Feature/GLiNER-module

.github/workflows/feature-testing.yml

@@ -27,16 +27,27 @@ jobs:
           version: "latest"
 
       - name: Install dependencies
-        run: uv sync
+        run: uv sync --all-groups
 
       - name: Start API for testing
         run: |
           uv run api.py &
           echo "API_PID=$!" >> $GITHUB_ENV
-          sleep 15  # wait for SpaCy model load + pre-warm
+
+          # Wait for API to be ready (with healthcheck loop, max 2 minutes)
+          for i in {1..120}; do
+            if curl -s http://localhost:8080/api/v1/health | grep -q pong; then
+              echo "API is ready"
+              exit 0
+            fi
+            echo "Waiting for API... ($i/120)"
+            sleep 1
+          done
+          echo "API failed to start within 120 seconds"
+          exit 1
 
       - name: Run test suite
-        run: pytest tests/ -q
+        run: uv run pytest tests/ -q
 
       - name: Stop API
         if: always()

CHANGELOG.md

@@ -4,6 +4,13 @@ Alle belangrijke wijzigingen in dit project worden in dit bestand gedocumenteerd
 
 De opmaak is gebaseerd op [Keep a Changelog](https://keepachangelog.com/en/1.1.0/) en dit project maakt gebruik van [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.4.0] - 2026-04-03
+
+### Changed
+- Presidio GLiNER Recognizer toegevoegd als module
+- Custom pattern recognizers (BSN/IBAN/etc.) op false gezet
+- Skip pattern recognizers tests
+
 ## [1.3.0] - 2026-03-02
 
 ### Removed

pyproject.toml

@@ -11,10 +11,13 @@ dependencies = [
     "presidio-anonymizer>=2.2.358",
     "nl_core_news_lg",
     "python-dotenv>=1.1.0",
+    "gliner>=0.1.13"
 ]
+
 authors = [
     { name = "Mark Westerweel", email = "mark.westerweel@conduction.nl" },
     { name = "Razo van Berkel", email = "razo.van.berkel@centric.eu" },
+    { name = "Nena Meijer", email = "nena.meijer@centric.eu" },
 ]
 
 [tool.uv.sources]

src/api/plugins.yaml

@@ -20,51 +20,66 @@ ner:
 recognizers:
   - name: DutchPhoneNumberRecognizer
     type: pattern
-    enabled: true
+    enabled: false
 
   - name: DutchIBANRecognizer
     type: pattern
-    enabled: true
+    enabled: false
 
   - name: DutchBSNRecognizer
     type: pattern
-    enabled: true
+    enabled: false
 
   - name: DutchDateRecognizer
     type: pattern
-    enabled: true
+    enabled: false
 
   - name: EmailRecognizer
     type: pattern
-    enabled: true
+    enabled: false
 
   - name: DutchPassportIdRecognizer
     type: pattern
-    enabled: true
+    enabled: false
 
   - name: DutchDriversLicenseRecognizer
     type: pattern
-    enabled: true
+    enabled: false
 
   - name: DutchVATRecognizer
     type: pattern
-    enabled: true
+    enabled: false
 
   - name: DutchKvKRecognizer
     type: pattern
-    enabled: true
+    enabled: false
 
   - name: DutchLicensePlateRecognizer
     type: pattern
-    enabled: true
+    enabled: false
 
   - name: IPv4Recognizer
     type: pattern
-    enabled: true
+    enabled: false
 
   - name: CaseNumberRecognizer
     type: pattern
+    enabled: false
+
+  - name: GLiNERRecognizer
+    type: gliner
     enabled: true
+    model: "urchade/gliner_multi_pii-v1"
+    entity_mapping:
+      person: PERSON
+      name: PERSON
+      organization: ORGANIZATION
+      location: LOCATION
+      gpe: LOCATION
+    flat_ner: false
+    multi_label: true
+    map_location: "cpu"
+    supported_language: "nl"
 
   # Optionele modules — standaard uitgeschakeld.
   # Schakel in door enabled: true te zetten en de vereiste packages te installeren.

src/api/services/text_analyzer.py

@@ -55,6 +55,15 @@ def _build_analyzer() -> AnalyzerEngine:
     for recognizer in plugin_cfg.recognizers:
         registry.add_recognizer(recognizer)
 
+    # If GLiNER is enabled, remove spaCy recognizer to avoid NER coming from spaCy
+    has_gliner = any(
+        recognizer.__class__.__name__ == "GLiNERRecognizer"
+        for recognizer in plugin_cfg.recognizers
+    )
+    if has_gliner:
+        registry.remove_recognizer("SpacyRecognizer")
+
+
     engine = AnalyzerEngine(
         nlp_engine=nlp_engine,
         registry=registry,

src/api/utils/plugin_loader.py

@@ -91,6 +91,24 @@ def _load_llm_recognizer(cfg: dict[str, Any]) -> EntityRecognizer:
     )
 
 
+def _load_gliner_recognizer(cfg: dict[str, Any]) -> EntityRecognizer:
+    "Laad de GLiNER recognizer (lazy import)."
+    try: from presidio_analyzer.predefined_recognizers import GLiNERRecognizer
+    except ImportError as exc:
+        raise ImportError(
+            f"GLiNER plugin '{cfg['name']}' vereist package: 'presidio-analyzer[gliner]'. "
+            f"Fout: {exc}"
+        ) from exc
+    return GLiNERRecognizer(
+        model_name=cfg["model"],
+        supported_language=cfg["supported_language"],
+        entity_mapping=cfg.get("entity_mapping", {}),
+        flat_ner=cfg.get("flat_ner", False),
+        multi_label=cfg.get("multi_label", True),
+        map_location=cfg.get("map_location", "cpu"),
+    )
+
+
 def load_plugins(plugins_path: Path | None = None) -> PluginConfig:
     """Laad plugin-configuratie en instantieer alle actieve recognizers.
 
@@ -163,6 +181,11 @@ def load_plugins(plugins_path: Path | None = None) -> PluginConfig:
                 recognizer = _load_llm_recognizer(plugin)
                 recognizers.append(recognizer)
                 logger.debug("LLM plugin geladen: %s", name)
+
+            elif plugin_type == "gliner":
+                recognizer = _load_gliner_recognizer(plugin)
+                recognizers.append(recognizer)
+                logger.debug("GLiNER plugin geladen: %s", name)
 
             else:
                 logger.warning("Onbekend plugin type '%s' voor '%s', overgeslagen.", plugin_type, name)

tests/test_string_endpoints.py

@@ -5,6 +5,7 @@
 """
 
 import httpx
+import pytest
 
 
 class TestHealth:
@@ -140,6 +141,7 @@ def test_entity_filter_preserves_non_targeted(self, client: httpx.Client) -> Non
 
 
 class TestAnonymizeStrategies:
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_replace_uses_entity_placeholder(self, client: httpx.Client) -> None:
         r = client.post(
             "/api/v1/anonymize",
@@ -155,6 +157,7 @@ def test_replace_uses_entity_placeholder(self, client: httpx.Client) -> None:
         assert "<PHONE_NUMBER>" in data["anonymized_text"]
         assert "0612345678" not in data["anonymized_text"]
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_redact_removes_value(self, client: httpx.Client) -> None:
         r = client.post(
             "/api/v1/anonymize",
@@ -168,6 +171,7 @@ def test_redact_removes_value(self, client: httpx.Client) -> None:
         assert r.status_code == 200
         assert "0612345678" not in r.json()["anonymized_text"]
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_hash_replaces_with_hex_string(self, client: httpx.Client) -> None:
         r = client.post(
             "/api/v1/anonymize",
@@ -183,6 +187,7 @@ def test_hash_replaces_with_hex_string(self, client: httpx.Client) -> None:
         assert "NL91ABNA0417164300" not in data["anonymized_text"]
         assert data["anonymized_text"] != data["original_text"]
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_mask_inserts_asterisks(self, client: httpx.Client) -> None:
         r = client.post(
             "/api/v1/anonymize",

tests/test_supported_entities.py

@@ -7,6 +7,7 @@
 """
 
 import httpx
+import pytest
 
 
 def _detected_types(client: httpx.Client, text: str, entities: list[str]) -> set[str]:
@@ -70,19 +71,22 @@ def test_ner_scores_are_floats(self, client: httpx.Client) -> None:
 class TestPatternEntities:
     """Regex pattern recognizer entities."""
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_phone_mobile(self, client: httpx.Client) -> None:
         types = _detected_types(
             client, "Bel ons op 0612345678.", ["PHONE_NUMBER"]
         )
         assert "PHONE_NUMBER" in types
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_phone_international_0031(self, client: httpx.Client) -> None:
         # +31 after a space doesn't trigger \b (both non-word chars) — use 0031 form.
         types = _detected_types(
             client, "Bereikbaar op 0031612345678.", ["PHONE_NUMBER"]
         )
         assert "PHONE_NUMBER" in types
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_email_detected(self, client: httpx.Client) -> None:
         texts = _detected_texts(
             client,
@@ -91,28 +95,33 @@ def test_email_detected(self, client: httpx.Client) -> None:
         )
         assert "support@example.nl" in texts
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_iban_nl(self, client: httpx.Client) -> None:
         texts = _detected_texts(
             client, "IBAN: NL91ABNA0417164300.", ["IBAN"]
         )
         assert "NL91ABNA0417164300" in texts
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_iban_nl_spaced(self, client: httpx.Client) -> None:
         texts = _detected_texts(
             client, "Rekeningnummer: NL91 ABNA 0417 1643 00.", ["IBAN"]
         )
         assert any("NL91" in t for t in texts)
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_bsn_nine_digits(self, client: httpx.Client) -> None:
         types = _detected_types(client, "BSN: 111222333.", ["BSN"])
         assert "BSN" in types
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_date_dd_mm_yyyy(self, client: httpx.Client) -> None:
         types = _detected_types(
             client, "Geboortedatum: 15-03-1990.", ["DATE_TIME"]
         )
         assert "DATE_TIME" in types
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_date_spelled_out(self, client: httpx.Client) -> None:
         types = _detected_types(
             client,
@@ -121,30 +130,35 @@ def test_date_spelled_out(self, client: httpx.Client) -> None:
         )
         assert "DATE_TIME" in types
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_id_no_passport(self, client: httpx.Client) -> None:
         types = _detected_types(
             client, "Paspoortnummer: AB1234561.", ["ID_NO"]
         )
         assert "ID_NO" in types
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_vat_number(self, client: httpx.Client) -> None:
         texts = _detected_texts(
             client, "BTW-nummer: NL123456789B01.", ["VAT_NUMBER"]
         )
         assert "NL123456789B01" in texts
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_license_plate(self, client: httpx.Client) -> None:
         types = _detected_types(
             client, "Kenteken AB-12-CD werd gesignaleerd.", ["LICENSE_PLATE"]
         )
         assert "LICENSE_PLATE" in types
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_ip_address(self, client: httpx.Client) -> None:
         texts = _detected_texts(
             client, "Verbonden via IP 192.168.1.1.", ["IP_ADDRESS"]
         )
         assert "192.168.1.1" in texts
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_case_no_z_format(self, client: httpx.Client) -> None:
         types = _detected_types(
             client,
@@ -153,18 +167,21 @@ def test_case_no_z_format(self, client: httpx.Client) -> None:
         )
         assert "CASE_NO" in types
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_case_no_awb_format(self, client: httpx.Client) -> None:
         types = _detected_types(
             client, "Bezwaar AWB 21/12345 ingediend.", ["CASE_NO"]
         )
         assert "CASE_NO" in types
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_drivers_license(self, client: httpx.Client) -> None:
         types = _detected_types(
             client, "Rijbewijsnummer: 1234567890.", ["DRIVERS_LICENSE"]
         )
         assert "DRIVERS_LICENSE" in types
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_kvk_number(self, client: httpx.Client) -> None:
         types = _detected_types(
             client, "KvK-nummer: 12345678.", ["KVK_NUMBER"]
@@ -175,6 +192,7 @@ def test_kvk_number(self, client: httpx.Client) -> None:
 class TestEmailNotOrganization:
     """Regression: SpaCy NER must not tag emails as ORGANIZATION."""
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_no_organization_overlapping_email(self, client: httpx.Client) -> None:
         r = client.post(
             "/api/v1/analyze",
@@ -202,6 +220,7 @@ def test_no_organization_overlapping_email(self, client: httpx.Client) -> None:
                     "SpaCy NER false positive should be filtered"
                 )
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_email_entity_is_detected(self, client: httpx.Client) -> None:
         """The email itself must still be detected as EMAIL."""
         r = client.post(
@@ -226,6 +245,7 @@ class TestPhoneVsDriversLicense:
     so both are returned. This is expected behavior.
     """
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_phone_number_is_detected(self, client: httpx.Client) -> None:
         r = client.post(
             "/api/v1/analyze",
@@ -239,6 +259,7 @@ def test_phone_number_is_detected(self, client: httpx.Client) -> None:
         types = {e["entity_type"] for e in r.json()["pii_entities"]}
         assert "PHONE_NUMBER" in types
 
+    @pytest.mark.skip(reason="Pattern recognizers disabled in plugins.yaml")
     def test_phone_has_higher_score_than_drivers_license(
         self, client: httpx.Client
     ) -> None: