feat(indexation): disable topic tagging by default

openrag/core/config/indexation_pipeline.py

@@ -52,8 +52,10 @@ class IndexationPipelineConfig(BaseModel):
         default_factory=lambda: ["person", "organization", "location", "event"],
     )
 
-    # Topic tagging
-    enable_topic_tagging: bool = True
+    # Topic tagging. Off by default: the generated tags are not yet surfaced in
+    # the API or used for retrieval, so enabling it only buys an extra LLM call
+    # per document. Partitions that want the tags opt in on their preset.
+    enable_topic_tagging: bool = False
     max_topic_tags: int = Field(default=7, ge=1, le=50)
     topic_tagging_llm: str | None = None
 

openrag/services/orchestrators/preset_service.py

@@ -51,7 +51,7 @@
             # pymupdf-configured deployment. Named presets below opt in explicitly.
             "enable_image_captioning": True,
             "enable_entity_extraction": True,
-            "enable_topic_tagging": True,
+            "enable_topic_tagging": False,
         },
         "legal": {
             "chunking": {"name": "recursive_splitter", "chunk_size": 1024, "chunk_overlap_rate": 0.25},

openrag/services/workers/indexer_actor.py

@@ -159,7 +159,10 @@ async def _replace_topic_tags_if_needed(
         return
 
     has_topic_tags = "topic_tags" in row
-    topic_tagging_disabled = indexation_config is not None and indexation_config.get("enable_topic_tagging") is False
+    # Mirrors IndexationPipelineConfig.enable_topic_tagging: an absent key means
+    # disabled, so a re-index under a preset that never enabled tagging still
+    # clears tags left behind by an earlier run.
+    topic_tagging_disabled = indexation_config is not None and not indexation_config.get("enable_topic_tagging", False)
     if not has_topic_tags and not topic_tagging_disabled:
         return
 

openrag/services/workers/indexer_pool.py

@@ -359,7 +359,7 @@ def _required_llm_names(indexation_config: dict[str, Any] | None) -> list[str]:
     names: list[str] = []
     if indexation_config.get("enable_contextualization"):
         names.append(indexation_config.get("contextualization_llm") or "default")
-    if indexation_config.get("enable_topic_tagging", True):
+    if indexation_config.get("enable_topic_tagging", False):
         names.append(indexation_config.get("topic_tagging_llm") or "default")
     return names
 

tests/unit/core/config/test_pipeline_configs.py

@@ -41,6 +41,11 @@ def test_indexation_pipeline_rejects_unknown_contextualization_mode():
         IndexationPipelineConfig(contextualization_mode="verbose")
 
 
+def test_indexation_pipeline_topic_tagging_defaults_off():
+    """Topic tagging is opt-in: tags are not yet surfaced or used in retrieval."""
+    assert IndexationPipelineConfig().enable_topic_tagging is False
+
+
 def test_retrieval_pipeline_rejects_unknown_type():
     """Retrieval presets reject unsupported retrieval modes."""
     with pytest.raises(ValidationError):

tests/unit/services/orchestrators/test_preset_service.py

@@ -155,6 +155,17 @@ async def test_seed_defaults_inserts_six_presets_when_empty():
     assert len(upserts) == 6  # 3 indexation + 3 retrieval
 
 
+@pytest.mark.asyncio
+async def test_seed_defaults_indexation_leaves_topic_tagging_off():
+    repo = _FakePresetRepo()
+    svc = _make_service(repo)
+    await svc.seed_defaults()
+
+    idx = [r for r in repo._store.values() if r["preset_type"] == "indexation"]
+    assert idx  # sanity: indexation presets were seeded
+    assert all(r["config"].get("enable_topic_tagging", False) is False for r in idx)
+
+
 @pytest.mark.asyncio
 async def test_seed_defaults_retrieval_inherits_reranker_enabled():
     from core.config.root import Settings

tests/unit/services/workers/test_indexer_pool.py

@@ -480,9 +480,10 @@ def test_required_llm_names_mirrors_pipeline_selection() -> None:
     from services.workers.indexer_pool import _required_llm_names
 
     assert _required_llm_names(None) == []
-    # Topic tagging defaults on, contextualization defaults off.
-    assert _required_llm_names({}) == ["default"]
+    # Both topic tagging and contextualization default off.
+    assert _required_llm_names({}) == []
     assert _required_llm_names({"enable_topic_tagging": False}) == []
+    assert _required_llm_names({"enable_topic_tagging": True}) == ["default"]
     assert _required_llm_names(
         {
             "enable_contextualization": True,

ui/src/pages/admin/presets.tsx

@@ -402,7 +402,7 @@ function IndexationPresetForm({
         />
         <FeatureToggle
           label="Topic tagging"
-          enabled={configGet(config, "enable_topic_tagging", true)}
+          enabled={configGet(config, "enable_topic_tagging", false)}
           onToggle={(on) => toggleFeature("enable_topic_tagging", "topic_tagging_llm", on)}
           // Postponed: tags are generated but not yet surfaced or used in
           // retrieval, so the control is disabled until the feature ships.