update sdui, new performance tab, start patching llama.cpp UI, lock openwebui to 0.6.41, add a new "clar data" button for openwebui select, new smartcache + pipelineparallel UI performance options

package.json

@@ -1,7 +1,7 @@
 {
   "name": "gerbil",
   "productName": "Gerbil",
-  "version": "1.16.4",
+  "version": "1.17.0",
   "description": "Run Large Language Models locally",
   "main": "out/main/index.js",
   "homepage": "./",
@@ -41,7 +41,7 @@
   "dependencies": {
     "@codemirror/search": "^6.5.11",
     "@codemirror/theme-one-dark": "^6.1.3",
-    "@codemirror/view": "^6.39.4",
+    "@codemirror/view": "^6.39.5",
     "@fontsource/inter": "^5.2.8",
     "@huggingface/gguf": "^0.3.2",
     "@mantine/core": "^8.3.10",
@@ -49,7 +49,7 @@
     "@uiw/react-codemirror": "^4.25.4",
     "electron-updater": "^6.6.2",
     "execa": "^9.6.1",
-    "lucide-react": "^0.561.0",
+    "lucide-react": "^0.562.0",
     "mime-types": "^3.0.2",
     "react": "^19.2.3",
     "react-dom": "^19.2.3",
@@ -115,6 +115,10 @@
       {
         "from": "assets/kcpp_sdui.embd",
         "to": "assets/kcpp_sdui.embd"
+      },
+      {
+        "from": "assets/lcpp.gz.embd",
+        "to": "assets/lcpp.gz.embd"
       }
     ],
     "extraResources": [

src/components/screens/Launch/AdvancedTab.tsx

@@ -3,7 +3,6 @@ import {
   Group,
   Text,
   TextInput,
-  NumberInput,
   Button,
   SimpleGrid,
   ActionIcon,
@@ -19,29 +18,14 @@ export const AdvancedTab = () => {
   const {
     additionalArguments,
     preLaunchCommands,
-    noshift,
-    flashattention,
     noavx2,
     failsafe,
-    lowvram,
-    quantmatmul,
-    usemmap,
     debugmode,
-    acceleration,
-    moecpu,
-    moeexperts,
     setAdditionalArguments,
     setPreLaunchCommands,
-    setNoshift,
-    setFlashattention,
     setNoavx2,
     setFailsafe,
-    setLowvram,
-    setQuantmatmul,
-    setUsemmap,
     setDebugmode,
-    setMoecpu,
-    setMoeexperts,
   } = useLaunchConfigStore();
   const [commandLineModalOpen, setCommandLineModalOpen] = useState(false);
   const [backendSupport, setBackendSupport] = useState<{
@@ -58,8 +42,6 @@ export const AdvancedTab = () => {
     setAdditionalArguments(updatedArgs);
   };
 
-  const isGpuAcceleration = acceleration === 'cuda' || acceleration === 'rocm';
-
   useEffect(() => {
     const detectAccelerationSupport = async () => {
       const support =
@@ -84,20 +66,6 @@ export const AdvancedTab = () => {
     <Stack gap="md">
       <div>
         <SimpleGrid cols={3} spacing="lg" verticalSpacing="md">
-          <CheckboxWithTooltip
-            checked={!noshift}
-            onChange={(checked) => setNoshift(!checked)}
-            label="Context Shift"
-            tooltip="Use Context Shifting to reduce reprocessing."
-          />
-
-          <CheckboxWithTooltip
-            checked={noshift}
-            onChange={setNoshift}
-            label="No Shift"
-            tooltip="Don't use GPU layer shifting for incomplete offloads, which may reduce model performance."
-          />
-
           <CheckboxWithTooltip
             checked={noavx2}
             onChange={setNoavx2}
@@ -110,25 +78,6 @@ export const AdvancedTab = () => {
             disabled={isLoading || !backendSupport?.noavx2}
           />
 
-          <CheckboxWithTooltip
-            checked={usemmap}
-            onChange={setUsemmap}
-            label="MMAP"
-            tooltip="Use MMAP to load models when enabled."
-          />
-
-          <CheckboxWithTooltip
-            checked={quantmatmul && isGpuAcceleration}
-            onChange={setQuantmatmul}
-            label="QuantMatMul"
-            tooltip={
-              !isGpuAcceleration
-                ? 'QuantMatMul is only available for CUDA and ROCm accelerations.'
-                : 'Enable MMQ mode to use finetuned kernels instead of default CuBLAS/HipBLAS for prompt processing.'
-            }
-            disabled={!isGpuAcceleration}
-          />
-
           <CheckboxWithTooltip
             checked={failsafe}
             onChange={setFailsafe}
@@ -141,25 +90,6 @@ export const AdvancedTab = () => {
             disabled={isLoading || !backendSupport?.failsafe}
           />
 
-          <CheckboxWithTooltip
-            checked={flashattention}
-            onChange={setFlashattention}
-            label="Flash Attention"
-            tooltip="Enable flash attention to reduce memory usage. May produce incorrect answers for some prompts, but improves performance."
-          />
-
-          <CheckboxWithTooltip
-            checked={lowvram && isGpuAcceleration}
-            onChange={setLowvram}
-            label="Low VRAM"
-            tooltip={
-              !isGpuAcceleration
-                ? 'Low VRAM mode is only available for CUDA and ROCm accelerations.'
-                : 'Avoid offloading KV Cache or scratch buffers to VRAM. Allows more layers to fit, but may result in a speed loss.'
-            }
-            disabled={!isGpuAcceleration}
-          />
-
           <CheckboxWithTooltip
             checked={debugmode}
             onChange={setDebugmode}
@@ -169,46 +99,6 @@ export const AdvancedTab = () => {
         </SimpleGrid>
       </div>
 
-      <div>
-        <Stack gap="md">
-          <Group gap="lg" align="flex-start" wrap="nowrap">
-            <div style={{ flex: 1, minWidth: 200 }}>
-              <Group gap="xs" align="center" mb="xs">
-                <Text size="sm" fw={500}>
-                  MoE Experts
-                </Text>
-                <InfoTooltip label="How many experts to use for MoE models. Set to -1 to follow GGUF metadata (default), or specify a specific number of experts." />
-              </Group>
-              <NumberInput
-                value={moeexperts}
-                onChange={(value) => setMoeexperts(Number(value))}
-                min={-1}
-                max={128}
-                step={1}
-                size="sm"
-              />
-            </div>
-
-            <div style={{ flex: 1, minWidth: 200 }}>
-              <Group gap="xs" align="center" mb="xs">
-                <Text size="sm" fw={500}>
-                  MoE CPU Layers
-                </Text>
-                <InfoTooltip label="Keep the Mixture of Experts (MoE) weights of the first N layers in the CPU. Set to 0 to disable (default), or specify the number of layers to keep on CPU." />
-              </Group>
-              <NumberInput
-                value={moecpu}
-                onChange={(value) => setMoecpu(Number(value) || 0)}
-                min={0}
-                max={999}
-                step={1}
-                size="sm"
-              />
-            </div>
-          </Group>
-        </Stack>
-      </div>
-
       <div>
         <Group mb="xs" justify="space-between">
           <Group>

src/components/screens/Launch/CommandLineArgumentsModal.tsx

@@ -67,6 +67,8 @@ const UI_COVERED_ARGS = new Set([
   '--tensor_split',
   '--debugmode',
   '--lowvram',
+  '--smartcache',
+  '--pipelineparallel',
 ] as const) as ReadonlySet<string>;
 
 const IGNORED_ARGS = new Set([
@@ -88,6 +90,7 @@ const IGNORED_ARGS = new Set([
   '--no-mmap',
   '--sdnotile',
   '--testmemory',
+  '--forceversion',
 ] as const) as ReadonlySet<string>;
 
 const COMMAND_LINE_ARGUMENTS = [
@@ -377,7 +380,21 @@ const COMMAND_LINE_ARGUMENTS = [
     description:
       'How many tokens to generate by default, if not specified. Must be smaller than context size. Usually, your frontend GUI will override this.',
     type: 'int',
-    default: 768,
+    default: 896,
+    category: 'Performance',
+  },
+  {
+    flag: '--smartcache',
+    description:
+      'Enables intelligent context switching by saving KV cache snapshots to RAM. Requires fast forwarding.',
+    type: 'boolean',
+    category: 'Performance',
+  },
+  {
+    flag: '--pipelineparallel',
+    description:
+      'Enable Pipeline Parallelism for faster multigpu speeds but using more memory, only active for multigpu.',
+    type: 'boolean',
     category: 'Performance',
   },
   {
@@ -433,15 +450,6 @@ const COMMAND_LINE_ARGUMENTS = [
     type: 'boolean',
     category: 'Advanced',
   },
-  {
-    flag: '--forceversion',
-    description:
-      'If the model file format detection fails (e.g. rogue modified model) you can set this to override the detected format (enter desired version, e.g. 401 for GPTNeoX-Type2).',
-    metavar: '[version]',
-    type: 'int',
-    default: 0,
-    category: 'Advanced',
-  },
   {
     flag: '--smartcontext',
     description:

src/components/screens/Launch/PerformanceTab.tsx

@@ -0,0 +1,153 @@
+import { Stack, Group, Text, NumberInput, SimpleGrid } from '@mantine/core';
+import { InfoTooltip } from '@/components/InfoTooltip';
+import { CheckboxWithTooltip } from '@/components/CheckboxWithTooltip';
+import { useLaunchConfigStore } from '@/stores/launchConfig';
+
+export const PerformanceTab = () => {
+  const {
+    noshift,
+    flashattention,
+    lowvram,
+    quantmatmul,
+    usemmap,
+    acceleration,
+    moecpu,
+    moeexperts,
+    smartcache,
+    pipelineparallel,
+    setNoshift,
+    setFlashattention,
+    setLowvram,
+    setQuantmatmul,
+    setUsemmap,
+    setMoecpu,
+    setMoeexperts,
+    setSmartcache,
+    setPipelineparallel,
+  } = useLaunchConfigStore();
+
+  const isGpuAcceleration = acceleration === 'cuda' || acceleration === 'rocm';
+
+  return (
+    <Stack gap="md">
+      <div>
+        <SimpleGrid cols={3} spacing="lg" verticalSpacing="md">
+          <CheckboxWithTooltip
+            checked={!noshift}
+            onChange={(checked) => setNoshift(!checked)}
+            label="Context Shift"
+            tooltip="Use Context Shifting to reduce reprocessing and improve performance with long contexts."
+          />
+
+          <CheckboxWithTooltip
+            checked={noshift}
+            onChange={setNoshift}
+            label="No Shift"
+            tooltip="Disable context shifting. May reduce performance but can help with compatibility issues."
+          />
+
+          <CheckboxWithTooltip
+            checked={smartcache}
+            onChange={setSmartcache}
+            label="Smart Cache"
+            tooltip="Enables intelligent context switching by saving KV cache snapshots to RAM. Requires fast forwarding."
+          />
+        </SimpleGrid>
+      </div>
+
+      <div>
+        <SimpleGrid cols={3} spacing="lg" verticalSpacing="md">
+          <CheckboxWithTooltip
+            checked={flashattention}
+            onChange={setFlashattention}
+            label="Flash Attention"
+            tooltip="Enable flash attention to reduce memory usage and improve performance. May produce incorrect answers for some prompts."
+          />
+
+          <CheckboxWithTooltip
+            checked={lowvram && isGpuAcceleration}
+            onChange={setLowvram}
+            label="Low VRAM"
+            tooltip={
+              !isGpuAcceleration
+                ? 'Low VRAM mode is only available for CUDA and ROCm accelerations.'
+                : 'Avoid offloading KV Cache or scratch buffers to VRAM. Allows more layers to fit, but may result in a speed loss.'
+            }
+            disabled={!isGpuAcceleration}
+          />
+
+          <CheckboxWithTooltip
+            checked={quantmatmul && isGpuAcceleration}
+            onChange={setQuantmatmul}
+            label="QuantMatMul"
+            tooltip={
+              !isGpuAcceleration
+                ? 'QuantMatMul is only available for CUDA and ROCm accelerations.'
+                : 'Enable MMQ mode to use finetuned kernels instead of default CuBLAS/HipBLAS for prompt processing.'
+            }
+            disabled={!isGpuAcceleration}
+          />
+
+          <CheckboxWithTooltip
+            checked={pipelineparallel && isGpuAcceleration}
+            onChange={setPipelineparallel}
+            label="Pipeline Parallel"
+            tooltip={
+              !isGpuAcceleration
+                ? 'Pipeline Parallelism is only available for multi-GPU setups.'
+                : 'Enable Pipeline Parallelism for faster multi-GPU speeds but using more memory. Only active for multi-GPU setups.'
+            }
+            disabled={!isGpuAcceleration}
+          />
+
+          <CheckboxWithTooltip
+            checked={usemmap}
+            onChange={setUsemmap}
+            label="MMAP"
+            tooltip="Use memory-mapped file I/O for faster model loading. Recommended for most systems."
+          />
+        </SimpleGrid>
+      </div>
+
+      <div>
+        <Stack gap="md">
+          <Group gap="lg" align="flex-start" wrap="nowrap">
+            <div style={{ flex: 1, minWidth: 200 }}>
+              <Group gap="xs" align="center" mb="xs">
+                <Text size="sm" fw={500}>
+                  MoE Experts
+                </Text>
+                <InfoTooltip label="How many experts to use for MoE models. Set to -1 to follow GGUF metadata (default), or specify a specific number of experts." />
+              </Group>
+              <NumberInput
+                value={moeexperts}
+                onChange={(value) => setMoeexperts(Number(value))}
+                min={-1}
+                max={128}
+                step={1}
+                size="sm"
+              />
+            </div>
+
+            <div style={{ flex: 1, minWidth: 200 }}>
+              <Group gap="xs" align="center" mb="xs">
+                <Text size="sm" fw={500}>
+                  MoE CPU Layers
+                </Text>
+                <InfoTooltip label="Keep the Mixture of Experts (MoE) weights of the first N layers in the CPU. Set to 0 to disable (default), or specify the number of layers to keep on CPU." />
+              </Group>
+              <NumberInput
+                value={moecpu}
+                onChange={(value) => setMoecpu(Number(value) || 0)}
+                min={0}
+                max={999}
+                step={1}
+                size="sm"
+              />
+            </div>
+          </Group>
+        </Stack>
+      </div>
+    </Stack>
+  );
+};

