diff --git a/.claude/memory.md b/.claude/memory.md
index 11523061b..96d9e8c5e 100644
--- a/.claude/memory.md
+++ b/.claude/memory.md
@@ -25,6 +25,15 @@ Quick reference for anyone starting with Claude on this project. Updated by the
 - **Ask user when in doubt** — never assume scope or approach
 - **PRs target upstream** — `tinyhumansai/openhuman` main branch, not fork
 
+## Local AI Presets & Daemon Gotcha
+
+- **Tier system lives in `src/openhuman/local_ai/presets.rs`** — single source of truth for tier→model ID mapping. To change default models for a release, edit `all_presets()` there.
+- **Device detection** uses `sysinfo` crate (`src/openhuman/local_ai/device.rs`). Apple Silicon = GPU always; others = best-effort.
+- **`OPENHUMAN_LOCAL_AI_TIER` env var** overrides the selected tier at config load time (in `load.rs`).
+- **Frontend tier selector** is in `LocalModelPanel.tsx` under Settings > Local AI Model. Uses `coreRpcClient` to call 3 RPC methods: `local_ai_device_profile`, `local_ai_presets`, `local_ai_apply_preset`.
+- **Default config maps to Medium tier** (`gemma3:4b-it-qat`). If someone changes `model_ids.rs` defaults, they should keep `presets.rs` in sync.
+- **Daemon binary gotcha** — A daemon process (`openhuman-aarch64-apple-darwin run`) auto-starts on port 7788 and respawns on kill. `yarn tauri dev` reuses it if already running. When adding new RPC methods, you must replace this binary: `cp -f target/debug/openhuman-core app/src-tauri/binaries/openhuman-aarch64-apple-darwin`, then kill the old PID so it respawns with the new binary.
+
 ## Environment
 
 - **Core sidecar port** — `7788` (default). Check with `lsof -i :7788`.
diff --git a/.env.example b/.env.example
index fc2cfcd82..f2bb43a74 100644
--- a/.env.example
+++ b/.env.example
@@ -88,6 +88,13 @@ OPENHUMAN_PROXY_SCOPE=
 # [optional] Comma-separated services to proxy
 OPENHUMAN_PROXY_SERVICES=
 
+# ---------------------------------------------------------------------------
+# Local AI model tier
+# ---------------------------------------------------------------------------
+# [optional] Override selected model tier: low, medium, high
+# Applies the corresponding preset at config load time (overrides config.toml).
+OPENHUMAN_LOCAL_AI_TIER=
+
 # ---------------------------------------------------------------------------
 # Local AI binary overrides
 # ---------------------------------------------------------------------------
diff --git a/Cargo.lock b/Cargo.lock
index 1ee57c74b..01b85c5fe 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -2458,7 +2458,7 @@ dependencies = [
  "js-sys",
  "log",
  "wasm-bindgen",
- "windows-core",
+ "windows-core 0.62.2",
 ]
 
 [[package]]
@@ -3746,6 +3746,15 @@ dependencies = [
  "nom 8.0.0",
 ]
 
+[[package]]
+name = "ntapi"
+version = "0.4.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c3b335231dfd352ffb0f8017f3b6027a4917f7df785ea2143d8af2adc66980ae"
+dependencies = [
+ "winapi",
+]
+
 [[package]]
 name = "nu-ansi-term"
 version = "0.46.0"
@@ -3968,6 +3977,7 @@ dependencies = [
  "sha2",
  "shellexpand",
  "socketioxide",
+ "sysinfo",
  "tempfile",
  "thiserror 2.0.18",
  "tokio",
@@ -6053,6 +6063,19 @@ dependencies = [
  "syn 2.0.117",
 ]
 
+[[package]]
+name = "sysinfo"
+version = "0.33.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4fc858248ea01b66f19d8e8a6d55f41deaf91e9d495246fd01368d99935c6c01"
+dependencies = [
+ "core-foundation-sys",
+ "libc",
+ "memchr",
+ "ntapi",
+ "windows",
+]
+
 [[package]]
 name = "tagptr"
 version = "0.2.0"
@@ -7459,19 +7482,52 @@ version = "0.4.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "712e227841d057c1ee1cd2fb22fa7e5a5461ae8e48fa2ca79ec42cfc1931183f"
 
+[[package]]
+name = "windows"
+version = "0.57.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "12342cb4d8e3b046f3d80effd474a7a02447231330ef77d71daa6fbc40681143"
+dependencies = [
+ "windows-core 0.57.0",
+ "windows-targets 0.52.6",
+]
+
+[[package]]
+name = "windows-core"
+version = "0.57.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d2ed2439a290666cd67ecce2b0ffaad89c2a56b976b736e6ece670297897832d"
+dependencies = [
+ "windows-implement 0.57.0",
+ "windows-interface 0.57.0",
+ "windows-result 0.1.2",
+ "windows-targets 0.52.6",
+]
+
 [[package]]
 name = "windows-core"
 version = "0.62.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b8e83a14d34d0623b51dce9581199302a221863196a1dde71a7663a4c2be9deb"
 dependencies = [
- "windows-implement",
- "windows-interface",
+ "windows-implement 0.60.2",
+ "windows-interface 0.59.3",
  "windows-link",
- "windows-result",
+ "windows-result 0.4.1",
  "windows-strings",
 ]
 
+[[package]]
+name = "windows-implement"
+version = "0.57.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9107ddc059d5b6fbfbffdfa7a7fe3e22a226def0b2608f72e9d552763d3e1ad7"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
+]
+
 [[package]]
 name = "windows-implement"
 version = "0.60.2"
@@ -7483,6 +7539,17 @@ dependencies = [
  "syn 2.0.117",
 ]
 
+[[package]]
+name = "windows-interface"
+version = "0.57.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "29bee4b38ea3cde66011baa44dba677c432a78593e202392d1e9070cf2a7fca7"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
+]
+
 [[package]]
 name = "windows-interface"
 version = "0.59.3"
@@ -7500,6 +7567,15 @@ version = "0.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f0805222e57f7521d6a62e36fa9163bc891acd422f971defe97d64e70d0a4fe5"
 
+[[package]]
+name = "windows-result"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5e383302e8ec8515204254685643de10811af0ed97ea37210dc26fb0032647f8"
+dependencies = [
+ "windows-targets 0.52.6",
+]
+
 [[package]]
 name = "windows-result"
 version = "0.4.1"
diff --git a/Cargo.toml b/Cargo.toml
index cc81b3a32..3e75b6f2c 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -67,6 +67,7 @@ rustls = { version = "0.23", features = ["ring"] }
 rustls-pki-types = "1.14.0"
 tokio-rustls = "0.26.4"
 webpki-roots = "1.0.6"
+sysinfo = { version = "0.33", default-features = false, features = ["system"] }
 clap = { version = "4.5", features = ["derive"] }
 clap_complete = "4.5"
 lettre = { version = "0.11.19", default-features = false, features = ["builder", "smtp-transport", "rustls-tls"] }
diff --git a/app/src-tauri/Cargo.lock b/app/src-tauri/Cargo.lock
index 9b07a0ef3..e4906d677 100644
--- a/app/src-tauri/Cargo.lock
+++ b/app/src-tauri/Cargo.lock
@@ -4,7 +4,7 @@ version = 4
 
