fix(setup): filter LM Studio models by state == "loaded_instance"

LM Studio's /v1/models returns all downloaded models, not just loaded
ones. Use /api/v0/models with state filtering in both fetch_loaded_models()
and _fetch_local_models() so only RAM-resident models are shown as loaded.
This also restores the _choose_model() fallback that offers downloaded-but-
unloaded models when nothing is active in LM Studio.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/pyra/setup/wizard.py

@@ -386,7 +386,14 @@ def fetch_loaded_models(provider: Provider) -> list[str]:
             resp = httpx.get(f"{provider.base_url}/api/ps", timeout=3.0)
             resp.raise_for_status()
             return [m["name"] for m in resp.json().get("models", [])]
-        else:
+        elif provider.id == "lmstudio":
+            resp = httpx.get("http://localhost:1234/api/v0/models", timeout=3.0)
+            resp.raise_for_status()
+            return [
+                m["id"] for m in resp.json().get("data", [])
+                if m.get("state") == "loaded_instance"
+            ]
+        else:  # llamacpp — /models returns only the active loaded model
             resp = httpx.get(f"{provider.base_url}/models", timeout=3.0)
             resp.raise_for_status()
             return [m["id"] for m in resp.json().get("data", [])]
@@ -407,7 +414,7 @@ def _show_local_model_status(provider: Provider) -> None:
 
 
 def _fetch_local_models(provider: Provider) -> list[str]:
-    """Return currently loaded/available models from a local provider's API."""
+    """Return currently loaded models from a local provider's API."""
     if not provider.base_url:
         return []
     try:
@@ -415,6 +422,13 @@ def _fetch_local_models(provider: Provider) -> list[str]:
             resp = httpx.get(f"{provider.base_url}/api/tags", timeout=3.0)
             resp.raise_for_status()
             return [m["name"] for m in resp.json().get("models", [])]
+        elif provider.id == "lmstudio":
+            resp = httpx.get("http://localhost:1234/api/v0/models", timeout=3.0)
+            resp.raise_for_status()
+            return [
+                m["id"] for m in resp.json().get("data", [])
+                if m.get("state") == "loaded_instance"
+            ]
         else:
             resp = httpx.get(f"{provider.base_url}/models", timeout=3.0)
             resp.raise_for_status()

tests/unit/test_setup_wizard.py

@@ -96,14 +96,34 @@ def test_suggest_plugins_multiple_categories(monkeypatch):
 
 # ── _fetch_local_models ────────────────────────────────────────────────────────
 
-def test_fetch_local_models_lmstudio_returns_model_ids(monkeypatch):
+def test_fetch_local_models_lmstudio_returns_loaded_model_ids(monkeypatch):
     import pyra.setup.wizard as wiz
     mock_resp = MagicMock()
-    mock_resp.json.return_value = {"data": [{"id": "gemma-4b"}, {"id": "llama3"}]}
+    mock_resp.json.return_value = {
+        "data": [
+            {"id": "gemma-4b", "state": "loaded_instance"},
+            {"id": "llama3", "state": "not_loaded"},
+        ]
+    }
     mock_resp.raise_for_status = lambda: None
     monkeypatch.setattr(wiz.httpx, "get", lambda *a, **kw: mock_resp)
     from pyra.setup.providers import get_provider
-    assert wiz._fetch_local_models(get_provider("lmstudio")) == ["gemma-4b", "llama3"]
+    assert wiz._fetch_local_models(get_provider("lmstudio")) == ["gemma-4b"]
+
+
+def test_fetch_local_models_lmstudio_filters_unloaded(monkeypatch):
+    import pyra.setup.wizard as wiz
+    mock_resp = MagicMock()
+    mock_resp.json.return_value = {
+        "data": [
+            {"id": "model-a", "state": "not_loaded"},
+            {"id": "model-b", "state": "not_loaded"},
+        ]
+    }
+    mock_resp.raise_for_status = lambda: None
+    monkeypatch.setattr(wiz.httpx, "get", lambda *a, **kw: mock_resp)
+    from pyra.setup.providers import get_provider
+    assert wiz._fetch_local_models(get_provider("lmstudio")) == []
 
 
 def test_fetch_local_models_ollama_returns_model_names(monkeypatch):
@@ -392,17 +412,37 @@ def test_fetch_loaded_models_ollama_uses_api_ps(monkeypatch):
     assert any("/api/ps" in u for u in calls)
 
 
-def test_fetch_loaded_models_lmstudio_uses_models_endpoint(monkeypatch):
+def test_fetch_loaded_models_lmstudio_uses_beta_api_and_filters(monkeypatch):
     import pyra.setup.wizard as wiz
     from pyra.setup.providers import get_provider
     mock_resp = MagicMock()
-    mock_resp.json.return_value = {"data": [{"id": "gemma-4b"}]}
+    mock_resp.json.return_value = {
+        "data": [
+            {"id": "gemma-4b", "state": "loaded_instance"},
+            {"id": "llama3", "state": "not_loaded"},
+        ]
+    }
     mock_resp.raise_for_status = lambda: None
     calls = []
     monkeypatch.setattr(wiz.httpx, "get", lambda url, **kw: (calls.append(url), mock_resp)[1])
     result = wiz.fetch_loaded_models(get_provider("lmstudio"))
     assert result == ["gemma-4b"]
-    assert any("/models" in u for u in calls)
+    assert any("/api/v0/models" in u for u in calls)
+
+
+def test_fetch_loaded_models_lmstudio_filters_unloaded(monkeypatch):
+    import pyra.setup.wizard as wiz
+    from pyra.setup.providers import get_provider
+    mock_resp = MagicMock()
+    mock_resp.json.return_value = {
+        "data": [
+            {"id": "model-a", "state": "not_loaded"},
+            {"id": "model-b", "state": "not_loaded"},
+        ]
+    }
+    mock_resp.raise_for_status = lambda: None
+    monkeypatch.setattr(wiz.httpx, "get", lambda url, **kw: mock_resp)
+    assert wiz.fetch_loaded_models(get_provider("lmstudio")) == []
 
 
 def test_fetch_loaded_models_returns_empty_on_error(monkeypatch):