src/components/screens/Launch/index.tsx

@@ -6,6 +6,7 @@ import { useLaunchLogic } from '@/hooks/useLaunchLogic';
 import { useWarnings } from '@/hooks/useWarnings';
 import { GeneralTab } from '@/components/screens/Launch/GeneralTab/index';
 import { AdvancedTab } from '@/components/screens/Launch/AdvancedTab';
+import { PerformanceTab } from '@/components/screens/Launch/PerformanceTab';
 import { NetworkTab } from '@/components/screens/Launch/NetworkTab';
 import { ImageGenerationTab } from '@/components/screens/Launch/ImageGenerationTab';
 import { WarningDisplay } from '@/components/WarningDisplay';
@@ -63,6 +64,8 @@ export const LaunchScreen = ({ onLaunch }: LaunchScreenProps) => {
     sdclipgpu,
     moecpu,
     moeexperts,
+    smartcache,
+    pipelineparallel,
     parseAndApplyConfigFile,
     loadConfigFromFile,
     setModel,
@@ -177,6 +180,8 @@ export const LaunchScreen = ({ onLaunch }: LaunchScreenProps) => {
     debugmode,
     moecpu,
     moeexperts,
+    smartcache,
+    pipelineparallel,
     usecuda: acceleration === 'cuda' || acceleration === 'rocm',
     usevulkan: acceleration === 'vulkan',
     useclblast: acceleration === 'clblast',
@@ -316,6 +321,8 @@ export const LaunchScreen = ({ onLaunch }: LaunchScreenProps) => {
       sdclipgpu,
       moecpu,
       moeexperts,
+      smartcache,
+      pipelineparallel,
     });
   }, [
     handleLaunch,
@@ -354,6 +361,8 @@ export const LaunchScreen = ({ onLaunch }: LaunchScreenProps) => {
     sdclipgpu,
     moecpu,
     moeexperts,
+    smartcache,
+    pipelineparallel,
   ]);
 
