Refactor inference around external Ollama routing

Cargo.toml

@@ -169,6 +169,8 @@ rppal = { version = "0.22", optional = true }
 # crates we never use (and that bloat the dev Cargo.lock noticeably).
 # TestTransport only needs the `test` feature.
 sentry = { version = "0.47.0", default-features = false, features = ["test"] }
+# Mock HTTP server for provider E2E tests (inference_provider_e2e).
+wiremock = "0.6"
 
 [features]
 sandbox-landlock = ["dep:landlock"]

app/src/components/settings/panels/LocalModelDebugPanel.tsx

@@ -17,13 +17,10 @@ import {
   type LocalAiTtsResult,
   openhumanLocalAiAssetsStatus,
   openhumanLocalAiDiagnostics,
-  openhumanLocalAiDownload,
-  openhumanLocalAiDownloadAllAssets,
   openhumanLocalAiDownloadAsset,
   openhumanLocalAiDownloadsProgress,
   openhumanLocalAiEmbed,
   openhumanLocalAiPrompt,
-  openhumanLocalAiSetOllamaPath,
   openhumanLocalAiStatus,
   openhumanLocalAiSummarize,
   openhumanLocalAiTranscribe,
@@ -60,8 +57,6 @@ const LocalModelDebugPanel = () => {
   const [assets, setAssets] = useState<LocalAiAssetsStatus | null>(null);
   const [downloads, setDownloads] = useState<LocalAiDownloadsProgress | null>(null);
   const [statusError, setStatusError] = useState<string>('');
-  const [isTriggeringDownload, setIsTriggeringDownload] = useState(false);
-  const [bootstrapMessage, setBootstrapMessage] = useState<string>('');
   const [assetDownloadBusy, setAssetDownloadBusy] = useState<Record<string, boolean>>({});
 
   const [summaryInput, setSummaryInput] = useState('');
@@ -97,8 +92,6 @@ const LocalModelDebugPanel = () => {
   const [diagnosticsError, setDiagnosticsError] = useState('');
 
   const [showErrorDetail, setShowErrorDetail] = useState(false);
-  const [ollamaPathInput, setOllamaPathInput] = useState('');
-  const [isSettingPath, setIsSettingPath] = useState(false);
 
   const progress = useMemo(() => {
     const downloadProgress = progressFromDownloads(downloads);
@@ -158,29 +151,6 @@ const LocalModelDebugPanel = () => {
     };
   }, []);
 
-  const triggerDownload = async (force: boolean) => {
-    if (!runtimeEnabled) return;
-    setIsTriggeringDownload(true);
-    setStatusError('');
-    setBootstrapMessage('');
-    try {
-      await openhumanLocalAiDownload(force);
-      await openhumanLocalAiDownloadAllAssets(force);
-      const freshStatus = await openhumanLocalAiStatus();
-      setStatus(freshStatus.result);
-      if (freshStatus.result?.state === 'ready') {
-        setBootstrapMessage(force ? 'Re-bootstrap complete' : 'Models verified');
-      }
-      setTimeout(() => setBootstrapMessage(''), 3000);
-    } catch (err) {
-      const message =
-        err instanceof Error ? err.message : 'Failed to trigger local model bootstrap';
-      setStatusError(message);
-    } finally {
-      setIsTriggeringDownload(false);
-    }
-  };
-
   const runSummaryTest = async () => {
     if (!runtimeEnabled || !summaryInput.trim()) return;
     setIsSummaryLoading(true);
@@ -305,32 +275,6 @@ const LocalModelDebugPanel = () => {
     }
   };
 
-  const handleSetOllamaPath = async () => {
-    setIsSettingPath(true);
-    setStatusError('');
-    try {
-      await openhumanLocalAiSetOllamaPath(ollamaPathInput);
-      await loadStatus();
-    } catch (err) {
-      setStatusError(err instanceof Error ? err.message : 'Failed to set Ollama path');
-    } finally {
-      setIsSettingPath(false);
-    }
-  };
-
-  const handleClearOllamaPath = async () => {
-    setOllamaPathInput('');
-    setIsSettingPath(true);
-    try {
-      await openhumanLocalAiSetOllamaPath('');
-      await loadStatus();
-    } catch (err) {
-      setStatusError(err instanceof Error ? err.message : 'Failed to clear Ollama path');
-    } finally {
-      setIsSettingPath(false);
-    }
-  };
-
   const handleRunDiagnostics = async () => {
     setIsDiagnosticsLoading(true);
     setDiagnosticsError('');
@@ -361,25 +305,25 @@ const LocalModelDebugPanel = () => {
           isDiagnosticsLoading={isDiagnosticsLoading}
           diagnosticsError={diagnosticsError}
           statusError={statusError}
-          isTriggeringDownload={isTriggeringDownload}
-          bootstrapMessage={bootstrapMessage}
+          isTriggeringDownload={false}
+          bootstrapMessage=""
           progress={progress}
           isIndeterminateDownload={isIndeterminateDownload}
           isInstalling={isInstalling}
           isInstallError={isInstallError}
           showErrorDetail={showErrorDetail}
-          ollamaPathInput={ollamaPathInput}
-          isSettingPath={isSettingPath}
+          ollamaPathInput=""
+          isSettingPath={false}
           downloadedText={downloadedText}
           speedText={speedText}
           etaText={etaText}
           statusTone={statusTone}
           runtimeEnabled={runtimeEnabled}
           onRefreshStatus={() => void loadStatus()}
-          onTriggerDownload={force => void triggerDownload(force)}
-          onSetOllamaPath={() => void handleSetOllamaPath()}
-          onClearOllamaPath={() => void handleClearOllamaPath()}
-          onSetOllamaPathInput={setOllamaPathInput}
+          onTriggerDownload={() => {}}
+          onSetOllamaPath={() => {}}
+          onClearOllamaPath={() => {}}
+          onSetOllamaPathInput={() => {}}
           onToggleErrorDetail={() => setShowErrorDetail(v => !v)}
           onRunDiagnostics={() => void handleRunDiagnostics()}
         />

app/src/components/settings/panels/local-model/DeviceCapabilitySection.test.tsx

@@ -0,0 +1,87 @@
+import { fireEvent, render, screen, waitFor } from '@testing-library/react';
+import { beforeEach, describe, expect, it, vi } from 'vitest';
+
+import DeviceCapabilitySection from './DeviceCapabilitySection';
+
+const mockApplyPreset = vi.fn();
+
+vi.mock('../../../../utils/tauriCommands', () => ({
+  openhumanLocalAiApplyPreset: (...args: unknown[]) => mockApplyPreset(...args),
+}));
+
+const makePresetsData = (overrides: Record<string, unknown> = {}) => ({
+  presets: [
+    {
+      tier: 'ram_2_4gb',
+      label: '2-4 GB',
+      description: 'Small local tier',
+      chat_model_id: 'gemma3:1b-it-qat',
+      vision_model_id: '',
+      embedding_model_id: 'bge-m3',
+      quantization: 'q4',
+      vision_mode: 'disabled',
+      supports_screen_summary: false,
+      target_ram_gb: 4,
+      min_ram_gb: 2,
+      approx_download_gb: 1.2,
+    },
+  ],
+  recommended_tier: 'ram_2_4gb',
+  current_tier: 'ram_2_4gb',
+  selected_tier: 'ram_2_4gb',
+  recommend_disabled: false,
+  local_ai_enabled: true,
+  device: {
+    total_ram_bytes: 16 * 1024 * 1024 * 1024,
+    cpu_count: 8,
+    cpu_brand: 'Test CPU',
+    os_name: 'macOS',
+    os_version: '15',
+    has_gpu: true,
+    gpu_description: 'Test GPU',
+  },
+  ...overrides,
+});
+
+describe('DeviceCapabilitySection', () => {
+  beforeEach(() => {
+    mockApplyPreset.mockReset();
+  });
+
+  it('renders external runtime guidance when ollama is unavailable', () => {
+    render(
+      <DeviceCapabilitySection
+        presetsData={makePresetsData()}
+        presetsLoading={false}
+        presetError=""
+        presetSuccess={null}
+        formatRamGb={() => '16 GB'}
+        ollamaAvailable={false}
+      />
+    );
+
+    expect(screen.getByText(/Run Ollama first/i)).toBeTruthy();
+    expect(screen.getByRole('link', { name: 'Ollama docs' })).toBeTruthy();
+    expect(screen.getByTitle('Run Ollama first to use this tier')).toBeTruthy();
+  });
+
+  it('allows selecting the disabled cloud fallback tier', async () => {
+    mockApplyPreset.mockResolvedValueOnce({ applied_tier: 'disabled' });
+
+    render(
+      <DeviceCapabilitySection
+        presetsData={makePresetsData({ local_ai_enabled: false })}
+        presetsLoading={false}
+        presetError=""
+        presetSuccess={null}
+        formatRamGb={() => '16 GB'}
+      />
+    );
+
+    fireEvent.click(screen.getByRole('button', { name: /Disabled.*0 GB/i }));
+
+    await waitFor(() => {
+      expect(mockApplyPreset).toHaveBeenCalledWith('disabled');
+    });
+  });
+});

app/src/components/settings/panels/local-model/DeviceCapabilitySection.tsx

@@ -14,30 +14,15 @@ interface DeviceCapabilitySectionProps {
   formatRamGb: (bytes: number) => string;
   onPresetApplied?: (result: ApplyPresetResult) => void;
   /**
-   * When `false`, the Ollama runtime isn't installed yet. Local tiers
-   * require Ollama, so they're rendered disabled with a notice that
-   * lets the user install Ollama in place. The "Disabled (cloud
-   * fallback)" option stays enabled since it doesn't need Ollama.
+   * When `false`, the external Ollama runtime isn't reachable yet. Local tiers
+   * stay disabled until the user runs Ollama themselves. The "Disabled (cloud
+   * fallback)" option stays enabled since it doesn't depend on Ollama.
    */
   ollamaAvailable?: boolean;
-  /**
-   * Triggers the same install pipeline the Runtime Status section uses.
-   * Wired only when `ollamaAvailable === false` to surface an inline
-   * Install Ollama button next to the locked tiers.
-   */
   onTriggerOllamaInstall?: () => void;
-  /** True while an install pipeline is already running. */
   isTriggeringInstall?: boolean;
-  /**
-   * Live state from `local_ai_status` so the notice can show real install
-   * progress: `installing`, `downloading`, `degraded`, etc. The button's
-   * own `isTriggeringInstall` only covers the RPC round-trip (~ms);
-   * `installState` covers the entire backend pipeline (~60s).
-   */
   installState?: string;
-  /** Latest `status.warning` text — shown under the progress label. */
   installWarning?: string | null;
-  /** Latest `status.error_detail` — shown when state is `degraded`. */
   installError?: string | null;
 }
 
@@ -57,9 +42,13 @@ const DeviceCapabilitySection = ({
   installWarning,
   installError,
 }: DeviceCapabilitySectionProps) => {
-  const installInProgress =
-    installState === 'installing' || installState === 'downloading' || installState === 'loading';
-  const installFailed = installState === 'degraded';
+  void onTriggerOllamaInstall;
+  void isTriggeringInstall;
+  void installState;
+  void installWarning;
+  void installError;
+  const installInProgress = false;
+  const installFailed = false;
   const [applying, setApplying] = useState<string | null>(null);
   const [applyError, setApplyError] = useState<string>('');
   const [applySuccess, setApplySuccess] = useState<ApplyPresetResult | null>(null);
@@ -187,26 +176,18 @@ const DeviceCapabilitySection = ({
           ) : (
             <>
               <div className="text-xs text-amber-800">
-                <span className="font-semibold text-amber-900">Install Ollama first.</span> Local
-                tiers run on the Ollama runtime, which isn&apos;t installed yet. The &ldquo;Disabled
-                (cloud fallback)&rdquo; option stays available either way.
+                <span className="font-semibold text-amber-900">Run Ollama first.</span> Local tiers
+                depend on an externally managed Ollama endpoint. Start it yourself, pull the models
+                you want, and keep using &ldquo;Disabled (cloud fallback)&rdquo; until the runtime
+                is reachable.
               </div>
               <div className="flex items-center gap-2">
-                {onTriggerOllamaInstall && (
-                  <button
-                    type="button"
-                    onClick={onTriggerOllamaInstall}
-                    disabled={isTriggeringInstall}
-                    className="px-3 py-1.5 text-xs rounded-md bg-amber-600 hover:bg-amber-700 disabled:opacity-60 text-white font-medium">
-                    {isTriggeringInstall ? 'Starting…' : 'Install Ollama'}
-                  </button>
-                )}
                 <a
                   href="https://ollama.com"
                   target="_blank"
                   rel="noopener noreferrer"
                   className="px-3 py-1.5 text-xs rounded-md border border-amber-300 hover:border-amber-400 text-amber-800">
-                  Install manually
+                  Ollama docs
                 </a>
               </div>
             </>
@@ -257,7 +238,7 @@ const DeviceCapabilitySection = ({
                 key={preset.tier}
                 onClick={() => void handleApply(preset.tier)}
                 disabled={applying !== null || locked}
-                title={locked ? 'Install Ollama first to use this tier' : undefined}
+                title={locked ? 'Run Ollama first to use this tier' : undefined}
                 className={`w-full text-left rounded-lg border p-3 transition-colors ${
                   isCurrent
                     ? 'border-primary-400 bg-primary-50'