fix(cookbook): avoid launching Ollama during Windows cache scan (#4368)

routes/cookbook_helpers.py

@@ -505,6 +505,8 @@ def _cached_model_scan_script(model_dirs: list[str] | None = None, add_hf_cache:
         "    if u.startswith('KB'): return int(n * 1024)",
         "    return int(n)",
         "def scan_ollama():",
+        "    if any(m.get('is_ollama') for m in models): return",
+        "    if os.name == 'nt' and not os.environ.get('ODYSSEUS_ALLOW_OLLAMA_CLI_SCAN'): return",
         "    if not shutil.which('ollama'): return",
         "    try:",
         "        p = subprocess.run(['ollama', 'list'], stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True, timeout=6)",
@@ -535,8 +537,8 @@ def _cached_model_scan_script(model_dirs: list[str] | None = None, add_hf_cache:
         "            models.append({'repo_id':name,'size_bytes':size_bytes,'nb_files':1,'has_incomplete':False,'path':'ollama','backend':'ollama','is_ollama':True})",
         "        return",
         "for _hf_cache in hf_cache_paths(): scan_hf(_hf_cache)",
-        "scan_ollama()",
         "scan_ollama_api()",
+        "scan_ollama()",
     ]
     for model_dir in model_dirs or []:
         lines.append(f"scan_dir(os.path.expanduser({model_dir!r}))")

tests/test_cookbook_helpers.py

@@ -786,6 +786,50 @@ def test_cached_model_scan_reports_plain_dir_gguf(tmp_path):
     assert ggufs[3]["quant"] == "BF16"
 
 
+def test_cached_model_scan_uses_ollama_api_before_cli_and_windows_opt_in():
+    script = _cached_model_scan_script()
+
+    assert "scan_ollama_api()\nscan_ollama()" in script
+    assert "if any(m.get('is_ollama') for m in models): return" in script
+    assert "os.name == 'nt'" in script
+    assert "ODYSSEUS_ALLOW_OLLAMA_CLI_SCAN" in script
+
+
+@pytest.mark.skipif(os.name != "nt", reason="Windows Ollama CLI startup guard")
+def test_cached_model_scan_does_not_launch_ollama_cli_on_windows(tmp_path):
+    """Official Ollama for Windows can auto-start the tray/server on `ollama list`.
+    The read-only cache scanner must not invoke that CLI unless explicitly opted in.
+    """
+    marker = tmp_path / "ollama-called.txt"
+    fake_ollama = tmp_path / "ollama.cmd"
+    fake_ollama.write_text(
+        "@echo off\r\n"
+        f'echo called>"{marker}"\r\n'
+        "echo NAME ID SIZE MODIFIED\r\n"
+        "echo local-model:latest abc 1 GB now\r\n",
+        encoding="utf-8",
+    )
+
+    empty_home = tmp_path / "home"
+    empty_home.mkdir()
+    scan_py = tmp_path / "scan_cache.py"
+    scan_py.write_text(_cached_model_scan_script(), encoding="utf-8")
+    env = dict(os.environ)
+    env["PATH"] = str(tmp_path) + os.pathsep + env.get("PATH", "")
+    env["HOME"] = str(empty_home)
+    env.pop("ODYSSEUS_ALLOW_OLLAMA_CLI_SCAN", None)
+    proc = subprocess.run(
+        [sys.executable, str(scan_py)],
+        check=True,
+        capture_output=True,
+        text=True,
+        env=env,
+    )
+
+    assert marker.exists() is False
+    assert all(m.get("backend") != "ollama" for m in json.loads(proc.stdout))
+
+
 def test_cached_model_scan_uses_huggingface_cache_env(tmp_path):
     """Docker recreates can leave the persisted HF cache outside HOME.
     The Serve scanner should honor the cache env path instead of only ~/.cache.