   return (
@@ -397,6 +406,7 @@ export const LaunchScreen = ({ onLaunch }: LaunchScreenProps) => {
               <Tabs.List>
                 <Tabs.Tab value="general">General</Tabs.Tab>
                 <Tabs.Tab value="image">Image Generation</Tabs.Tab>
+                <Tabs.Tab value="performance">Performance</Tabs.Tab>
                 <Tabs.Tab value="network">Network</Tabs.Tab>
                 <Tabs.Tab value="advanced">Advanced</Tabs.Tab>
               </Tabs.List>
@@ -405,16 +415,20 @@ export const LaunchScreen = ({ onLaunch }: LaunchScreenProps) => {
                 <GeneralTab configLoaded={configLoaded} />
               </Tabs.Panel>
 
-              <Tabs.Panel value="advanced">
+              <Tabs.Panel value="image">
-                <AdvancedTab />
+                <ImageGenerationTab />
+              </Tabs.Panel>
+
+              <Tabs.Panel value="performance">
+                <PerformanceTab />
               </Tabs.Panel>
 
               <Tabs.Panel value="network">
                 <NetworkTab />
               </Tabs.Panel>
 
-              <Tabs.Panel value="image">
+              <Tabs.Panel value="advanced">
-                <ImageGenerationTab />
+                <AdvancedTab />
               </Tabs.Panel>
             </Tabs>
 

src/components/settings/FrontendInterfaceSelector.tsx

@@ -1,5 +1,5 @@
 import { useState, useEffect, useCallback, useMemo } from 'react';
-import { Text, Box, Anchor, rem } from '@mantine/core';
+import { Text, Box, Anchor, rem, Button, Group, Stack } from '@mantine/core';
 import { Monitor, Image } from 'lucide-react';
 import { usePreferencesStore } from '@/stores/preferences';
 import type {
@@ -8,6 +8,7 @@ import type {
 } from '@/types';
 import { FRONTENDS } from '@/constants';
 import { Select } from '@/components/Select';
+import { Modal } from '@/components/Modal';
 
 interface FrontendRequirement {
   id: string;
@@ -40,6 +41,8 @@ export const FrontendInterfaceSelector = ({
     Map<string, boolean>
   >(new Map());
 
+  const [showClearDataModal, setShowClearDataModal] = useState(false);
+
   const frontendConfigs: FrontendConfig[] = useMemo(
     () => [
       {
@@ -133,6 +136,11 @@ export const FrontendInterfaceSelector = ({
     );
   };
 
+  const handleClearOpenWebUIData = async () => {
+    await window.electronAPI.dependencies.clearOpenWebUIData();
+    setShowClearDataModal(false);
+  };
+
   const renderDisabledFrontendWarnings = () => {
     const disabledFrontends = frontendConfigs.filter(
       (config) => !isFrontendAvailable(config.value)
@@ -256,6 +264,7 @@ export const FrontendInterfaceSelector = ({
           </Text>
         )}
 
+        <Group gap="xs" align="flex-end">
           <Select
             value={frontendPreference}
             onChange={handleFrontendPreferenceChange}
@@ -265,12 +274,62 @@ export const FrontendInterfaceSelector = ({
               label: config.label,
               disabled: !isFrontendAvailable(config.value),
             }))}
-          leftSection={<Monitor style={{ width: rem(16), height: rem(16) }} />}
+            leftSection={
+              <Monitor style={{ width: rem(16), height: rem(16) }} />
+            }
+            style={{ flex: 1 }}
           />
 
+          {frontendPreference === 'openwebui' && (
+            <Button
+              variant="light"
+              color="orange"
+              onClick={() => setShowClearDataModal(true)}
+              disabled={isOnInterfaceScreen}
+            >
+              Clear Data
+            </Button>
+          )}
+        </Group>
+
         {renderDisabledFrontendWarnings()}
       </div>
 
