Feat: Add and integrate enrichment model support across backend and frontend

backend/app_settings.py

@@ -12,6 +12,7 @@ DEFAULT_BACKEND_API_URL = "http://127.0.0.1:8000"
 DEFAULT_OLLAMA_API_URL = "http://127.0.0.1:11434"
 DEFAULT_EMBED_MODEL = "nomic-embed-text:latest"
 DEFAULT_RERANK_MODEL = DEFAULT_EMBED_MODEL
+DEFAULT_ENRICHMENT_MODEL = "qwen3:4b"
 DEFAULT_TRANSCRIPTION_MODEL = "base"
 BGE_EMBED_MODEL = "bge-m3:latest"
 DEFAULT_SETTINGS: Dict[str, Any] = {
@@ -21,6 +22,7 @@ DEFAULT_SETTINGS: Dict[str, Any] = {
     "visionModel": "",
     "embedModel": DEFAULT_EMBED_MODEL,
     "rerankModel": DEFAULT_RERANK_MODEL,
+    "enrichmentModel": DEFAULT_ENRICHMENT_MODEL,
     "transcriptionModel": DEFAULT_TRANSCRIPTION_MODEL,
 }
 
@@ -125,6 +127,10 @@ def load_app_settings() -> Dict[str, Any]:
         settings["visionModel"] = settings.get("chatModel", "")
     settings["embedModel"] = normalize_embed_model(settings.get("embedModel"))
     settings["rerankModel"] = normalize_rerank_model(settings.get("rerankModel"))
+    settings["enrichmentModel"] = normalize_model_name(
+        settings.get("enrichmentModel"),
+        DEFAULT_ENRICHMENT_MODEL,
+    )
     settings["chatModel"] = normalize_model_name(settings.get("chatModel"))
     settings["visionModel"] = normalize_model_name(settings.get("visionModel"))
     settings["transcriptionModel"] = normalize_transcription_model(settings.get("transcriptionModel"))
@@ -147,6 +153,11 @@ def get_rerank_model_preference() -> str:
     return normalize_rerank_model(settings.get("rerankModel"))
 
 
+def get_enrichment_model_preference() -> str:
+    settings = load_app_settings()
+    return normalize_model_name(settings.get("enrichmentModel"), DEFAULT_ENRICHMENT_MODEL)
+
+
 def get_transcription_model_preference() -> str:
     settings = load_app_settings()
     return normalize_transcription_model(settings.get("transcriptionModel"))

backend/local_rag.py

@@ -22,7 +22,9 @@ from pydantic import BaseModel
 
 from .app_settings import (
     DEFAULT_EMBED_MODEL as DEFAULT_EMBED_MODEL_SETTING,
+    DEFAULT_ENRICHMENT_MODEL,
     get_embed_model_preference,
+    get_enrichment_model_preference,
     get_ollama_api_url,
 )
 from .paths import library_root
@@ -35,7 +37,7 @@ LIB_ROOT = library_root()
 RAW_CORPUS_PROFILE = "per-file-default-v1"
 PREPARE_PROFILE = "selective-enrich-v2"
 DEFAULT_EMBED_MODEL = DEFAULT_EMBED_MODEL_SETTING
-DEFAULT_ENRICH_MODEL = "qwen3:4b"
+DEFAULT_ENRICH_MODEL = DEFAULT_ENRICHMENT_MODEL
 DEFAULT_ENRICH_MIN_CHARS = 240
 DEFAULT_ENRICH_MAX_TEXT = 6000
 DEFAULT_ENRICH_CONCURRENCY = max(1, min(4, (os.cpu_count() or 4) // 2))
@@ -106,6 +108,10 @@ def _default_embed_model() -> str:
     return get_embed_model_preference()
 
 
+def _default_enrichment_model() -> str:
+    return get_enrichment_model_preference()
+
+
 def _resolve_ollama_url(value: Optional[str] = None) -> str:
     if isinstance(value, str) and value.strip():
         return value.strip().rstrip("/")
@@ -631,13 +637,15 @@ def _persist_item_metadata(slug: str, enhanced_path: Path) -> None:
                 "status": "ready" if ok else "fallback",
                 "headline": str(record.get("headline") or ""),
                 "summary": str(record.get("summary") or ""),
-                "keywords": list(record.get("keywords") or [])[:12],
-                "entities": list(record.get("entities") or [])[:12],
+                "keywords": list(record.get("keywords") or []),
+                "entities": list(record.get("entities") or []),
+                "qa": list(record.get("qa") or []),
                 "language": record.get("lang"),
                 "level": str(enrichment_meta.get("level") or "standard"),
                 "model": enrichment_meta.get("model"),
                 "strategy": enrichment_meta.get("strategy"),
                 "qa_count": len(record.get("qa") or []),
+                "quality_flags": list(enrichment_meta.get("quality_flags") or []),
                 "updated_at": now_iso(),
                 "error": enrichment_meta.get("error") if not ok else None,
             }
@@ -997,7 +1005,7 @@ def _run_selected_enrichment(slug: str, on_progress=None, **opts) -> Dict[str, A
         shadow_out=paths["shadow_partial"],
         on_progress=on_progress,
         ollama=_resolve_ollama_url(opts.get("ollama")),
-        model=opts.get("enrich_model", DEFAULT_ENRICH_MODEL),
+        model=opts.get("enrich_model") or _default_enrichment_model() or DEFAULT_ENRICH_MODEL,
         summary_lang=opts.get("summary_lang", "auto"),
         concurrency=opts.get("enrich_concurrency", DEFAULT_ENRICH_CONCURRENCY),
         min_chars=opts.get("min_chars", DEFAULT_ENRICH_MIN_CHARS),
@@ -1060,7 +1068,7 @@ def _run_prepare_pipeline(slug: str, on_progress=None, **opts):
             slug,
             on_progress=enrich_progress,
             ollama=_resolve_ollama_url(opts.get("ollama")),
-            enrich_model=opts.get("enrich_model", DEFAULT_ENRICH_MODEL),
+            enrich_model=opts.get("enrich_model") or _default_enrichment_model() or DEFAULT_ENRICH_MODEL,
             summary_lang=opts.get("summary_lang", "auto"),
             enrich_concurrency=opts.get("enrich_concurrency", DEFAULT_ENRICH_CONCURRENCY),
             min_chars=opts.get("min_chars", DEFAULT_ENRICH_MIN_CHARS),

src-tauri/src/main.rs

@@ -28,6 +28,7 @@ const BACKEND_SIDECAR_NAME: &str = "heimgeist-backend";
 const BACKEND_STARTUP_TIMEOUT: Duration = Duration::from_secs(45);
 const DEFAULT_OLLAMA_API_URL: &str = "http://127.0.0.1:11434";
 const DEFAULT_EMBED_MODEL: &str = "nomic-embed-text:latest";
+const DEFAULT_ENRICHMENT_MODEL: &str = "qwen3:4b";
 const DEFAULT_TRANSCRIPTION_MODEL: &str = "base";
 const BGE_EMBED_MODEL: &str = "bge-m3:latest";
 const DEFAULT_UI_SCALE: f64 = 1.0;
@@ -121,6 +122,10 @@ fn default_settings() -> SettingsMap {
     settings.insert("visionModel".into(), json!(""));
     settings.insert("embedModel".into(), json!(DEFAULT_EMBED_MODEL));
     settings.insert("rerankModel".into(), json!(DEFAULT_EMBED_MODEL));
+    settings.insert(
+        "enrichmentModel".into(),
+        json!(DEFAULT_ENRICHMENT_MODEL),
+    );
     settings.insert(
         "transcriptionModel".into(),
         json!(DEFAULT_TRANSCRIPTION_MODEL),
@@ -244,6 +249,14 @@ fn migrate_settings(source: Option<SettingsMap>) -> (SettingsMap, bool) {
         migrated = true;
     }
 
+    if !source.contains_key("enrichmentModel") {
+        next.insert(
+            "enrichmentModel".into(),
+            json!(DEFAULT_ENRICHMENT_MODEL),
+        );
+        migrated = true;
+    }
+
     if !source.contains_key("transcriptionModel") {
         next.insert(
             "transcriptionModel".into(),
@@ -263,6 +276,10 @@ fn normalize_settings(settings: &mut SettingsMap) {
     let vision_model = normalize_model_name(settings.get("visionModel"), "");
     let embed_model = normalize_embed_model(settings.get("embedModel"));
     let rerank_model = normalize_embed_model(settings.get("rerankModel"));
+    let enrichment_model = normalize_model_name(
+        settings.get("enrichmentModel"),
+        DEFAULT_ENRICHMENT_MODEL,
+    );
     let transcription_model = normalize_model_name(
         settings.get("transcriptionModel"),
         DEFAULT_TRANSCRIPTION_MODEL,
@@ -280,6 +297,7 @@ fn normalize_settings(settings: &mut SettingsMap) {
     settings.insert("visionModel".into(), json!(vision_model));
     settings.insert("embedModel".into(), json!(embed_model));
     settings.insert("rerankModel".into(), json!(rerank_model));
+    settings.insert("enrichmentModel".into(), json!(enrichment_model));
     settings.insert("transcriptionModel".into(), json!(transcription_model));
     settings.insert("uiScale".into(), json!(ui_scale));
     settings.insert(