Add LM Studio model switch endpoint

Author: Tooru

gui/model-source.js

@@ -19,6 +19,42 @@
     let openrouterProviderBackup = providerInput?.value || '';
     let openrouterMaxTokensBackup = maxTokensInput?.value || '';
     let inFlight = null;
+    let activeLmStudioModel = '';
+
+    async function switchLmStudioModel(modelId) {
+      if (!modelId) return;
+
+      lmstudioModelSelect.disabled = true;
+      setLmStudioNote(`Loading '${modelId}' in LM Studio…`);
+
+      try {
+        const response = await fetch('/models/lmstudio/switch', {
+          method: 'POST',
+          headers: { 'Content-Type': 'application/json' },
+          body: JSON.stringify({ model_id: modelId })
+        });
+
+        if (!response.ok) {
+          let detail = '';
+          try {
+            const payload = await response.json();
+            detail = payload?.detail || '';
+          } catch {
+            // ignore
+          }
+          throw new Error(detail || response.statusText || 'Unable to switch LM Studio model');
+        }
+
+        activeLmStudioModel = modelId;
+      } catch (error) {
+        setLmStudioNote(error?.message || 'Unable to switch LM Studio model.');
+        return;
+      } finally {
+        lmstudioModelSelect.disabled = false;
+      }
+
+      applyLmStudioModelSelection();
+    }
 
     function getSource() {
       return sourceSelect.value === 'lmstudio' ? 'lmstudio' : 'openrouter';
@@ -104,6 +140,7 @@
 
           lmstudioModelSelect.disabled = false;
           applyLmStudioModelSelection();
+          activeLmStudioModel = lmstudioModelSelect.value;
         })
         .catch((error) => {
           lmstudioModelSelect.innerHTML = '';
@@ -120,6 +157,20 @@
       return inFlight;
     }
 
+    async function handleLmStudioModelChange() {
+      const selected = lmstudioModelSelect.value;
+      applyLmStudioModelSelection();
+
+      if (getSource() !== 'lmstudio') return;
+      if (!selected) {
+        activeLmStudioModel = '';
+        return;
+      }
+
+      if (selected === activeLmStudioModel) return;
+      await switchLmStudioModel(selected);
+    }
+
     function applyModelSourceUI() {
       const isLmstudio = getSource() === 'lmstudio';
       openrouterFields.hidden = isLmstudio;
@@ -166,7 +217,7 @@
 
     sourceSelect.addEventListener('change', applyModelSourceUI);
     sourceSelect.addEventListener('input', applyModelSourceUI);
-    lmstudioModelSelect.addEventListener('change', applyLmStudioModelSelection);
+    lmstudioModelSelect.addEventListener('change', handleLmStudioModelChange);
     applyModelSourceUI();
   }
 

server/routes/router.py

@@ -3,9 +3,13 @@
 import asyncio
 import datetime as dt
 import json
+import re
+import shutil
+import subprocess
 from pathlib import Path
 from typing import Any
 from urllib.error import HTTPError, URLError
+from urllib.parse import urlparse
 from urllib.request import Request, urlopen
 
 import structlog
@@ -368,6 +372,108 @@ def list_lmstudio_models() -> dict[str, Any]:
     }
 
 
+_LMSTUDIO_MODEL_ID_PATTERN = re.compile(r"[A-Za-z0-9][A-Za-z0-9._/@:+-]*\Z")
+
+
+def _resolve_lms_path() -> str | None:
+    resolved = shutil.which("lms")
+    if resolved:
+        return resolved
+    fallback = Path.home() / ".lmstudio" / "bin" / "lms"
+    if fallback.exists():
+        return str(fallback)
+    return None
+
+
+def _lmstudio_cli_instance_args(base_url: str) -> list[str]:
+    trimmed = (base_url or "").strip()
+    if not trimmed:
+        return []
+    if "://" not in trimmed:
+        trimmed = f"http://{trimmed}"
+    parsed = urlparse(trimmed)
+    host = parsed.hostname or "127.0.0.1"
+    port = parsed.port or 1234
+    return ["--host", host, "--port", str(port)]
+
+
+def _truncate_cli_output(value: str, *, limit: int = 2000) -> str:
+    cleaned = (value or "").strip()
+    if len(cleaned) > limit:
+        return f"{cleaned[:limit]}..."
+    return cleaned
+
+
+class LMStudioModelSwitchRequest(BaseModel):
+    model_id: str
+
+
+class LMStudioModelSwitchResponse(BaseModel):
+    model_id: str
+    message: str
+
+
+@router.post("/models/lmstudio/switch", response_model=LMStudioModelSwitchResponse, tags=["models"])
+def switch_lmstudio_model(
+    request: LMStudioModelSwitchRequest,
+    _: None = Depends(require_api_token),
+) -> LMStudioModelSwitchResponse:
+    model_id = (request.model_id or "").strip()
+    if not model_id:
+        raise HTTPException(status_code=422, detail="model_id is required")
+    if _LMSTUDIO_MODEL_ID_PATTERN.fullmatch(model_id) is None:
+        raise HTTPException(status_code=400, detail="Invalid model_id format")
+
+    base_url = (settings.lmstudio_base_url or "").strip()
+    if not base_url:
+        raise HTTPException(status_code=500, detail="LM Studio base URL is not configured")
+
+    lms_path = _resolve_lms_path()
+    if not lms_path:
+        raise HTTPException(
+            status_code=501,
+            detail="LM Studio CLI 'lms' was not found. Install LM Studio or add 'lms' to PATH to enable model switching.",
+        )
+
+    instance_args = _lmstudio_cli_instance_args(base_url)
+
+    try:
+        unload = subprocess.run(
+            [lms_path, "unload", "--all", *instance_args],
+            capture_output=True,
+            text=True,
+            timeout=30,
+        )
+    except subprocess.TimeoutExpired as exc:  # pragma: no cover
+        raise HTTPException(status_code=504, detail="Timed out unloading LM Studio models") from exc
+
+    if unload.returncode != 0:
+        detail = _truncate_cli_output(unload.stderr or unload.stdout)
+        raise HTTPException(
+            status_code=502,
+            detail=f"Unable to unload LM Studio models: {detail or 'unknown error'}",
+        )
+
+    try:
+        load = subprocess.run(
+            [lms_path, "load", model_id, "--exact", "-y", *instance_args],
+            capture_output=True,
+            text=True,
+            timeout=600,
+        )
+    except subprocess.TimeoutExpired as exc:  # pragma: no cover
+        raise HTTPException(status_code=504, detail=f"Timed out loading '{model_id}' in LM Studio") from exc
+
+    if load.returncode != 0:
+        detail = _truncate_cli_output(load.stderr or load.stdout)
+        raise HTTPException(
+            status_code=502,
+            detail=f"Unable to load '{model_id}' in LM Studio: {detail or 'unknown error'}",
+        )
+
+    return LMStudioModelSwitchResponse(model_id=model_id, message=f"Loaded '{model_id}' in LM Studio")
+
+
 class RetryApiErrorsResponse(BaseModel):
     retry_run_id: str
     original_run_id: str

tests/test_external_endpoint_config.py

@@ -327,6 +327,109 @@ def test_lmstudio_endpoint_fails_when_base_url_empty(monkeypatch) -> None:
     assert "not configured" in response.json()["detail"]
 
 
+def test_switch_lmstudio_model_invokes_lms_cli(monkeypatch) -> None:
+    import sys
+    from unittest.mock import MagicMock
+
+    from fastapi.testclient import TestClient
+
+    from server.api import create_app
+    from server.config import get_settings
+
+    get_settings.cache_clear()
+
+    router_module = sys.modules["server.routes.router"]
+
+    mock_settings = MagicMock()
+    mock_settings.lmstudio_base_url = "http://custom-lmstudio:7777/v1"
+    monkeypatch.setattr(router_module, "settings", mock_settings)
+
+    monkeypatch.setattr(router_module, "_resolve_lms_path", lambda: "/fake/lms")
+
+    calls: list[list[str]] = []
+
+    class FakeResult:
+        def __init__(self) -> None:
+            self.returncode = 0
+            self.stdout = ""
+            self.stderr = ""
+
+    def fake_run(args: list[str], *, capture_output: bool, text: bool, timeout: int) -> FakeResult:
+        assert capture_output is True
+        assert text is True
+        assert timeout > 0
+        calls.append(args)
+        return FakeResult()
+
+    monkeypatch.setattr(router_module.subprocess, "run", fake_run)
+
+    app = create_app()
+    client = TestClient(app, raise_server_exceptions=False)
+
+    response = client.post(
+        "/models/lmstudio/switch",
+        json={"model_id": "liquid/lfm2.5-1.2b"},
+    )
+
+    assert response.status_code == 200
+    payload = response.json()
+    assert payload["model_id"] == "liquid/lfm2.5-1.2b"
+
+    assert calls == [
+        [
+            "/fake/lms",
+            "unload",
+            "--all",
+            "--host",
+            "custom-lmstudio",
+            "--port",
+            "7777",
+        ],
+        [
+            "/fake/lms",
+            "load",
+            "liquid/lfm2.5-1.2b",
+            "--exact",
+            "-y",
+            "--host",
+            "custom-lmstudio",
+            "--port",
+            "7777",
+        ],
+    ]
+
+
+def test_switch_lmstudio_model_rejects_invalid_model_id(monkeypatch) -> None:
+    import sys
+    from unittest.mock import MagicMock
+
+    from fastapi.testclient import TestClient
+
+    from server.api import create_app
+    from server.config import get_settings
+
+    get_settings.cache_clear()
+
+    router_module = sys.modules["server.routes.router"]
+
+    mock_settings = MagicMock()
+    mock_settings.lmstudio_base_url = "http://custom-lmstudio:7777/v1"
+    monkeypatch.setattr(router_module, "settings", mock_settings)
+
+    monkeypatch.setattr(router_module, "_resolve_lms_path", lambda: "/fake/lms")
+
+    app = create_app()
+    client = TestClient(app, raise_server_exceptions=False)
+
+    response = client.post(
+        "/models/lmstudio/switch",
+        json={"model_id": "bad model"},
+    )
+
+    assert response.status_code == 400
+    assert "Invalid" in response.json()["detail"]
+
+
 # =============================================================================
 # Expert Questions LM Studio URL Configuration Tests
 # =============================================================================