+      <Modal
+        opened={showClearDataModal}
+        onClose={() => setShowClearDataModal(false)}
+        title="Clear Open WebUI Data?"
+      >
+        <Stack gap="md">
+          <Text size="sm" c="dimmed">
+            This will permanently delete all Open WebUI data including:
+          </Text>
+          <Box component="ul" pl="md">
+            <Text component="li" size="sm" c="dimmed">
+              Chat history
+            </Text>
+            <Text component="li" size="sm" c="dimmed">
+              User settings
+            </Text>
+            <Text component="li" size="sm" c="dimmed">
+              Database
+            </Text>
+          </Box>
+
+          <Group justify="flex-end" gap="sm">
+            <Button
+              variant="subtle"
+              onClick={() => setShowClearDataModal(false)}
+            >
+              Cancel
+            </Button>
+            <Button color="red" onClick={handleClearOpenWebUIData}>
+              Clear Data
+            </Button>
+          </Group>
+        </Stack>
+      </Modal>
+
       <div>
         <Text fw={500} mb="xs">
           Image Generation Frontend

src/hooks/useLaunchLogic.ts

@@ -43,6 +43,8 @@ interface LaunchArgs {
   sdclipgpu: boolean;
   moecpu: number;
   moeexperts: number;
+  smartcache: boolean;
+  pipelineparallel: boolean;
 }
 
 const buildModelArgs = (
@@ -149,6 +151,14 @@ const buildConfigArgs = (isImageMode: boolean, launchArgs: LaunchArgs) => {
     args.push('--moecpu', launchArgs.moecpu.toString());
   }
 
+  if (launchArgs.smartcache) {
+    args.push('--smartcache');
+  }
+
+  if (launchArgs.pipelineparallel) {
+    args.push('--pipelineparallel');
+  }
+
   return args;
 };
 

src/main/ipc.ts

@@ -290,6 +290,18 @@ export function setupIPCHandlers() {
 
   ipcMain.handle('dependencies:isUvAvailable', () => isUvAvailable());
 
+  ipcMain.handle('dependencies:clearOpenWebUIData', async () => {
+    const { rm } = await import('fs/promises');
+    const openWebUIDataDir = join(getInstallDir(), 'openwebui-data');
+    try {
+      await rm(openWebUIDataDir, { recursive: true, force: true });
+      return { success: true };
+    } catch (error) {
+      logError('Failed to clear Open WebUI data:', error as Error);
+      return { success: false, error: (error as Error).message };
+    }
+  });
+
   ipcMain.on('monitoring:start', () => startMonitoring(mainWindow));
 
   ipcMain.on('monitoring:stop', () => stopMonitoring());

src/main/modules/koboldcpp/launcher/index.ts

@@ -16,7 +16,12 @@ import {
 } from '@/main/modules/config';
 import { startFrontend as startSillyTavernFrontend } from '@/main/modules/sillytavern';
 import { startFrontend as startOpenWebUIFrontend } from '@/main/modules/openwebui';
-import { patchKliteEmbd, patchKcppSduiEmbd, filterSpam } from './patches';
+import {
+  patchKliteEmbd,
+  patchKcppSduiEmbd,
+  patchLcppGzEmbd,
+  filterSpam,
+} from './patches';
 import { startProxy, stopProxy } from '../proxy';
 import { startTunnel, stopTunnel } from '../tunnel';
 import { resolveModelPath, abortActiveDownloads } from '../model-download';
@@ -191,7 +196,11 @@ export async function launchKoboldCpp(
       if (isTextMode) {
         await patchKliteEmbd(binaryDir);
       }
-    } else if (isImageMode && imageGenerationFrontendPreference === 'builtin') {
+    } else if (frontendPreference === 'llamacpp') {
+      await patchLcppGzEmbd(binaryDir);
+    }
+
+    if (isImageMode && imageGenerationFrontendPreference === 'builtin') {
       await patchKcppSduiEmbd(binaryDir);
     }
 

src/main/modules/koboldcpp/launcher/patches.ts

@@ -112,6 +112,23 @@ export const patchKcppSduiEmbd = (unpackedDir: string) =>
     }
   }, 'Failed to patch kcpp_sdui.embd');
 
