optional semantic retrieval

tests/test_registry_service.py

@@ -3,6 +3,7 @@ import subprocess
 from repo_registry.core.service import RegistryService
 from repo_registry.llm_extraction import ExtractedAbility, ExtractedCapability
 from repo_registry.repo_ingestion.git import GitIngestionService
+from repo_registry.semantic import HashingEmbeddingProvider
 from repo_registry.storage.sqlite import NotFoundError, RegistryStore
 
 
@@ -290,6 +291,60 @@ def test_search_filters_by_status_language_and_framework(tmp_path):
     assert wrong_capability_results == []
 
 
+def test_semantic_search_adds_hybrid_matches_without_changing_text_default(tmp_path):
+    source = tmp_path / "repo"
+    source.mkdir()
+    (source / "README.md").write_text(
+        "# Queue Worker\n\nHandles postponed customer jobs.\n",
+        encoding="utf-8",
+    )
+
+    text_service = make_service(tmp_path)
+    repository = text_service.register_repository(
+        name="Queue Worker",
+        url=str(source),
+        description="Processes deferred customer work.",
+    )
+    ability_id = text_service.add_ability(
+        repository.id,
+        name="Background Job Processing",
+        description="Run deferred work outside request handling.",
+        confidence=0.8,
+    )
+    capability_id = text_service.add_capability(
+        repository.id,
+        ability_id,
+        name="Process Customer Tasks",
+        description="Execute queued customer tasks asynchronously.",
+        confidence=0.7,
+    )
+    text_service.add_feature(
+        repository.id,
+        capability_id,
+        name="worker task loop",
+        type="background worker",
+        location="worker.py",
+        confidence=0.6,
+    )
+    text_service.analyze_repository(repository.id)
+
+    assert text_service.search("customer queued") == []
+
+    semantic_service = RegistryService(
+        text_service.store,
+        ingestion=GitIngestionService(tmp_path / "checkouts"),
+        embedding_provider=HashingEmbeddingProvider(),
+    )
+    results = semantic_service.search("customer queued")
+
+    assert results
+    assert results[0].match_type in {"capability", "content_chunk"}
+    assert results[0].matched_field == "semantic"
+    assert results[0].vector_score > 0
+    assert results[0].hybrid_score >= results[0].vector_score * 0.35
+    assert any(result.match_type == "content_chunk" for result in results)
+
+
 def test_register_repository_imports_metadata_when_name_is_omitted(tmp_path):
     source = tmp_path / "metadata-source"
     source.mkdir()

tests/test_web_api.py

@@ -410,11 +410,25 @@ def test_api_service_settings_can_enable_llm_extractor(monkeypatch, tmp_path):
     assert service.llm_extractor is not None
 
 
+def test_api_service_settings_can_enable_hashing_embedding_provider(tmp_path):
+    service = get_service(
+        Settings(
+            database_path=str(tmp_path / "embedding-settings.sqlite3"),
+            checkout_root=str(tmp_path / "checkouts"),
+            embedding_provider="hashing",
+        )
+    )
+
+    assert service.embedding_provider is not None
+    assert service.embedding_provider.name == "hashing-v1"
+
+
 def test_settings_can_load_from_environment(monkeypatch):
     monkeypatch.setenv("REPO_REGISTRY_DATABASE_PATH", "var/env.sqlite3")
     monkeypatch.setenv("REPO_REGISTRY_CHECKOUT_ROOT", "var/env-checkouts")
     monkeypatch.setenv("REPO_REGISTRY_LLM_PROVIDER", "mock")
     monkeypatch.setenv("REPO_REGISTRY_LLM_MODEL", "demo-model")
+    monkeypatch.setenv("REPO_REGISTRY_EMBEDDING_PROVIDER", "hashing")
 
     settings = Settings()
 
@@ -422,6 +436,7 @@ def test_settings_can_load_from_environment(monkeypatch):
     assert settings.checkout_root == "var/env-checkouts"
     assert settings.llm_provider == "mock"
     assert settings.llm_model == "demo-model"
+    assert settings.embedding_provider == "hashing"
 
 
 def test_api_analysis_run_loop(tmp_path):