diff --git a/src/main/index.ts b/src/main/index.ts
index 422f20f..6c20418 100644
--- a/src/main/index.ts
+++ b/src/main/index.ts
@@ -65,7 +65,8 @@ import {
   getLlamaCppPty,
   validateLlamaCppProcess,
   checkLlamaCppUpdate,
-  updateLlamaCpp
+  updateLlamaCpp,
+  uninstallLlamaCpp
 } from './utils/llamacpp'
 
 import {
@@ -907,6 +908,28 @@ if (!gotTheLock) {
     ipcMain.handle('llamacpp:logs', () => getLlamaCppLog())
     ipcMain.handle('llamacpp:pty:connect', () => connectLlamaCppPtyPort())
 
+    ipcMain.handle('llamacpp:uninstall', async () => {
+      try {
+        const info = getLlamaCppInfo()
+        await uninstallLlamaCpp()
+        sendToRenderer('status:llamacpp', null)
+        // Unregister OpenAI endpoint if it was running
+        if (info.url) {
+          sendToRenderer('connections:openai', {
+            action: 'remove',
+            url: `${info.url}/v1`
+          })
+          setTimeout(() => sendToRenderer('models:refresh'), 500)
+        }
+        await setConfig({ llamaCpp: { ...CONFIG?.llamaCpp, enabled: false } })
+        CONFIG = await getConfig()
+        return true
+      } catch (error) {
+        log.error('Failed to uninstall llamacpp:', error)
+        return false
+      }
+    })
+
     // Hugging Face models
     ipcMain.handle('huggingface:models:list', () => listModels())
     ipcMain.handle('huggingface:models:dir', () => getModelsDir())
diff --git a/src/main/utils/llamacpp.ts b/src/main/utils/llamacpp.ts
index 0677f7b..dd9084e 100644
--- a/src/main/utils/llamacpp.ts
+++ b/src/main/utils/llamacpp.ts
@@ -505,3 +505,18 @@ export const validateLlamaCppProcess = (): boolean => {
   lock.release()
   return false
 }
+
+/**
+ * Uninstall llama.cpp — stop the server and remove all downloaded binaries.
+ */
+export const uninstallLlamaCpp = async (): Promise<void> => {
+  await stopLlamaCpp()
+
+  const cacheBase = path.join(getUserDataPath(), 'llama.cpp')
+  if (fs.existsSync(cacheBase)) {
+    fs.rmSync(cacheBase, { recursive: true, force: true })
+    log.info('Removed llama.cpp directory:', cacheBase)
+  }
+
+  binaryPath = null
+}
diff --git a/src/preload/index.ts b/src/preload/index.ts
index 21f2963..28d461c 100644
--- a/src/preload/index.ts
+++ b/src/preload/index.ts
@@ -175,6 +175,7 @@ const api = {
   },
   checkLlamaCppUpdate: () => ipcRenderer.invoke('llamacpp:check-update'),
   updateLlamaCpp: () => ipcRenderer.invoke('llamacpp:update'),
+  uninstallLlamaCpp: () => ipcRenderer.invoke('llamacpp:uninstall'),
 
   // Hugging Face models
   listHfModels: () => ipcRenderer.invoke('huggingface:models:list'),
diff --git a/src/renderer/src/lib/components/Main/Connections/GetStartedModal.svelte b/src/renderer/src/lib/components/Main/Connections/GetStartedModal.svelte
index 074629f..262f8b1 100644
--- a/src/renderer/src/lib/components/Main/Connections/GetStartedModal.svelte
+++ b/src/renderer/src/lib/components/Main/Connections/GetStartedModal.svelte
@@ -67,7 +67,7 @@
         <div>
           <div class="text-[13px] font-medium text-gray-700 dark:text-gray-300 flex items-center gap-1.5">
             {$i18n.t('main.getStarted.llamaCpp')}
-            <span class="text-[9px] px-1.5 py-0.5 rounded-full bg-amber-500/10 text-amber-600 dark:text-amber-400 uppercase tracking-wide font-medium">{$i18n.t('common.experimental')}</span>
+            <span class="text-[9px] opacity-30 uppercase tracking-wide">{$i18n.t('common.experimental')}</span>
           </div>
           <div class="text-[11px] text-gray-400 dark:text-gray-500 mt-0.5">{$i18n.t('main.getStarted.llamaCppDesc')}</div>
         </div>
diff --git a/src/renderer/src/lib/components/Main/Settings/InferenceRuntime.svelte b/src/renderer/src/lib/components/Main/Settings/InferenceRuntime.svelte
index 6e1f859..717389c 100644
--- a/src/renderer/src/lib/components/Main/Settings/InferenceRuntime.svelte
+++ b/src/renderer/src/lib/components/Main/Settings/InferenceRuntime.svelte
@@ -11,6 +11,8 @@
   let settingUp = $state(false)
   let loaded = $state(false)
   let setupStatus = $state('')
+  let uninstalling = $state(false)
+  let installing = $state(false)
 
   type UpdateStatus = 'idle' | 'checking' | 'available' | 'updating' | 'up-to-date' | 'error'
   let updateStatus = $state<UpdateStatus>('idle')
@@ -166,6 +168,8 @@
     deleting = null
   }
 
+  const installed = $derived(!!lsInfo?.binaryPath)
+
   const formatSize = (bytes: number): string => {
     if (bytes < 1024) return `${bytes} B`
     if (bytes < 1024 * 1024) return `${(bytes / 1024).toFixed(1)} KB`
@@ -176,6 +180,37 @@
 
 {#if !loaded}
   <div class="py-6 text-[12px] opacity-20 text-center">{$i18n.t('common.loading')}</div>
+{:else if !installed}
+  <div class="py-4 flex items-center justify-between">
+    <div>
+      <div class="text-[13px] opacity-40 flex items-center gap-1.5">
+        {$i18n.t('settings.inference.notInstalled')}
+        <span class="text-[9px] opacity-30 uppercase tracking-wide">{$i18n.t('common.experimental')}</span>
+      </div>
+      <div class="text-[11px] opacity-20 mt-0.5">{$i18n.t('settings.inference.notInstalledDesc')}</div>
+    </div>
+    <button
+      class="text-[12px] opacity-40 hover:opacity-70 px-3 py-1.5 bg-black/[0.04] dark:bg-white/[0.06] transition border-none text-[#1d1d1f] dark:text-[#fafafa] rounded-xl flex items-center gap-1.5 {installing ? 'pointer-events-none opacity-20' : ''}"
+      disabled={installing}
+      onclick={async () => {
+        installing = true
+        try {
+          await window.electronAPI.startLlamaCpp()
+          lsInfo = await window.electronAPI.getLlamaCppInfo()
+        } catch (e) {
+          console.error('Failed to install:', e)
+        }
+        installing = false
+      }}
+    >
+      {#if installing}
+        <div class="w-2.5 h-2.5 rounded-full border-[1.5px] border-black/20 dark:border-white/30 border-t-transparent animate-spin"></div>
+        {$i18n.t('common.installing')}
+      {:else}
+        {$i18n.t('common.install')}
+      {/if}
+    </button>
+  </div>
 {:else}
 <div class="flex flex-col divide-y divide-white/[0.04]">
   <!-- Server status & controls -->
@@ -184,7 +219,7 @@
       <div>
         <div class="text-[13px] opacity-70 flex items-center gap-1.5">
           {$i18n.t('settings.inference.llamaServer')}
-          <span class="text-[9px] px-1.5 py-0.5 rounded-full bg-amber-500/10 text-amber-600 dark:text-amber-400 uppercase tracking-wide font-medium">{$i18n.t('common.experimental')}</span>
+          <span class="text-[9px] opacity-30 uppercase tracking-wide">{$i18n.t('common.experimental')}</span>
         </div>
         <div class="text-[11px] opacity-25 mt-0.5">
           {$i18n.t('settings.inference.llamaServerDesc')}
@@ -430,5 +465,38 @@
     />
   </div>
 
+  <!-- Uninstall -->
+  {#if lsInfo?.binaryPath}
+  <div class="py-4 flex items-center justify-between">
+    <div>
+      <div class="text-[13px] opacity-70">{$i18n.t('settings.inference.uninstall')}</div>
+      <div class="text-[11px] opacity-25 mt-0.5">{$i18n.t('settings.inference.uninstallDesc')}</div>
+    </div>
+    <button
+      class="text-[12px] opacity-40 hover:opacity-70 px-3 py-1.5 bg-black/[0.04] dark:bg-white/[0.06] transition border-none text-[#1d1d1f] dark:text-[#fafafa] rounded-xl flex items-center gap-1.5 {uninstalling ? 'pointer-events-none opacity-20' : ''}"
+      disabled={uninstalling}
+      onclick={async () => {
+        if (confirm($i18n.t('settings.inference.uninstallConfirm'))) {
+          uninstalling = true
+          try {
+            await window.electronAPI.uninstallLlamaCpp()
+            lsInfo = await window.electronAPI.getLlamaCppInfo()
+          } catch (e) {
+            console.error('Failed to uninstall llama.cpp:', e)
+          }
+          uninstalling = false
+        }
+      }}
+    >
+      {#if uninstalling}
+        <div class="w-2.5 h-2.5 rounded-full border-[1.5px] border-black/20 dark:border-white/30 border-t-transparent animate-spin"></div>
+        {$i18n.t('common.uninstalling')}
+      {:else}
+        {$i18n.t('common.uninstall')}
+      {/if}
+    </button>
+  </div>
+  {/if}
+
 </div>
 {/if}
diff --git a/src/renderer/src/lib/i18n/locales/en-US/translation.json b/src/renderer/src/lib/i18n/locales/en-US/translation.json
index 7f64ea6..5ca0a02 100644
--- a/src/renderer/src/lib/i18n/locales/en-US/translation.json
+++ b/src/renderer/src/lib/i18n/locales/en-US/translation.json
@@ -235,6 +235,11 @@
   "settings.inference.updateAvailable": "Update available: {{version}}",
   "settings.inference.upToDate": "Up to date",
   "settings.inference.updating": "Updating…",
+  "settings.inference.uninstall": "Uninstall",
+  "settings.inference.uninstallDesc": "Remove llama.cpp binaries",
+  "settings.inference.uninstallConfirm": "This will stop llama-server and remove all downloaded binaries. Continue?",
+  "settings.inference.notInstalled": "llama.cpp is not installed",
+  "settings.inference.notInstalledDesc": "Download and set up llama.cpp for local model inference",
 
   "settings.models.downloadedModels": "Downloaded Models",
   "settings.models.noModels": "No models downloaded yet",