+export const patchLcppGzEmbd = (unpackedDir: string) =>
+  tryExecute(async () => {
+    const possiblePaths = [
+      join(unpackedDir, '_internal', 'embd_res', 'lcpp.gz.embd'),
+      join(unpackedDir, 'lcpp.gz.embd'),
+    ];
+
+    const sourceAssetPath = getAssetPath('lcpp.gz.embd');
+
+    for (const targetPath of possiblePaths) {
+      if (await pathExists(targetPath)) {
+        await copyFile(sourceAssetPath, targetPath);
+        break;
+      }
+    }
+  }, 'Failed to patch lcpp.gz.embd');
+
 export function filterSpam(output: string) {
   const spamPatterns = [
     /^print_info:/,

src/main/modules/openwebui.ts

@@ -15,7 +15,13 @@ import { PROXY } from '@/constants/proxy';
 
 let openWebUIProcess: ChildProcess | null = null;
 
-const OPENWEBUI_BASE_ARGS = ['--python', '3.11', 'open-webui@latest', 'serve'];
+const OPENWEBUI_VERSION = '0.6.41';
+const OPENWEBUI_BASE_ARGS = [
+  '--python',
+  '3.11',
+  `open-webui@${OPENWEBUI_VERSION}`,
+  'serve',
+];
 
 on('SIGINT', () => {
   void stopFrontend();

src/preload/index.ts

@@ -208,6 +208,8 @@ const logsAPI: LogsAPI = {
 const dependenciesAPI: DependenciesAPI = {
   isUvAvailable: () => ipcRenderer.invoke('dependencies:isUvAvailable'),
   isNpxAvailable: () => ipcRenderer.invoke('dependencies:isNpxAvailable'),
+  clearOpenWebUIData: () =>
+    ipcRenderer.invoke('dependencies:clearOpenWebUIData'),
 };
 
 const monitoringAPI: MonitoringAPI = {

src/stores/launchConfig.ts

@@ -41,6 +41,8 @@ interface LaunchConfigState {
   sdclipgpu: boolean;
   moecpu: number;
   moeexperts: number;
+  smartcache: boolean;
+  pipelineparallel: boolean;
   isImageGenerationMode: boolean;
   isTextMode: boolean;
 
@@ -81,6 +83,8 @@ interface LaunchConfigState {
   setSdclipgpu: (enabled: boolean) => void;
   setMoecpu: (moecpu: number) => void;
   setMoeexperts: (moeexperts: number) => void;
+  setSmartcache: (smartcache: boolean) => void;
+  setPipelineparallel: (pipelineparallel: boolean) => void;
 
   parseAndApplyConfigFile: (configPath: string) => Promise<void>;
   loadConfigFromFile: (
@@ -141,6 +145,8 @@ export const useLaunchConfigStore = create<LaunchConfigState>((set, get) => ({
   sdclipgpu: false,
   moecpu: 0,
   moeexperts: -1,
+  smartcache: false,
+  pipelineparallel: false,
 
   isImageGenerationMode: false,
   isTextMode: false,
@@ -195,6 +201,8 @@ export const useLaunchConfigStore = create<LaunchConfigState>((set, get) => ({
   setSdclipgpu: (enabled) => set({ sdclipgpu: enabled }),
   setMoecpu: (moeCpu) => set({ moecpu: moeCpu }),
   setMoeexperts: (moeExperts) => set({ moeexperts: moeExperts }),
+  setSmartcache: (smartcache) => set({ smartcache }),
+  setPipelineparallel: (pipelineparallel) => set({ pipelineparallel }),
 
   // eslint-disable-next-line sonarjs/cognitive-complexity
   parseAndApplyConfigFile: async (configPath: string) => {
@@ -426,6 +434,18 @@ export const useLaunchConfigStore = create<LaunchConfigState>((set, get) => ({
         updates.moeexperts = -1;
       }
 
+      if (typeof configData.smartcache === 'boolean') {
+        updates.smartcache = configData.smartcache;
+      } else {
+        updates.smartcache = false;
+      }
+
+      if (typeof configData.pipelineparallel === 'boolean') {
+        updates.pipelineparallel = configData.pipelineparallel;
+      } else {
+        updates.pipelineparallel = false;
+      }
+
       set(updates);
     }
   },

src/types/electron.d.ts

@@ -121,6 +121,8 @@ export interface KoboldConfig {
   preLaunchCommands?: string[];
   moecpu?: number;
   moeexperts?: number;
+  smartcache?: boolean;
+  pipelineparallel?: boolean;
   autoGpuLayers?: boolean;
   model?: string;
   backend?: string;
@@ -252,6 +254,7 @@ export interface LogsAPI {
 export interface DependenciesAPI {
   isNpxAvailable: () => Promise<boolean>;
   isUvAvailable: () => Promise<boolean>;
+  clearOpenWebUIData: () => Promise<{ success: boolean; error?: string }>;
 }
 
 export interface MonitoringAPI {

yarn.lock

@@ -315,15 +315,15 @@ __metadata:
   languageName: node
   linkType: hard
 
-"@codemirror/view@npm:^6.0.0, @codemirror/view@npm:^6.17.0, @codemirror/view@npm:^6.23.0, @codemirror/view@npm:^6.27.0, @codemirror/view@npm:^6.35.0, @codemirror/view@npm:^6.39.4":
+"@codemirror/view@npm:^6.0.0, @codemirror/view@npm:^6.17.0, @codemirror/view@npm:^6.23.0, @codemirror/view@npm:^6.27.0, @codemirror/view@npm:^6.35.0, @codemirror/view@npm:^6.39.5":
-  version: 6.39.4
+  version: 6.39.5
-  resolution: "@codemirror/view@npm:6.39.4"
+  resolution: "@codemirror/view@npm:6.39.5"
   dependencies:
     "@codemirror/state": "npm:^6.5.0"
     crelt: "npm:^1.0.6"
     style-mod: "npm:^4.1.0"
     w3c-keyname: "npm:^2.2.4"
-  checksum: 10c0/aa17cfbfb0714243cccc9e970f4f382c3e747435f2fc63d40125370379227436fb529473dde856a276b8b7df98f83c55f4cc5cd0c24a56335f24233103b7dfd0
+  checksum: 10c0/54fc600b54c336283e8c73b6a0a99156880682121d73c2d4b74fc5a93f956e33b54ba16cd7debbf5fc28a48b18aeb643217e5200e3af0588909ca90cb275e957
   languageName: node
   linkType: hard
 
@@ -4124,7 +4124,7 @@ __metadata:
   dependencies:
     "@codemirror/search": "npm:^6.5.11"
     "@codemirror/theme-one-dark": "npm:^6.1.3"
-    "@codemirror/view": "npm:^6.39.4"
+    "@codemirror/view": "npm:^6.39.5"
     "@eslint/js": "npm:^9.39.2"
     "@fontsource/inter": "npm:^5.2.8"
     "@huggingface/gguf": "npm:^0.3.2"
@@ -4154,7 +4154,7 @@ __metadata:
     execa: "npm:^9.6.1"
     globals: "npm:^16.5.0"
     jiti: "npm:^2.6.1"
-    lucide-react: "npm:^0.561.0"
+    lucide-react: "npm:^0.562.0"
     mime-types: "npm:^3.0.2"
     prettier: "npm:^3.7.4"
     react: "npm:^19.2.3"
@@ -5387,12 +5387,12 @@ __metadata:
   languageName: node
   linkType: hard
 
-"lucide-react@npm:^0.561.0":
+"lucide-react@npm:^0.562.0":
-  version: 0.561.0
+  version: 0.562.0
-  resolution: "lucide-react@npm:0.561.0"
+  resolution: "lucide-react@npm:0.562.0"
   peerDependencies:
     react: ^16.5.1 || ^17.0.0 || ^18.0.0 || ^19.0.0
-  checksum: 10c0/d23c96a543bdb1e11cec1f05a631f86f961f2ee17a10da42e7c297bb06c5f1587a1b66927f58082c67d8edc0749b0a5e44e543b461deab9db19d22fb9ca93cd7
+  checksum: 10c0/c713a2490916e42c678b9df0d7309d8cdb715af1d7139c7eb49c45547901b276a7cf722548012a668dc0e337eeadf5141d5af8c8fcda6721fc58fb2d288efcba
   languageName: node
   linkType: hard