fix(local): read the model manifests as UTF-8 on Windows

surfsense_local/backend/modules/llm/catalog/local/manifest/loader.py

@@ -8,7 +8,9 @@
 
 
 def load_local_manifest(path: Path | None = None) -> LocalManifest:
-    return LocalManifest.model_validate_json((path or MANIFEST_PATH).read_text())
+    return LocalManifest.model_validate_json(
+        (path or MANIFEST_PATH).read_text(encoding="utf-8")
+    )
 
 
 def empty_manifest() -> LocalManifest:

surfsense_local/backend/modules/llm/catalog/remote/manifest/loader.py

@@ -14,7 +14,9 @@
 
 
 def load_remote_manifest(path: Path | None = None) -> RemoteManifest:
-    return RemoteManifest.model_validate_json((path or MANIFEST).read_text())
+    return RemoteManifest.model_validate_json(
+        (path or MANIFEST).read_text(encoding="utf-8")
+    )
 
 
 @lru_cache

surfsense_local/backend/tests/unit/llm/catalog/remote/test_shipped_manifest.py

@@ -1,5 +1,8 @@
 """The committed manifest is the one every remote label comes from."""
 
+import subprocess
+import sys
+
 import pytest
 
 from modules.llm.catalog.remote.manifest.loader import (
@@ -19,6 +22,31 @@ def test_the_shipped_manifest_loads() -> None:
     assert "openrouter" in manifest.providers
 
 
+def test_the_shipped_manifest_loads_without_the_system_encoding() -> None:
+    """Windows reads cp1252 by default, and the manifest's UTF-8 left it empty."""
+    load = (
+        "from modules.llm.catalog.remote.manifest.loader import load_remote_manifest;"
+        "load_remote_manifest()"
+    )
+    run = subprocess.run(
+        [
+            sys.executable,
+            "-X",
+            "warn_default_encoding",
+            "-W",
+            "error::EncodingWarning",
+            "-c",
+            load,
+        ],
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+        check=False,
+    )
+
+    assert run.returncode == 0, run.stderr
+
+
 def test_the_shipped_manifest_knows_an_embedder_is_not_a_chat_model() -> None:
     """The old snapshot labelled text-embedding-3-small a chat model."""
     found = remote_lookup().classify("text-embedding-3-small", provider="openai")