feat: unified_search + ai_proactive plugins with Ollama Cloud DeepSeek V4

- unified_search: Hybride Suche (PostgreSQL FTS + pgvector + RRF Fusion) - 5 Search Providers (Contact, Company, Mail, File, Event) - KI Query Understanding (Fuzzy, Facetten via LiteLLM) - DMS Text-Extraction (PDF, DOCX, XLSX, PPTX) - Embedding Pipeline (ollama/nomic-embed-text, 768 Dim) - Background Jobs für Indexierung - Plugin-basierte Provider Registry - ai_proactive: Proaktiver KI-Agent - Context-Tracking (Frontend → Backend → Event Bus) - Proactive Engine mit LLM Suggestion-Generierung - SSE Real-time Push an Frontend - 6 AI Tools für Tool Registry - Rate-Limiting + User Settings - Deep Analysis Background Jobs - Frontend Integration: - useAIContext Hook, SuggestionSidebar, SuggestionBadge - ProactiveAISettings Page, Search API Client - Globale Suche auf neue API umgestellt - Tests: test_unified_search.py + test_ai_proactive.py (alle bestanden) - Config: Ollama Cloud DeepSeek V4 als Default, konfigurierbar - Dependencies: PyMuPDF, python-docx, python-pptx, pgvector - Bugfixes: notification type_key length, migration IF NOT EXISTS
2026-07-18 11:21:51 +02:00
parent 4c9295de21
commit 8cebb4f4e9
52 changed files with 6493 additions and 41 deletions
@@ -0,0 +1,101 @@
+"""KI query understanding and result aggregation via LiteLLM."""
+
+from __future__ import annotations
+
+import os
+import json
+import logging
+from typing import Any
+
+import litellm
+
+logger = logging.getLogger(__name__)
+
+DEFAULT_LLM_MODEL = os.environ.get('SEARCH_LLM_MODEL', 'ollama/deepseek-v4')
+OLLAMA_API_KEY = os.environ.get('API_KEY_OLLAMA_CLOUD', '')
+
+QUERY_ANALYZE_SYSTEM = (
+    "Du bist ein Query-Analyzer fuer ein CRM. "
+    "Analysiere die Suchanfrage und gib JSON zurueck: "
+    '{"normalized_query": str, "entities": {"person": str|null, "company": str|null, "topic": str|null}, '
+    '"intent": str, "semantic_terms": [str], "suggested_filters": {}}'
+)
+
+RESULT_AGGREGATE_SYSTEM = (
+    "Du bist ein Result-Aggregator. Fasse Ergebnisse zusammen und generiere Facetten: "
+    '{"summary": str, "facets": {"types": {}, "dates": {}, "people": []}, "suggestions": [str]}'
+)
+
+
+def _fallback_query_analysis(query: str) -> dict[str, Any]:
+    return {
+        "normalized_query": query,
+        "entities": {},
+        "intent": "search",
+        "semantic_terms": [],
+        "suggested_filters": {},
+    }
+
+
+def _fallback_aggregate(results: list[dict], query: str) -> dict[str, Any]:
+    return {
+        "summary": f"{len(results)} Ergebnisse gefunden",
+        "facets": {},
+        "suggestions": [],
+    }
+
+
+async def llm_analyze_query(query: str) -> dict[str, Any]:
+    """Analyze a search query using LLM for intent, entities, and semantic terms.
+
+    Falls back to a simple dict if LLM fails.
+    """
+    try:
+        response = await litellm.acompletion(
+            model=DEFAULT_LLM_MODEL,
+            messages=[
+                {"role": "system", "content": QUERY_ANALYZE_SYSTEM},
+                {"role": "user", "content": query},
+            ],
+            temperature=0.1,
+            max_tokens=500,
+            response_format={"type": "json_object"},
+            api_key=OLLAMA_API_KEY,
+        )
+        content = response.choices[0].message.content
+        return json.loads(content)
+    except Exception:
+        logger.warning("LLM query analysis failed, using fallback")
+        return _fallback_query_analysis(query)
+
+
+async def llm_aggregate_results(results: list[dict], query: str) -> dict[str, Any]:
+    """Aggregate search results using LLM for summary, facets, and suggestions.
+
+    Falls back to a simple dict if LLM fails.
+    """
+    if not results:
+        return _fallback_aggregate(results, query)
+    try:
+        # Truncate results to avoid token overflow
+        compact = [
+            {"entity_type": r.get("entity_type"), "title": r.get("title", "")[:100]}
+            for r in results[:50]
+        ]
+        user_msg = json.dumps({"query": query, "results": compact})
+        response = await litellm.acompletion(
+            model=DEFAULT_LLM_MODEL,
+            messages=[
+                {"role": "system", "content": RESULT_AGGREGATE_SYSTEM},
+                {"role": "user", "content": user_msg},
+            ],
+            temperature=0.1,
+            max_tokens=1000,
+            response_format={"type": "json_object"},
+            api_key=OLLAMA_API_KEY,
+        )
+        content = response.choices[0].message.content
+        return json.loads(content)
+    except Exception:
+        logger.warning("LLM result aggregation failed, using fallback")
+        return _fallback_aggregate(results, query)