fix(hausmeister-bot): Booking-Werbefilter und LangGraph-Aufräumen
Filtert campaign@/marketing@-Absender vor IMPORTANT_SENDERS (taegliche Booking-Alerts). Entfernt obsoletes deep_research.py (CT121 LangGraph) — Tiefenrecherche laeuft ueber Perplexity.
This commit is contained in:
Homelab Cursor
parent
f0013a9fd6
commit
04f69dfd81
6 changed files with 15 additions and 296 deletions
|
|
@ -37,7 +37,7 @@ 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
|
||||
web_search, get_wordpress_stats
|
||||
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -194,6 +194,17 @@ def search_mail(query: str, days: int = 30, limit: int = 15) -> list[dict]:
|
|||
m.logout()
|
||||
|
||||
|
||||
MARKETING_FROM_PATTERNS = [
|
||||
"email.campaign@", "campaign@", "marketing@", "newsletter@",
|
||||
"promotion@", "promo@", "no-reply@mail.",
|
||||
]
|
||||
|
||||
|
||||
def _is_marketing_mail(frm_lower: str) -> bool:
|
||||
"""Werbemails von sonst wichtigen Absendern (z.B. booking.com campaign@)."""
|
||||
return any(p in frm_lower for p in MARKETING_FROM_PATTERNS)
|
||||
|
||||
|
||||
def _is_spam_sender(frm_lower: str) -> bool:
|
||||
"""Prueft ob Absender in der Spam-Liste steht."""
|
||||
return any(s in frm_lower for s in SPAM_SENDERS)
|
||||
|
|
@ -217,6 +228,8 @@ def get_important_mails(hours: int = 24) -> list[dict]:
|
|||
frm_lower = parsed["from"].lower()
|
||||
if _is_spam_sender(frm_lower):
|
||||
continue
|
||||
if _is_marketing_mail(frm_lower):
|
||||
continue
|
||||
if any(s in frm_lower for s in IMPORTANT_SENDERS):
|
||||
results.append(parsed)
|
||||
results.reverse()
|
||||
|
|
|
|||
|
|
@ -1,45 +0,0 @@
|
|||
from pathlib import Path
|
||||
p = Path('/root/homelab-brain/homelab-ai-bot/tools/deep_research.py')
|
||||
text = p.read_text()
|
||||
|
||||
# 1. Add datetime import
|
||||
text = text.replace(
|
||||
'import logging\nimport re\nimport time',
|
||||
'import logging\nimport re\nimport time\nfrom datetime import date'
|
||||
)
|
||||
|
||||
# 2. Add _inject_date_context function before _price_report_quality
|
||||
text = text.replace(
|
||||
'def _price_report_quality(report: str):',
|
||||
'''def _inject_date_context(query):
|
||||
today = date.today()
|
||||
return (
|
||||
f"Heutiges Datum: {today.strftime('%d. %B %Y')}. "
|
||||
f"'Letzte 3 Monate' = {(today.replace(day=1) - __import__('datetime').timedelta(days=90)).strftime('%B %Y')} bis {today.strftime('%B %Y')}.\n\n"
|
||||
+ query
|
||||
)
|
||||
|
||||
|
||||
def _price_report_quality(report: str):'''
|
||||
)
|
||||
|
||||
# 3. Add year check in _price_report_quality before return
|
||||
text = text.replace(
|
||||
' return len(missing) == 0, missing',
|
||||
''' today = date.today()
|
||||
ok_years = {str(today.year), str(today.year - 1)}
|
||||
found = set(re.findall(r"20\d{2}", text))
|
||||
if found and not (found & ok_years):
|
||||
missing.append(f"Jahre falsch: {found} statt {ok_years}")
|
||||
|
||||
return len(missing) == 0, missing'''
|
||||
)
|
||||
|
||||
# 4. Inject date context in handle_deep_research
|
||||
text = text.replace(
|
||||
' log.info("deep_research gestartet: %s", query[:120])',
|
||||
' query = _inject_date_context(query)\n log.info("deep_research gestartet: %s", query[:200])'
|
||||
)
|
||||
|
||||
p.write_text(text)
|
||||
print('OK')
|
||||
|
|
@ -1,27 +0,0 @@
|
|||
from pathlib import Path
|
||||
p = Path('/root/homelab-brain/homelab-ai-bot/tools/deep_research.py')
|
||||
text = p.read_text()
|
||||
|
||||
# Fix the broken _inject_date_context
|
||||
old_fn = '''def _inject_date_context(query):
|
||||
today = date.today()
|
||||
return (
|
||||
f"Heutiges Datum: {today.strftime('%d. %B %Y')}. "
|
||||
f"'Letzte 3 Monate' = {(today.replace(day=1) - __import__('datetime').timedelta(days=90)).strftime('%B %Y')} bis {today.strftime('%B %Y')}.\n\n"
|
||||
+ query
|
||||
)'''
|
||||
|
||||
new_fn = '''def _inject_date_context(query):
|
||||
from datetime import timedelta
|
||||
today = date.today()
|
||||
three_months_ago = today - timedelta(days=90)
|
||||
prefix = (
|
||||
"Heutiges Datum: " + today.strftime("%d. %B %Y") + ". "
|
||||
"'Letzte 3 Monate' = " + three_months_ago.strftime("%B %Y")
|
||||
+ " bis " + today.strftime("%B %Y") + ".\n\n"
|
||||
)
|
||||
return prefix + query'''
|
||||
|
||||
text = text.replace(old_fn, new_fn)
|
||||
p.write_text(text)
|
||||
print('OK')
|
||||
|
|
@ -1,222 +0,0 @@
|
|||
"""Deep Research Tool — Open Deep Research (CT 121) via LangGraph API."""
|
||||
|
||||
import logging
|
||||
import re
|
||||
import time
|
||||
|
||||
import requests
|
||||
|
||||
log = logging.getLogger("deep_research")
|
||||
|
||||
DEEP_RESEARCH_URL = "http://10.10.10.121:2024"
|
||||
ASSISTANT_ID = "e9a5370f-7a53-55a8-ada8-6ab9ef15bb5b"
|
||||
RESEARCH_MODEL = "openai/gpt-4o-mini"
|
||||
POLL_INTERVAL = 10
|
||||
MAX_WAIT = 600
|
||||
|
||||
SYSTEM_PROMPT_EXTRA = """DEEP RESEARCH:
|
||||
Du hast Zugriff auf deep_research — eine KI-gestuetzte Tiefenrecherche die 20-30 Quellen durchsucht.
|
||||
Nutze es NUR wenn der User explizit "deep research" oder "tiefenrecherche" sagt. Fuer alles andere: web_search.
|
||||
NICHT fuer einfache Fakten oder Homelab-Fragen.
|
||||
WICHTIG: deep_research dauert 2-5 Minuten. Das ist normal. Warte auf das Ergebnis.
|
||||
Das Ergebnis ist ein ausfuehrlicher Report. Fasse ihn fuer Telegram zusammen (max ~3000 Zeichen).
|
||||
|
||||
QUALITAET BEI PREISFRAGEN:
|
||||
- Liefere konkrete Zahlen statt allgemeiner Markttexte.
|
||||
- Zeige Zeitraum, Preis damals/heute, Delta in % und Quellen.
|
||||
- Wenn keine belastbaren Daten vorhanden sind, sage es explizit."""
|
||||
|
||||
TOOLS = [
|
||||
{
|
||||
"type": "function",
|
||||
"function": {
|
||||
"name": "deep_research",
|
||||
"description": "KI-gestuetzte Tiefenrecherche (20-30 Quellen, 2-5 Min). NUR wenn User explizit deep research oder tiefenrecherche sagt.",
|
||||
"parameters": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"query": {
|
||||
"type": "string",
|
||||
"description": "Die Recherche-Frage"
|
||||
}
|
||||
},
|
||||
"required": ["query"]
|
||||
},
|
||||
},
|
||||
},
|
||||
]
|
||||
|
||||
|
||||
def _create_thread():
|
||||
r = requests.post(f"{DEEP_RESEARCH_URL}/threads", json={}, timeout=10)
|
||||
r.raise_for_status()
|
||||
return r.json()["thread_id"]
|
||||
|
||||
|
||||
def _start_run(thread_id, query):
|
||||
payload = {
|
||||
"assistant_id": ASSISTANT_ID,
|
||||
"input": {"messages": [{"role": "user", "content": query}]},
|
||||
"config": {
|
||||
"configurable": {
|
||||
"summarization_model": f"openai:{RESEARCH_MODEL}",
|
||||
"research_model": f"openai:{RESEARCH_MODEL}",
|
||||
"compression_model": f"openai:{RESEARCH_MODEL}",
|
||||
"final_report_model": f"openai:{RESEARCH_MODEL}",
|
||||
"allow_clarification": False,
|
||||
}
|
||||
},
|
||||
}
|
||||
r = requests.post(
|
||||
f"{DEEP_RESEARCH_URL}/threads/{thread_id}/runs", json=payload, timeout=30
|
||||
)
|
||||
r.raise_for_status()
|
||||
return r.json()["run_id"]
|
||||
|
||||
|
||||
def _poll_run(thread_id, run_id):
|
||||
elapsed = 0
|
||||
while elapsed < MAX_WAIT:
|
||||
time.sleep(POLL_INTERVAL)
|
||||
elapsed += POLL_INTERVAL
|
||||
try:
|
||||
r = requests.get(
|
||||
f"{DEEP_RESEARCH_URL}/threads/{thread_id}/runs/{run_id}", timeout=10
|
||||
)
|
||||
r.raise_for_status()
|
||||
data = r.json()
|
||||
status = data.get("status", "unknown")
|
||||
log.info("Poll %ds: status=%s", elapsed, status)
|
||||
if status == "success":
|
||||
return True, None
|
||||
if status in ("error", "failed"):
|
||||
err = data.get("error", "Unbekannter Fehler")
|
||||
log.error("Run failed: %s", err)
|
||||
return False, err
|
||||
if status == "interrupted":
|
||||
return False, "Research wurde unterbrochen"
|
||||
except Exception as e:
|
||||
log.warning("Poll error at %ds: %s", elapsed, e)
|
||||
return False, f"Timeout nach {MAX_WAIT}s"
|
||||
|
||||
|
||||
def _get_result(thread_id):
|
||||
r = requests.get(f"{DEEP_RESEARCH_URL}/threads/{thread_id}/state", timeout=30)
|
||||
r.raise_for_status()
|
||||
state = r.json()
|
||||
|
||||
messages = state.get("values", {}).get("messages", [])
|
||||
log.info("Messages in result: %d", len(messages))
|
||||
for i, msg in enumerate(messages):
|
||||
content = msg.get("content", "")
|
||||
clen = len(content) if isinstance(content, str) else 0
|
||||
log.info(" msg[%d] type=%s len=%d", i, msg.get("type", "?"), clen)
|
||||
|
||||
for msg in reversed(messages):
|
||||
content = msg.get("content", "")
|
||||
if isinstance(content, str) and len(content) > 100:
|
||||
return content
|
||||
|
||||
return "Kein Report generiert."
|
||||
|
||||
|
||||
def _is_price_query(query: str) -> bool:
|
||||
q = (query or "").lower()
|
||||
needles = [
|
||||
"preis",
|
||||
"preise",
|
||||
"kosten",
|
||||
"teuer",
|
||||
"guenstig",
|
||||
"ram",
|
||||
"ddr4",
|
||||
"ddr5",
|
||||
"entwicklung",
|
||||
]
|
||||
return any(n in q for n in needles)
|
||||
|
||||
|
||||
def _price_report_quality(report: str):
|
||||
text = report or ""
|
||||
links = re.findall(r"https?://\S+", text)
|
||||
has_percent = bool(re.search(r"[-+]?\d+[\.,]?\d*\s*%", text))
|
||||
has_currency = bool(re.search(r"(?:\d+[\.,]?\d*\s?(?:€|eur|\$))|(?:€\s?\d+)", text, re.I))
|
||||
has_comparison = bool(
|
||||
re.search(r"(damals|heute|vor\s+\d+\s+(?:monaten|wochen)|aktuell|delta)", text, re.I)
|
||||
)
|
||||
|
||||
missing = []
|
||||
if len(links) < 3:
|
||||
missing.append("mindestens 3 konkrete Quellen-Links")
|
||||
if not has_percent:
|
||||
missing.append("Delta in %")
|
||||
if not has_currency:
|
||||
missing.append("konkrete Preise mit Waehrung")
|
||||
if not has_comparison:
|
||||
missing.append("Preisvergleich damals/heute")
|
||||
|
||||
return len(missing) == 0, missing
|
||||
|
||||
|
||||
def _run_research(query: str):
|
||||
thread_id = _create_thread()
|
||||
log.info("Thread erstellt: %s", thread_id)
|
||||
run_id = _start_run(thread_id, query)
|
||||
log.info("Run gestartet: %s", run_id)
|
||||
ok, error = _poll_run(thread_id, run_id)
|
||||
if not ok:
|
||||
return False, f"Deep Research fehlgeschlagen: {error}"
|
||||
report = _get_result(thread_id)
|
||||
log.info("Report erhalten: %d Zeichen", len(report))
|
||||
return True, report
|
||||
|
||||
|
||||
def handle_deep_research(query: str, **kw):
|
||||
log.info("deep_research gestartet: %s", query[:120])
|
||||
try:
|
||||
ok, report = _run_research(query)
|
||||
if not ok:
|
||||
return report
|
||||
|
||||
# Harte Qualitaetspruefung fuer Preisfragen
|
||||
if _is_price_query(query):
|
||||
good, missing = _price_report_quality(report)
|
||||
if not good:
|
||||
log.warning("Preisreport zu schwach, starte Retry. Missing: %s", ", ".join(missing))
|
||||
stricter_query = (
|
||||
query
|
||||
+ "\n\nLIEFERE NUR belastbare Preisdaten im Format:\n"
|
||||
+ "1) Zeitraum (exakt)\n"
|
||||
+ "2) Preis damals -> Preis heute (EUR)\n"
|
||||
+ "3) Delta in %\n"
|
||||
+ "4) 3-5 konkrete Quellen-Links (keine Startseiten).\n"
|
||||
+ "Wenn unklar: explizit keine belastbaren Preisdaten gefunden."
|
||||
)
|
||||
ok2, report2 = _run_research(stricter_query)
|
||||
if ok2:
|
||||
good2, missing2 = _price_report_quality(report2)
|
||||
if good2:
|
||||
report = report2
|
||||
else:
|
||||
return (
|
||||
"keine belastbaren Preisdaten gefunden. "
|
||||
"Es fehlen: "
|
||||
+ ", ".join(missing2)
|
||||
+ ". Bitte Anfrage enger formulieren (Produktklasse + Region + Zeitraum)."
|
||||
)
|
||||
else:
|
||||
return report2
|
||||
|
||||
if len(report) > 6000:
|
||||
report = report[:6000] + "\n\n[... Report gekuerzt]"
|
||||
return report
|
||||
|
||||
except requests.ConnectionError:
|
||||
log.error("CT 121 nicht erreichbar")
|
||||
return "Deep Research (CT 121) nicht erreichbar. Service laeuft moeglicherweise nicht."
|
||||
except Exception as e:
|
||||
log.exception("Deep Research Fehler")
|
||||
return f"Deep Research Fehler: {e}"
|
||||
|
||||
|
||||
HANDLERS = {"deep_research": handle_deep_research}
|
||||
|
|
@ -40,7 +40,7 @@ Gut: "DDR5 RAM 32GB preis 2026"
|
|||
Schlecht: "Was kostet aktuell eine Feinunze Gold in Euro?"
|
||||
Nutze web_search bei aktuellen Fakten, Preisen, News, Vergleichen.
|
||||
Mache 2-3 Suchen mit verschiedenen Keywords fuer bessere Ergebnisse.
|
||||
Nutze deep_research NUR wenn User explizit danach fragt."""
|
||||
Bei expliziter Tiefenrecherche (deep research / tiefenrecherche) antwortet das System automatisch via Perplexity Sonar — kein separates Tool."""
|
||||
|
||||
|
||||
def _search_once(base_url: str, query: str):
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue