen (3 Min Wartezeit, kein Ergebnis). Sonar Deep Research
|
||||
liefert in 75s strukturierte Reports mit Quellen.
|
||||
|
||||
---
|
||||
|
||||
## KI-Server (RTX 3090, Muldenstein, 100.84.255.83)
|
||||
|
||||
| Modell | Typ | Groesse | Zweck |
|
||||
|--------|-----|---------|-------|
|
||||
| qwen3:30b-a3b | Text, MoE | 18.5 GB | Standard + Tools |
|
||||
| qwen3-vl:32b | Vision+Text | 20.9 GB | Bilder, OCR, Dokumente |
|
||||
| qwen2.5:14b | Text | 9 GB | Timeout-Fallback |
|
||||
|
||||
---
|
||||
|
||||
## OpenRouter API (Stand 21.03.2026)
|
||||
|
||||
Guthaben: $45.00 | Verbraucht: ~$24.50 | Rest: ~$20.50
|
||||
Verbrauch heute: $1.43 | Woche: $1.81
|
||||
Modelle: perplexity/sonar, perplexity/sonar-deep-research
|
||||
|
||||
---
|
||||
|
||||
## Deploy-Workflow (PFLICHT bei jeder Aenderung)
|
||||
|
||||
1. Auto-Sync deaktivieren (crontab + /etc/crontab)
|
||||
2. Editieren in /root/homelab-brain/ (= /opt/homelab-brain/)
|
||||
3. NIEMALS direkt in /opt/homelab-ai-bot/ editieren (Bind-Mount)
|
||||
4. git add -A && git commit && git push origin main
|
||||
5. pct exec 116 -- systemctl restart hausmeister-bot
|
||||
6. Logs pruefen: journalctl -u hausmeister-bot -n 20
|
||||
7. Auto-Sync wieder aktivieren
|
||||
|
||||
---
|
||||
|
||||
## Bekannte Eigenheiten
|
||||
|
||||
- Local-Override muss VOR Web-Trigger in _route_model stehen
|
||||
- /no_think wird an Ollama-Anfragen angehaengt (schnellere Antworten)
|
||||
- Passthrough-Tools: get_temperaturen, get_energie, get_heizung (Grafana direkt)
|
||||
- tool_loader cached Module — nach Code-Aenderungen Service neu starten
|
||||
- sync_state.py macht git reset --hard FETCH_HEAD — vernichtet nicht gepushte Commits
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue