[LLM|Models] Ship the local model manifest in the Studio worker

docs/architecture/packaging.md

@@ -39,7 +39,7 @@ What else each spec names, and why the analyser cannot find it on its own:
 | `api.spec` | `onnxruntime` and `tokenizers` libraries | the query encoder's native libraries load from C |
 | `api.spec` | the local model manifest `catalog/local/manifest/models.json`, the remote model manifest `catalog/remote/manifest/models.json`, the chat prompts | read by path or through `importlib.resources` |
 | `api.spec` | excludes Docling, torch, torchvision, transformers, pandas, scipy and OpenCV | only the worker parses files, and the analyser cannot tell these are optional |
-| `worker.spec` | the remote model manifest | Studio classifies a remote model through the same discovery the API uses, and the file is read by path |
+| `worker.spec` | the local model manifest, the remote model manifest | Studio finds its chosen image and audio models through the local catalog, and classifies a remote model through the same discovery the API uses; both files are read by path |
 | `worker.spec` | Docling and its packages, RapidOCR, transformers, torchvision | lazy and native imports Docling reaches only on the first PDF |
 | `worker.spec` | python-docx, python-pptx, xlsxwriter, reportlab | the Office formats run model-written code that imports them, so no static import exists |
 | `worker.spec` | `modules.documents.tasks`, `modules.artifacts.tasks` | Huey resolves a task by its name |
@@ -109,8 +109,8 @@ All five carry the `packaging` marker, which `pyproject.toml` excludes by defaul
 | Test | Proves |
 |---|---|
 | `test_frozen_boot.py` | a minimal frozen entry, `sys.frozen` true, migrates a real database from the bundled revisions, loads `vec0` and round-trips a vector through `chunk_vectors` |
-| `test_real_binaries.py` | the real API binary passes its retrieval import check and answers `/health`; the real worker passes its vision import check and both queue consumers stay up; both binaries ship the remote manifest, and the frozen API serves a curated row |
-| `test_spec_data_files.py` | every literal `datas` path in the specs exists, so a renamed file cannot ship missing |
+| `test_real_binaries.py` | the real API binary passes its retrieval import check and answers `/health`; the real worker passes its vision import check and both queue consumers stay up; both binaries ship the remote manifest, the worker ships the local one, and the frozen API serves a curated row |
+| `test_spec_data_files.py` | every literal `datas` path in the specs exists, so a renamed file cannot ship missing, and both `api.spec` and `worker.spec` bundle the local model manifest |
 | `test_license_key.py` | the compiled license keys exclude the fixture key |
 | `test_audiocpp_voicing.py` | the staged audio.cpp server, started with the sidecar's flags from the `server.json` the app writes, voices two turns of each curated model through the app's adapter, at the sample rate its entry names, then holds no model loaded; it runs only when `SURFSENSE_TEST_AUDIO_MODELS` names a folder of the pinned files, each checked against its sha256, and then fails without a staged server; `build-audiocpp.yml` runs it on both builds |
 
@@ -124,4 +124,3 @@ Two run in CI: `test_license_key.py` inside the release workflow, and `test_audi
 - No release has built or packaged audio.cpp yet, and nothing has staged its macOS archive on a Mac.
 - No workflow runs the `surfsense_local` unit or integration tests on pull requests; only the two packaging tests above run in CI.
 - No test ingests a PDF with networking disabled.
-- `test_the_curated_manifest_is_one_of_them` fails on Windows: `literal_data_paths()` joins the spec's path with `Path`, which gives backslashes there, and the test looks for a forward-slash string.

surfsense_local/backend/bundling/worker.spec

@@ -29,6 +29,16 @@ datas.append(
     )
 )
 
+# Studio resolves its chosen image and audio models through the local catalog,
+# which knows a curated model only from this manifest; without it an installed,
+# chosen model reads as "not installed".
+datas.append(
+    (
+        str(BACKEND / "modules" / "llm" / "catalog" / "local" / "manifest" / "models.json"),
+        "modules/llm/catalog/local/manifest",
+    )
+)
+
 for package in (
     "onnxruntime",
     "docling",

surfsense_local/backend/tests/packaging/test_real_binaries.py

@@ -68,6 +68,23 @@ def _assert_capability_catalogue_shipped(binary: Path) -> None:
     assert json.loads(manifest.read_text())["providers"]
 
 
+def _assert_curated_manifest_shipped(binary: Path) -> None:
+    """Studio finds its chosen image and audio models through the curated list;
+    without it the frozen worker calls an installed model "not installed"."""
+    manifest = (
+        binary.parent
+        / "_internal"
+        / "modules"
+        / "llm"
+        / "catalog"
+        / "local"
+        / "manifest"
+        / "models.json"
+    )
+    assert manifest.is_file(), f"local model manifest missing from {binary.parent}"
+    assert json.loads(manifest.read_text())["models"]
+
+
 def _free_port() -> int:
     with socket.socket() as sock:
         sock.bind(("127.0.0.1", 0))
@@ -130,6 +147,7 @@ def test_worker_binary_starts(tmp_path: Path) -> None:
     """Freeze the worker and assert its lazy vision imports and consumer boot."""
     binary = _freeze("worker.spec", tmp_path)
     _assert_capability_catalogue_shipped(binary)
+    _assert_curated_manifest_shipped(binary)
     vision = subprocess.run(
         [str(binary), "--check-vision-runtime"],
         env=_env(tmp_path),

surfsense_local/backend/tests/packaging/test_spec_data_files.py

@@ -47,7 +47,8 @@ def literal_data_paths(spec: Path) -> list[str]:
                     parts.insert(0, str(current.right.value))
                 current = current.left
             if parts and getattr(current, "id", None) == "BACKEND":
-                found.append(str(Path(*parts)))
+                # Forward slashes on every platform, as the specs' targets are.
+                found.append(Path(*parts).as_posix())
     return found
 
 
@@ -59,9 +60,11 @@ def test_every_bundled_data_path_exists(spec: Path) -> None:
     assert missing == [], f"{spec.name} bundles paths that do not exist: {missing}"
 
 
-def test_the_curated_manifest_is_one_of_them() -> None:
-    """The file that was missed. Without it a frozen build has no curated list,
-    and the model screen is empty on the machine that most needs it."""
-    bundled = {rel for spec in SPECS for rel in literal_data_paths(spec)}
+@pytest.mark.parametrize("name", ["api.spec", "worker.spec"])
+def test_every_binary_that_reads_the_curated_manifest_ships_it(name: str) -> None:
+    """Without it the catalog falls back to an empty list: the API shows no
+    curated models, and the Studio worker calls an installed, chosen image or
+    audio model "not installed"."""
+    bundled = literal_data_paths(BACKEND / "bundling" / name)
 
     assert "modules/llm/catalog/local/manifest/models.json" in bundled