 [[package]]
 name = "OpenHuman"
-version = "0.49.31"
+version = "0.49.32"
 dependencies = [
  "env_logger",
  "log",
diff --git a/app/src/components/settings/SettingsHome.tsx b/app/src/components/settings/SettingsHome.tsx
index f97329adf..253a9b030 100644
--- a/app/src/components/settings/SettingsHome.tsx
+++ b/app/src/components/settings/SettingsHome.tsx
@@ -271,6 +271,23 @@ const SettingsHome = () => {
     //   onClick: handleViewEncryptionKey,
     //   dangerous: false,
     // },
+    {
+      id: 'local-model',
+      title: 'Local AI Model',
+      description: 'Choose model tier by device capability and manage downloads',
+      icon: (
+        <svg className="w-5 h-5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
+          <path
+            strokeLinecap="round"
+            strokeLinejoin="round"
+            strokeWidth={2}
+            d="M9 3v2m6-2v2M9 19v2m6-2v2M5 9H3m2 6H3m18-6h-2m2 6h-2M7 19h10a2 2 0 002-2V7a2 2 0 00-2-2H7a2 2 0 00-2 2v10a2 2 0 002 2zM9 9h6v6H9V9z"
+          />
+        </svg>
+      ),
+      onClick: () => navigateToSettings('local-model'),
+      dangerous: false,
+    },
     {
       id: 'team',
       title: 'Team',
diff --git a/app/src/components/settings/panels/DeveloperOptionsPanel.tsx b/app/src/components/settings/panels/DeveloperOptionsPanel.tsx
index 40d61275c..89fe40470 100644
--- a/app/src/components/settings/panels/DeveloperOptionsPanel.tsx
+++ b/app/src/components/settings/panels/DeveloperOptionsPanel.tsx
@@ -35,22 +35,6 @@ const developerItems = [
       </svg>
     ),
   },
-  {
-    id: 'local-model',
-    title: 'Local Model Runtime',
-    description: 'Monitor download/load status and test local model calls',
-    route: 'local-model',
-    icon: (
-      <svg className="w-5 h-5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
-        <path
-          strokeLinecap="round"
-          strokeLinejoin="round"
-          strokeWidth={2}
-          d="M9 17v-6m3 6V7m3 10v-4m5 6H4a2 2 0 01-2-2V7a2 2 0 012-2h16a2 2 0 012 2v10a2 2 0 01-2 2z"
-        />
-      </svg>
-    ),
-  },
   {
     id: 'tauri-commands',
     title: 'Tauri Command Console',
diff --git a/app/src/components/settings/panels/LocalModelPanel.tsx b/app/src/components/settings/panels/LocalModelPanel.tsx
index 24536e3ea..90a801b8f 100644
--- a/app/src/components/settings/panels/LocalModelPanel.tsx
+++ b/app/src/components/settings/panels/LocalModelPanel.tsx
@@ -1,6 +1,7 @@
 import { useEffect, useMemo, useState } from 'react';
 
 import {
+  type ApplyPresetResult,
   type LocalAiAssetsStatus,
   type LocalAiDownloadsProgress,
   type LocalAiEmbeddingResult,
@@ -8,12 +9,14 @@ import {
   type LocalAiStatus,
   type LocalAiSuggestion,
   type LocalAiTtsResult,
+  openhumanLocalAiApplyPreset,
   openhumanLocalAiAssetsStatus,
   openhumanLocalAiDownload,
   openhumanLocalAiDownloadAllAssets,
   openhumanLocalAiDownloadAsset,
   openhumanLocalAiDownloadsProgress,
   openhumanLocalAiEmbed,
+  openhumanLocalAiPresets,
   openhumanLocalAiPrompt,
   openhumanLocalAiStatus,
   openhumanLocalAiSuggestQuestions,
@@ -21,6 +24,7 @@ import {
   openhumanLocalAiTranscribe,
   openhumanLocalAiTts,
   openhumanLocalAiVisionPrompt,
+  type PresetsResponse,
 } from '../../../utils/tauriCommands';
 import SettingsHeader from '../components/SettingsHeader';
 import { useSettingsNavigation } from '../hooks/useSettingsNavigation';
@@ -148,6 +152,13 @@ const LocalModelPanel = () => {
   const [ttsOutput, setTtsOutput] = useState<LocalAiTtsResult | null>(null);
   const [isTtsLoading, setIsTtsLoading] = useState(false);
 
+  const [presetsData, setPresetsData] = useState<PresetsResponse | null>(null);
+  const [presetsLoading, setPresetsLoading] = useState(true);
+  const [isApplyingPreset, setIsApplyingPreset] = useState(false);
+  const [presetError, setPresetError] = useState('');
+  const [presetSuccess, setPresetSuccess] = useState<ApplyPresetResult | null>(null);
+  const [showAdvanced, setShowAdvanced] = useState(false);
+
   const progress = useMemo(() => {
     const downloadProgress = progressFromDownloads(downloads);
     if (downloadProgress != null) return downloadProgress;
@@ -189,8 +200,41 @@ const LocalModelPanel = () => {
     }
   };
 
+  const loadPresets = async () => {
+    setPresetsLoading(true);
+    try {
+      const data = await openhumanLocalAiPresets();
+      setPresetsData(data);
+      setPresetError('');
+    } catch (err) {
+      const msg = err instanceof Error ? err.message : 'Failed to load presets';
+      console.warn('[LocalModelPanel] failed to load presets:', msg);
+      setPresetError(msg);
+    } finally {
+      setPresetsLoading(false);
+    }
+  };
+
+  const applyPreset = async (tier: string) => {
+    setIsApplyingPreset(true);
+    setPresetError('');
+    setPresetSuccess(null);
+    try {
+      const result = await openhumanLocalAiApplyPreset(tier);
+      setPresetSuccess(result);
+      await loadPresets();
+      await loadStatus();
+    } catch (err) {
+      const msg = err instanceof Error ? err.message : 'Failed to apply preset';
+      setPresetError(msg);
+    } finally {
+      setIsApplyingPreset(false);
+    }
+  };
+
   useEffect(() => {
     void loadStatus();
+    void loadPresets();
     const timer = setInterval(() => {
       void loadStatus();
     }, 1500);
@@ -360,356 +404,497 @@ const LocalModelPanel = () => {
     }
   };
 
+  const formatRamGb = (bytes: number): string => {
+    const gb = bytes / (1024 * 1024 * 1024);
+    return gb >= 10 ? `${Math.round(gb)} GB` : `${gb.toFixed(1)} GB`;
+  };
+
   return (
     <div className="h-full flex flex-col">
       <SettingsHeader title="Local Model" showBackButton={true} onBack={navigateBack} />
 
       <div className="flex-1 overflow-y-auto px-6 pb-10 space-y-6">
+        {/* --- Model Tier Selection --- */}
         <section className="space-y-3">
-          <div className="flex items-center justify-between">
-            <h3 className="text-lg font-semibold text-white">Runtime Status</h3>
-            <button
-              onClick={() => void loadStatus()}
-              className="text-sm text-blue-400 hover:text-blue-300 transition-colors">
-              Refresh
-            </button>
-          </div>
+          <h3 className="text-lg font-semibold text-white">Model Tier</h3>
 
-          <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
-            <div className="flex items-center justify-between text-sm">
-              <span className="text-gray-400">State</span>
-              <span className={`font-medium ${statusTone(status?.state ?? 'idle')}`}>
-                {status ? statusLabel(downloads?.state ?? status.state) : 'Unavailable'}
-              </span>
+          {/* Loading / error states */}
+          {presetsLoading && !presetsData && (
+            <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 text-sm text-stone-400 animate-pulse">
+              Loading device info and presets…
             </div>
-
-            <div className="h-2 rounded-full bg-stone-800 overflow-hidden">
-              <div
-                className={`h-full bg-gradient-to-r from-blue-500 to-cyan-400 transition-all duration-500 ${
-                  isIndeterminateDownload ? 'animate-pulse' : ''
-                }`}
-                style={{ width: `${Math.round((isIndeterminateDownload ? 1 : progress) * 100)}%` }}
-              />
+          )}
+          {!presetsLoading && !presetsData && presetError && (
+            <div className="bg-gray-900 rounded-lg border border-red-700/40 p-4 text-sm text-red-300">
+              Could not load presets: {presetError}
             </div>
+          )}
 
-            <div className="flex flex-wrap items-center gap-x-3 gap-y-1 text-xs text-stone-400">
-              <span>
-                Progress:{' '}
-                {isIndeterminateDownload
-                  ? 'Downloading (size unknown)'
-                  : `${Math.round(progress * 100)}%`}
-              </span>
-              {downloadedText && <span className="text-stone-300">{downloadedText}</span>}
-              {speedText && <span className="text-blue-300">{speedText}</span>}
-              {etaText && <span className="text-cyan-300">ETA {etaText}</span>}
-            </div>
-
-            <div className="grid grid-cols-1 sm:grid-cols-2 gap-3 text-sm">
-              <div className="rounded-md border border-gray-700 p-2">
-                <div className="text-stone-400 text-xs uppercase tracking-wide">Provider</div>
-                <div className="text-stone-100 mt-1">{status?.provider ?? 'n/a'}</div>
-              </div>
-              <div className="rounded-md border border-gray-700 p-2">
-                <div className="text-stone-400 text-xs uppercase tracking-wide">Model</div>
-                <div className="text-stone-100 mt-1">{status?.model_id ?? 'n/a'}</div>
-              </div>
-            </div>
-
-            <div className="grid grid-cols-1 sm:grid-cols-3 gap-3 text-sm">
-              <div className="rounded-md border border-gray-700 p-2">
-                <div className="text-stone-400 text-xs uppercase tracking-wide">Backend</div>
-                <div className="text-stone-100 mt-1">{status?.active_backend ?? 'cpu'}</div>
-              </div>
-              <div className="rounded-md border border-gray-700 p-2">
-                <div className="text-stone-400 text-xs uppercase tracking-wide">Last Latency</div>
-                <div className="text-stone-100 mt-1">
-                  {typeof status?.last_latency_ms === 'number'
-                    ? `${status.last_latency_ms} ms`
-                    : 'n/a'}
-                </div>
-              </div>
-              <div className="rounded-md border border-gray-700 p-2">
-                <div className="text-stone-400 text-xs uppercase tracking-wide">Generation TPS</div>
-                <div className="text-stone-100 mt-1">
-                  {typeof status?.gen_toks_per_sec === 'number'
-                    ? `${status.gen_toks_per_sec.toFixed(1)} tok/s`
-                    : 'n/a'}
-                </div>
-              </div>
-            </div>
-
-            {status?.model_path && (
-              <div className="text-xs text-stone-400 break-all">Artifact: {status.model_path}</div>
-            )}
-
-            {status?.backend_reason && (
-              <div className="text-xs text-blue-300">{status.backend_reason}</div>
-            )}
-            {status?.warning && <div className="text-xs text-amber-300">{status.warning}</div>}
-            {statusError && <div className="text-xs text-red-300">{statusError}</div>}
-
-            <div className="flex items-center gap-2 pt-1">
-              <button
-                onClick={() => void triggerDownload(false)}
-                disabled={isTriggeringDownload}
-                className="px-3 py-1.5 text-xs rounded-md bg-blue-600 hover:bg-blue-700 disabled:opacity-60 text-white">
-                {isTriggeringDownload ? 'Triggering...' : 'Bootstrap / Resume'}
-              </button>
-              <button
-                onClick={() => void triggerDownload(true)}
-                disabled={isTriggeringDownload}
-                className="px-3 py-1.5 text-xs rounded-md border border-gray-600 hover:border-gray-500 disabled:opacity-60 text-stone-200">
-                Force Re-bootstrap
-              </button>
-            </div>
-          </div>
-        </section>
-
-        <section className="space-y-3">
-          <h3 className="text-lg font-semibold text-white">Capability Assets</h3>
-          <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
-            <div className="text-xs text-stone-400">
-              Quantization preference: {assets?.quantization ?? 'q4'}
-            </div>
-            <div className="grid grid-cols-1 sm:grid-cols-2 gap-3 text-sm">
-              {[
-                { label: 'Chat', key: 'chat' as const, item: assets?.chat },
-                { label: 'Vision', key: 'vision' as const, item: assets?.vision },
-                { label: 'Embedding', key: 'embedding' as const, item: assets?.embedding },
-                { label: 'STT', key: 'stt' as const, item: assets?.stt },
-                { label: 'TTS', key: 'tts' as const, item: assets?.tts },
-              ].map(({ label, key, item }) => (
-                <div key={String(label)} className="rounded-md border border-gray-700 p-2">
-                  <div className="text-stone-400 text-xs uppercase tracking-wide">{label}</div>
-                  <div className="text-stone-100 mt-1 break-all">{item?.id ?? 'n/a'}</div>
-                  <div className={`text-xs mt-1 ${statusTone(item?.state ?? 'idle')}`}>
-                    {statusLabel(item?.state ?? 'idle')}
+          {/* Device info */}
+          {presetsData?.device && (
+            <div className="bg-gray-900 rounded-lg border border-gray-700 p-3">
+              <div className="grid grid-cols-3 gap-3 text-xs">
+                <div>
+                  <div className="text-stone-400 uppercase tracking-wide">RAM</div>
+                  <div className="text-stone-100 mt-0.5 font-medium">
+                    {formatRamGb(presetsData.device.total_ram_bytes)}
                   </div>
-                  {item?.path && (
-                    <div className="text-[10px] text-stone-500 mt-1 break-all">{item.path}</div>
-                  )}
-                  <button
-                    onClick={() => void triggerAssetDownload(key)}
-                    disabled={assetDownloadBusy[key]}
-                    className="mt-2 px-2 py-1 text-[10px] rounded border border-gray-600 hover:border-gray-500 disabled:opacity-60 text-stone-200">
-                    {assetDownloadBusy[key] ? 'Downloading...' : 'Download'}
-                  </button>
                 </div>
-              ))}
-            </div>
-          </div>
-        </section>
-
-        <section className="space-y-3">
-          <h3 className="text-lg font-semibold text-white">Test Summarization</h3>
-          <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
-            <textarea
-              value={summaryInput}
-              onChange={e => setSummaryInput(e.target.value)}
-              placeholder="Paste text to summarize with the local model..."
-              className="w-full min-h-28 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
-            />
-            <div className="flex items-center justify-between">
-              <div className="text-xs text-stone-400">
-                Calls `openhuman.local_ai_summarize` via Rust core
-              </div>
-              <button
-                onClick={() => void runSummaryTest()}
-                disabled={isSummaryLoading || !summaryInput.trim()}
-                className="px-3 py-1.5 text-xs rounded-md bg-emerald-600 hover:bg-emerald-700 disabled:opacity-60 text-white">
-                {isSummaryLoading ? 'Running...' : 'Run Summary Test'}
-              </button>
-            </div>
-            {summaryOutput && (
-              <pre className="whitespace-pre-wrap rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200">
-                {summaryOutput}
-              </pre>
-            )}
-          </div>
-        </section>
-
-        <section className="space-y-3">
-          <h3 className="text-lg font-semibold text-white">Test Suggested Prompts</h3>
-          <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
-            <textarea
-              value={suggestInput}
-              onChange={e => setSuggestInput(e.target.value)}
-              placeholder="Paste conversation context to generate suggestions..."
-              className="w-full min-h-28 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
-            />
-            <div className="flex items-center justify-between">
-              <div className="text-xs text-stone-400">
-                Calls `openhuman.local_ai_suggest_questions` via Rust core
-              </div>
-              <button
-                onClick={() => void runSuggestTest()}
-                disabled={isSuggestLoading || !suggestInput.trim()}
-                className="px-3 py-1.5 text-xs rounded-md bg-cyan-600 hover:bg-cyan-700 disabled:opacity-60 text-white">
-                {isSuggestLoading ? 'Running...' : 'Run Suggestion Test'}
-              </button>
-            </div>
-
-            {suggestions.length > 0 && (
-              <div className="space-y-2">
-                {suggestions.map(suggestion => (
+                <div>
+                  <div className="text-stone-400 uppercase tracking-wide">CPU</div>
                   <div
-                    key={`${suggestion.text}-${suggestion.confidence}`}
-                    className="rounded-md border border-gray-700 bg-stone-950 p-3">
-                    <div className="text-sm text-stone-100">{suggestion.text}</div>
-                    <div className="text-xs text-stone-500 mt-1">
-                      Confidence: {(suggestion.confidence * 100).toFixed(0)}%
+                    className="text-stone-100 mt-0.5 font-medium truncate"
+                    title={presetsData.device.cpu_brand}>
+                    {presetsData.device.cpu_count} cores
+                  </div>
+                </div>
+                <div>
+                  <div className="text-stone-400 uppercase tracking-wide">GPU</div>
+                  <div
+                    className="text-stone-100 mt-0.5 font-medium truncate"
+                    title={presetsData.device.gpu_description ?? undefined}>
+                    {presetsData.device.has_gpu
+                      ? (presetsData.device.gpu_description ?? 'Detected')
+                      : 'Not detected'}
+                  </div>
+                </div>
+              </div>
+            </div>
+          )}
+
+          {/* Tier cards */}
+          {presetsData && (
+            <div className="space-y-2">
+              {presetsData.presets.map(preset => {
+                const isRecommended = preset.tier === presetsData.recommended_tier;
+                const isCurrent = preset.tier === presetsData.current_tier;
+                return (
+                  <button
+                    key={preset.tier}
+                    type="button"
+                    onClick={() => void applyPreset(preset.tier)}
+                    disabled={isApplyingPreset || isCurrent}
+                    className={`w-full text-left rounded-lg border p-3 transition-colors ${
+                      isCurrent
+                        ? 'border-blue-500 bg-blue-500/10'
+                        : 'border-gray-700 bg-gray-900 hover:border-gray-500'
+                    } ${isApplyingPreset ? 'opacity-60' : ''}`}>
+                    <div className="flex items-center justify-between">
+                      <div className="flex items-center gap-2">
+                        <span className="text-sm font-semibold text-white">{preset.label}</span>
+                        {isRecommended && (
+                          <span className="px-1.5 py-0.5 text-[10px] font-medium rounded bg-emerald-600/30 text-emerald-300 uppercase tracking-wide">
+                            Recommended
+                          </span>
+                        )}
+                        {isCurrent && (
+                          <span className="px-1.5 py-0.5 text-[10px] font-medium rounded bg-blue-600/30 text-blue-300 uppercase tracking-wide">
+                            Active
+                          </span>
+                        )}
+                      </div>
+                      <span className="text-xs text-stone-400">
+                        ~{preset.approx_download_gb} GB
+                      </span>
+                    </div>
+                    <div className="text-xs text-stone-400 mt-1">{preset.description}</div>
+                    <div className="text-[10px] text-stone-500 mt-1">
+                      Chat: {preset.chat_model_id} &middot; Min RAM: {preset.min_ram_gb} GB
+                    </div>
+                  </button>
+                );
+              })}
+
+              {presetsData.current_tier === 'custom' && (
+                <div className="rounded-lg border border-amber-600/30 bg-amber-600/5 p-3 text-xs text-amber-300">
+                  You are using custom model IDs that do not match any built-in preset.
+                </div>
+              )}
+            </div>
+          )}
+
+          {presetError && <div className="text-xs text-red-300">{presetError}</div>}
+          {presetSuccess && (
+            <div className="text-xs text-green-300">
+              Applied {presetSuccess.applied_tier} tier: {presetSuccess.chat_model_id}
+            </div>
+          )}
+        </section>
+
+        {/* Advanced toggle */}
+        <button
+          type="button"
+          onClick={() => setShowAdvanced(prev => !prev)}
+          className="flex items-center gap-2 text-sm text-stone-400 hover:text-stone-200 transition-colors">
+          <svg
+            className={`w-4 h-4 transition-transform ${showAdvanced ? 'rotate-90' : ''}`}
+            fill="none"
+            stroke="currentColor"
+            viewBox="0 0 24 24">
+            <path strokeLinecap="round" strokeLinejoin="round" strokeWidth={2} d="M9 5l7 7-7 7" />
+          </svg>
+          {showAdvanced ? 'Hide Advanced' : 'Show Advanced'}
+        </button>
+
+        {showAdvanced && (
+          <>
+            <section className="space-y-3">
+              <div className="flex items-center justify-between">
+                <h3 className="text-lg font-semibold text-white">Runtime Status</h3>
+                <button
+                  onClick={() => void loadStatus()}
+                  className="text-sm text-blue-400 hover:text-blue-300 transition-colors">
+                  Refresh
+                </button>
+              </div>
+
+              <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
+                <div className="flex items-center justify-between text-sm">
+                  <span className="text-gray-400">State</span>
+                  <span className={`font-medium ${statusTone(status?.state ?? 'idle')}`}>
+                    {status ? statusLabel(downloads?.state ?? status.state) : 'Unavailable'}
+                  </span>
+                </div>
+
+                <div className="h-2 rounded-full bg-stone-800 overflow-hidden">
+                  <div
+                    className={`h-full bg-gradient-to-r from-blue-500 to-cyan-400 transition-all duration-500 ${
+                      isIndeterminateDownload ? 'animate-pulse' : ''
+                    }`}
+                    style={{
+                      width: `${Math.round((isIndeterminateDownload ? 1 : progress) * 100)}%`,
+                    }}
+                  />
+                </div>
+
+                <div className="flex flex-wrap items-center gap-x-3 gap-y-1 text-xs text-stone-400">
+                  <span>
+                    Progress:{' '}
+                    {isIndeterminateDownload
+                      ? 'Downloading (size unknown)'
+                      : `${Math.round(progress * 100)}%`}
+                  </span>
+                  {downloadedText && <span className="text-stone-300">{downloadedText}</span>}
+                  {speedText && <span className="text-blue-300">{speedText}</span>}
+                  {etaText && <span className="text-cyan-300">ETA {etaText}</span>}
+                </div>
+
+                <div className="grid grid-cols-1 sm:grid-cols-2 gap-3 text-sm">
+                  <div className="rounded-md border border-gray-700 p-2">
+                    <div className="text-stone-400 text-xs uppercase tracking-wide">Provider</div>
+                    <div className="text-stone-100 mt-1">{status?.provider ?? 'n/a'}</div>
+                  </div>
+                  <div className="rounded-md border border-gray-700 p-2">
+                    <div className="text-stone-400 text-xs uppercase tracking-wide">Model</div>
+                    <div className="text-stone-100 mt-1">{status?.model_id ?? 'n/a'}</div>
+                  </div>
+                </div>
+
+                <div className="grid grid-cols-1 sm:grid-cols-3 gap-3 text-sm">
+                  <div className="rounded-md border border-gray-700 p-2">
+                    <div className="text-stone-400 text-xs uppercase tracking-wide">Backend</div>
+                    <div className="text-stone-100 mt-1">{status?.active_backend ?? 'cpu'}</div>
+                  </div>
+                  <div className="rounded-md border border-gray-700 p-2">
+                    <div className="text-stone-400 text-xs uppercase tracking-wide">
+                      Last Latency
+                    </div>
+                    <div className="text-stone-100 mt-1">
+                      {typeof status?.last_latency_ms === 'number'
+                        ? `${status.last_latency_ms} ms`
+                        : 'n/a'}
                     </div>
                   </div>
-                ))}
-              </div>
-            )}
-          </div>
-        </section>
+                  <div className="rounded-md border border-gray-700 p-2">
+                    <div className="text-stone-400 text-xs uppercase tracking-wide">
+                      Generation TPS
+                    </div>
+                    <div className="text-stone-100 mt-1">
+                      {typeof status?.gen_toks_per_sec === 'number'
+                        ? `${status.gen_toks_per_sec.toFixed(1)} tok/s`
+                        : 'n/a'}
+                    </div>
+                  </div>
+                </div>
 
-        <section className="space-y-3">
-          <h3 className="text-lg font-semibold text-white">Test Custom Prompt</h3>
-          <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
-            <textarea
-              value={promptInput}
-              onChange={e => setPromptInput(e.target.value)}
-              placeholder="Type any prompt and run it against the local model..."
-              className="w-full min-h-28 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
-            />
-            <div className="flex flex-wrap items-center justify-between gap-2">
-              <label className="flex items-center gap-2 text-xs text-stone-300">
-                <input
-                  type="checkbox"
-                  checked={promptNoThink}
-                  onChange={e => setPromptNoThink(e.target.checked)}
-                  className="h-3.5 w-3.5 rounded border-gray-600 bg-stone-900 text-blue-500 focus:ring-blue-500"
+                {status?.model_path && (
+                  <div className="text-xs text-stone-400 break-all">
+                    Artifact: {status.model_path}
+                  </div>
+                )}
+
+                {status?.backend_reason && (
+                  <div className="text-xs text-blue-300">{status.backend_reason}</div>
+                )}
+                {status?.warning && <div className="text-xs text-amber-300">{status.warning}</div>}
+                {statusError && <div className="text-xs text-red-300">{statusError}</div>}
+
+                <div className="flex items-center gap-2 pt-1">
+                  <button
+                    onClick={() => void triggerDownload(false)}
+                    disabled={isTriggeringDownload}
+                    className="px-3 py-1.5 text-xs rounded-md bg-blue-600 hover:bg-blue-700 disabled:opacity-60 text-white">
+                    {isTriggeringDownload ? 'Triggering...' : 'Bootstrap / Resume'}
+                  </button>
+                  <button
+                    onClick={() => void triggerDownload(true)}
+                    disabled={isTriggeringDownload}
+                    className="px-3 py-1.5 text-xs rounded-md border border-gray-600 hover:border-gray-500 disabled:opacity-60 text-stone-200">
+                    Force Re-bootstrap
+                  </button>
+                </div>
+              </div>
+            </section>
+
+            <section className="space-y-3">
+              <h3 className="text-lg font-semibold text-white">Capability Assets</h3>
+              <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
+                <div className="text-xs text-stone-400">
+                  Quantization preference: {assets?.quantization ?? 'q4'}
+                </div>
+                <div className="grid grid-cols-1 sm:grid-cols-2 gap-3 text-sm">
+                  {[
+                    { label: 'Chat', key: 'chat' as const, item: assets?.chat },
+                    { label: 'Vision', key: 'vision' as const, item: assets?.vision },
+                    { label: 'Embedding', key: 'embedding' as const, item: assets?.embedding },
+                    { label: 'STT', key: 'stt' as const, item: assets?.stt },
+                    { label: 'TTS', key: 'tts' as const, item: assets?.tts },
+                  ].map(({ label, key, item }) => (
+                    <div key={String(label)} className="rounded-md border border-gray-700 p-2">
+                      <div className="text-stone-400 text-xs uppercase tracking-wide">{label}</div>
+                      <div className="text-stone-100 mt-1 break-all">{item?.id ?? 'n/a'}</div>
+                      <div className={`text-xs mt-1 ${statusTone(item?.state ?? 'idle')}`}>
+                        {statusLabel(item?.state ?? 'idle')}
+                      </div>
+                      {item?.path && (
+                        <div className="text-[10px] text-stone-500 mt-1 break-all">{item.path}</div>
+                      )}
+                      <button
+                        onClick={() => void triggerAssetDownload(key)}
+                        disabled={assetDownloadBusy[key]}
+                        className="mt-2 px-2 py-1 text-[10px] rounded border border-gray-600 hover:border-gray-500 disabled:opacity-60 text-stone-200">
+                        {assetDownloadBusy[key] ? 'Downloading...' : 'Download'}
+                      </button>
+                    </div>
+                  ))}
+                </div>
+              </div>
+            </section>
+
+            <section className="space-y-3">
+              <h3 className="text-lg font-semibold text-white">Test Summarization</h3>
+              <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
+                <textarea
+                  value={summaryInput}
+                  onChange={e => setSummaryInput(e.target.value)}
+                  placeholder="Paste text to summarize with the local model..."
+                  className="w-full min-h-28 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
                 />
-                No-think mode
-              </label>
-              <button
-                onClick={() => void runPromptTest()}
-                disabled={isPromptLoading || !promptInput.trim()}
-                className="px-3 py-1.5 text-xs rounded-md bg-blue-600 hover:bg-blue-700 disabled:opacity-60 text-white">
-                {isPromptLoading ? 'Running...' : 'Run Prompt Test'}
-              </button>
-            </div>
-            <div className="text-xs text-stone-400">
-              Calls `openhuman.local_ai_prompt` via Rust core
-            </div>
-            {promptOutput && (
-              <pre className="whitespace-pre-wrap rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200">
-                {promptOutput}
-              </pre>
-            )}
-          </div>
-        </section>
-
-        <section className="space-y-3">
-          <h3 className="text-lg font-semibold text-white">Test Vision Prompt</h3>
-          <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
-            <textarea
-              value={visionPromptInput}
-              onChange={e => setVisionPromptInput(e.target.value)}
-              placeholder="Enter a prompt for the vision model..."
-              className="w-full min-h-20 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
-            />
-            <textarea
-              value={visionImageInput}
-              onChange={e => setVisionImageInput(e.target.value)}
-              placeholder="One image reference per line (data URI, URL, or local path marker)"
-              className="w-full min-h-20 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
-            />
-            <button
-              onClick={() => void runVisionTest()}
-              disabled={isVisionLoading || !visionPromptInput.trim() || !visionImageInput.trim()}
-              className="px-3 py-1.5 text-xs rounded-md bg-indigo-600 hover:bg-indigo-700 disabled:opacity-60 text-white">
-              {isVisionLoading ? 'Running...' : 'Run Vision Test'}
-            </button>
-            {visionOutput && (
-              <pre className="whitespace-pre-wrap rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200">
-                {visionOutput}
-              </pre>
-            )}
-          </div>
-        </section>
-
-        <section className="space-y-3">
-          <h3 className="text-lg font-semibold text-white">Test Embeddings</h3>
-          <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
-            <textarea
-              value={embeddingInput}
-              onChange={e => setEmbeddingInput(e.target.value)}
-              placeholder="One input string per line..."
-              className="w-full min-h-20 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
-            />
-            <button
-              onClick={() => void runEmbeddingTest()}
-              disabled={isEmbeddingLoading || !embeddingInput.trim()}
-              className="px-3 py-1.5 text-xs rounded-md bg-teal-600 hover:bg-teal-700 disabled:opacity-60 text-white">
-              {isEmbeddingLoading ? 'Running...' : 'Run Embedding Test'}
-            </button>
-            {embeddingOutput && (
-              <div className="rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200 space-y-1">
-                <div>Model: {embeddingOutput.model_id}</div>
-                <div>Dimensions: {embeddingOutput.dimensions}</div>
-                <div>Vectors: {embeddingOutput.vectors.length}</div>
+                <div className="flex items-center justify-between">
+                  <div className="text-xs text-stone-400">
+                    Calls `openhuman.local_ai_summarize` via Rust core
+                  </div>
+                  <button
+                    onClick={() => void runSummaryTest()}
+                    disabled={isSummaryLoading || !summaryInput.trim()}
+                    className="px-3 py-1.5 text-xs rounded-md bg-emerald-600 hover:bg-emerald-700 disabled:opacity-60 text-white">
+                    {isSummaryLoading ? 'Running...' : 'Run Summary Test'}
+                  </button>
+                </div>
+                {summaryOutput && (
+                  <pre className="whitespace-pre-wrap rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200">
+                    {summaryOutput}
+                  </pre>
+                )}
               </div>
-            )}
-          </div>
-        </section>
+            </section>
 
-        <section className="space-y-3">
-          <h3 className="text-lg font-semibold text-white">Test Voice Input (STT)</h3>
-          <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
-            <input
-              value={audioPathInput}
-              onChange={e => setAudioPathInput(e.target.value)}
-              placeholder="Absolute path to audio file"
-              className="w-full rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
-            />
-            <button
-              onClick={() => void runTranscribeTest()}
-              disabled={isTranscribeLoading || !audioPathInput.trim()}
-              className="px-3 py-1.5 text-xs rounded-md bg-purple-600 hover:bg-purple-700 disabled:opacity-60 text-white">
-              {isTranscribeLoading ? 'Running...' : 'Run Transcription Test'}
-            </button>
-            {transcribeOutput && (
-              <div className="rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200 space-y-1">
-                <div>Model: {transcribeOutput.model_id}</div>
-                <pre className="whitespace-pre-wrap">{transcribeOutput.text}</pre>
-              </div>
-            )}
-          </div>
-        </section>
+            <section className="space-y-3">
+              <h3 className="text-lg font-semibold text-white">Test Suggested Prompts</h3>
+              <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
+                <textarea
+                  value={suggestInput}
+                  onChange={e => setSuggestInput(e.target.value)}
+                  placeholder="Paste conversation context to generate suggestions..."
+                  className="w-full min-h-28 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
+                />
+                <div className="flex items-center justify-between">
+                  <div className="text-xs text-stone-400">
+                    Calls `openhuman.local_ai_suggest_questions` via Rust core
+                  </div>
+                  <button
+                    onClick={() => void runSuggestTest()}
+                    disabled={isSuggestLoading || !suggestInput.trim()}
+                    className="px-3 py-1.5 text-xs rounded-md bg-cyan-600 hover:bg-cyan-700 disabled:opacity-60 text-white">
+                    {isSuggestLoading ? 'Running...' : 'Run Suggestion Test'}
+                  </button>
+                </div>
 
-        <section className="space-y-3">
-          <h3 className="text-lg font-semibold text-white">Test Voice Output (TTS)</h3>
-          <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
-            <textarea
-              value={ttsInput}
-              onChange={e => setTtsInput(e.target.value)}
-              placeholder="Enter text to synthesize..."
-              className="w-full min-h-20 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
-            />
-            <input
-              value={ttsOutputPath}
-              onChange={e => setTtsOutputPath(e.target.value)}
-              placeholder="Optional output WAV path"
-              className="w-full rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
-            />
-            <button
-              onClick={() => void runTtsTest()}
-              disabled={isTtsLoading || !ttsInput.trim()}
-              className="px-3 py-1.5 text-xs rounded-md bg-rose-600 hover:bg-rose-700 disabled:opacity-60 text-white">
-              {isTtsLoading ? 'Running...' : 'Run TTS Test'}
-            </button>
-            {ttsOutput && (
-              <div className="rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200 space-y-1">
-                <div>Voice: {ttsOutput.voice_id}</div>
-                <div className="break-all">Output: {ttsOutput.output_path}</div>
+                {suggestions.length > 0 && (
+                  <div className="space-y-2">
+                    {suggestions.map(suggestion => (
+                      <div
+                        key={`${suggestion.text}-${suggestion.confidence}`}
+                        className="rounded-md border border-gray-700 bg-stone-950 p-3">
+                        <div className="text-sm text-stone-100">{suggestion.text}</div>
+                        <div className="text-xs text-stone-500 mt-1">
+                          Confidence: {(suggestion.confidence * 100).toFixed(0)}%
+                        </div>
+                      </div>
+                    ))}
+                  </div>
+                )}
               </div>
-            )}
-          </div>
-        </section>
+            </section>
+
+            <section className="space-y-3">
+              <h3 className="text-lg font-semibold text-white">Test Custom Prompt</h3>
+              <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
+                <textarea
+                  value={promptInput}
+                  onChange={e => setPromptInput(e.target.value)}
+                  placeholder="Type any prompt and run it against the local model..."
+                  className="w-full min-h-28 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
+                />
+                <div className="flex flex-wrap items-center justify-between gap-2">
+                  <label className="flex items-center gap-2 text-xs text-stone-300">
+                    <input
+                      type="checkbox"
+                      checked={promptNoThink}
+                      onChange={e => setPromptNoThink(e.target.checked)}
+                      className="h-3.5 w-3.5 rounded border-gray-600 bg-stone-900 text-blue-500 focus:ring-blue-500"
+                    />
+                    No-think mode
+                  </label>
+                  <button
+                    onClick={() => void runPromptTest()}
+                    disabled={isPromptLoading || !promptInput.trim()}
+                    className="px-3 py-1.5 text-xs rounded-md bg-blue-600 hover:bg-blue-700 disabled:opacity-60 text-white">
+                    {isPromptLoading ? 'Running...' : 'Run Prompt Test'}
+                  </button>
+                </div>
+                <div className="text-xs text-stone-400">
+                  Calls `openhuman.local_ai_prompt` via Rust core
+                </div>
+                {promptOutput && (
+                  <pre className="whitespace-pre-wrap rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200">
+                    {promptOutput}
+                  </pre>
+                )}
+              </div>
+            </section>
+
+            <section className="space-y-3">
+              <h3 className="text-lg font-semibold text-white">Test Vision Prompt</h3>
+              <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
+                <textarea
+                  value={visionPromptInput}
+                  onChange={e => setVisionPromptInput(e.target.value)}
+                  placeholder="Enter a prompt for the vision model..."
+                  className="w-full min-h-20 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
+                />
+                <textarea
+                  value={visionImageInput}
+                  onChange={e => setVisionImageInput(e.target.value)}
+                  placeholder="One image reference per line (data URI, URL, or local path marker)"
+                  className="w-full min-h-20 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
+                />
+                <button
+                  onClick={() => void runVisionTest()}
+                  disabled={
+                    isVisionLoading || !visionPromptInput.trim() || !visionImageInput.trim()
+                  }
+                  className="px-3 py-1.5 text-xs rounded-md bg-indigo-600 hover:bg-indigo-700 disabled:opacity-60 text-white">
+                  {isVisionLoading ? 'Running...' : 'Run Vision Test'}
+                </button>
+                {visionOutput && (
+                  <pre className="whitespace-pre-wrap rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200">
+                    {visionOutput}
+                  </pre>
+                )}
+              </div>
+            </section>
+
+            <section className="space-y-3">
+              <h3 className="text-lg font-semibold text-white">Test Embeddings</h3>
+              <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
+                <textarea
+                  value={embeddingInput}
+                  onChange={e => setEmbeddingInput(e.target.value)}
+                  placeholder="One input string per line..."
+                  className="w-full min-h-20 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
+                />
+                <button
+                  onClick={() => void runEmbeddingTest()}
+                  disabled={isEmbeddingLoading || !embeddingInput.trim()}
+                  className="px-3 py-1.5 text-xs rounded-md bg-teal-600 hover:bg-teal-700 disabled:opacity-60 text-white">
+                  {isEmbeddingLoading ? 'Running...' : 'Run Embedding Test'}
+                </button>
+                {embeddingOutput && (
+                  <div className="rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200 space-y-1">
+                    <div>Model: {embeddingOutput.model_id}</div>
+                    <div>Dimensions: {embeddingOutput.dimensions}</div>
+                    <div>Vectors: {embeddingOutput.vectors.length}</div>
+                  </div>
+                )}
+              </div>
+            </section>
+
+            <section className="space-y-3">
+              <h3 className="text-lg font-semibold text-white">Test Voice Input (STT)</h3>
+              <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
+                <input
+                  value={audioPathInput}
+                  onChange={e => setAudioPathInput(e.target.value)}
+                  placeholder="Absolute path to audio file"
+                  className="w-full rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
+                />
+                <button
+                  onClick={() => void runTranscribeTest()}
+                  disabled={isTranscribeLoading || !audioPathInput.trim()}
+                  className="px-3 py-1.5 text-xs rounded-md bg-purple-600 hover:bg-purple-700 disabled:opacity-60 text-white">
+                  {isTranscribeLoading ? 'Running...' : 'Run Transcription Test'}
+                </button>
+                {transcribeOutput && (
+                  <div className="rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200 space-y-1">
+                    <div>Model: {transcribeOutput.model_id}</div>
+                    <pre className="whitespace-pre-wrap">{transcribeOutput.text}</pre>
+                  </div>
+                )}
+              </div>
+            </section>
+
+            <section className="space-y-3">
+              <h3 className="text-lg font-semibold text-white">Test Voice Output (TTS)</h3>
+              <div className="bg-gray-900 rounded-lg border border-gray-700 p-4 space-y-3">
+                <textarea
+                  value={ttsInput}
+                  onChange={e => setTtsInput(e.target.value)}
+                  placeholder="Enter text to synthesize..."
+                  className="w-full min-h-20 rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
+                />
+                <input
+                  value={ttsOutputPath}
+                  onChange={e => setTtsOutputPath(e.target.value)}
+                  placeholder="Optional output WAV path"
+                  className="w-full rounded-md bg-stone-950 border border-gray-700 px-3 py-2 text-sm text-stone-100 placeholder:text-stone-500 focus:outline-none focus:ring-1 focus:ring-blue-500"
+                />
+                <button
+                  onClick={() => void runTtsTest()}
+                  disabled={isTtsLoading || !ttsInput.trim()}
+                  className="px-3 py-1.5 text-xs rounded-md bg-rose-600 hover:bg-rose-700 disabled:opacity-60 text-white">
+                  {isTtsLoading ? 'Running...' : 'Run TTS Test'}
+                </button>
+                {ttsOutput && (
+                  <div className="rounded-md bg-stone-950 border border-gray-700 p-3 text-xs text-stone-200 space-y-1">
+                    <div>Voice: {ttsOutput.voice_id}</div>
+                    <div className="break-all">Output: {ttsOutput.output_path}</div>
+                  </div>
+                )}
+              </div>
+            </section>
+          </>
+        )}
       </div>
     </div>
   );
diff --git a/app/src/components/skills/SkillSetupWizard.tsx b/app/src/components/skills/SkillSetupWizard.tsx
index 727b28536..50ab38a6f 100644
--- a/app/src/components/skills/SkillSetupWizard.tsx
+++ b/app/src/components/skills/SkillSetupWizard.tsx
@@ -53,7 +53,10 @@ export default function SkillSetupWizard({
       isConnected
     ) {
       setSetupComplete(skillId, true).catch(() => {});
-      setState({ phase: "complete", message: "Successfully connected!" });
+      // Schedule state update to avoid synchronous setState inside an effect
+      setTimeout(() => {
+        setState({ phase: "complete", message: "Successfully connected!" });
+      }, 0);
     }
   }, [isConnected, state.phase, skillId]);
 
diff --git a/app/src/utils/tauriCommands.ts b/app/src/utils/tauriCommands.ts
index 612b3b250..b45210b85 100644
--- a/app/src/utils/tauriCommands.ts
+++ b/app/src/utils/tauriCommands.ts
@@ -1294,6 +1294,60 @@ export async function openhumanLocalAiDownloadAsset(
   });
 }
 
+// --- Device profile & model tier presets ---
+
+export interface DeviceProfileResult {
+  total_ram_bytes: number;
+  cpu_count: number;
+  cpu_brand: string;
+  os_name: string;
+  os_version: string;
+  has_gpu: boolean;
+  gpu_description: string | null;
+}
+
+export interface ModelPresetResult {
+  tier: string;
+  label: string;
+  description: string;
+  chat_model_id: string;
+  vision_model_id: string;
+  embedding_model_id: string;
+  quantization: string;
+  min_ram_gb: number;
+  approx_download_gb: number;
+}
+
+export interface PresetsResponse {
+  presets: ModelPresetResult[];
+  recommended_tier: string;
+  current_tier: string;
+  device: DeviceProfileResult;
+}
+
+export interface ApplyPresetResult {
+  applied_tier: string;
+  chat_model_id: string;
+  vision_model_id: string;
+  embedding_model_id: string;
+  quantization: string;
+}
+
+export async function openhumanLocalAiDeviceProfile(): Promise<DeviceProfileResult> {
+  return await callCoreRpc<DeviceProfileResult>({ method: 'openhuman.local_ai_device_profile' });
+}
+
+export async function openhumanLocalAiPresets(): Promise<PresetsResponse> {
+  return await callCoreRpc<PresetsResponse>({ method: 'openhuman.local_ai_presets' });
+}
+
+export async function openhumanLocalAiApplyPreset(tier: string): Promise<ApplyPresetResult> {
+  return await callCoreRpc<ApplyPresetResult>({
+    method: 'openhuman.local_ai_apply_preset',
+    params: { tier },
+  });
+}
+
 export async function aiGetConfig(): Promise<AIPreview> {
   return {
     soul: {
diff --git a/src/openhuman/config/schema/load.rs b/src/openhuman/config/schema/load.rs
index c5ccc0dfe..9af009ab9 100644
--- a/src/openhuman/config/schema/load.rs
+++ b/src/openhuman/config/schema/load.rs
@@ -570,6 +570,28 @@ impl Config {
             self.proxy.enabled = false;
         }
 
+        if let Ok(tier_str) = std::env::var("OPENHUMAN_LOCAL_AI_TIER") {
+            let tier_str = tier_str.trim().to_ascii_lowercase();
+            if !tier_str.is_empty() {
+                if let Some(tier) =
+                    crate::openhuman::local_ai::presets::ModelTier::from_str_opt(&tier_str)
+                {
+                    if tier != crate::openhuman::local_ai::presets::ModelTier::Custom {
+                        crate::openhuman::local_ai::presets::apply_preset_to_config(
+                            &mut self.local_ai,
+                            tier,
+                        );
+                        tracing::debug!(tier = %tier_str, "applied local AI tier from OPENHUMAN_LOCAL_AI_TIER");
+                    }
+                } else {
+                    tracing::warn!(
+                        tier = %tier_str,
+                        "ignoring invalid OPENHUMAN_LOCAL_AI_TIER (valid: low, medium, high)"
+                    );
+                }
+            }
+        }
+
         if self.proxy.enabled && self.proxy.scope == ProxyScope::Environment {
             self.proxy.apply_to_process_env();
         }
diff --git a/src/openhuman/config/schema/local_ai.rs b/src/openhuman/config/schema/local_ai.rs
index b24116ef1..9fa7de03e 100644
--- a/src/openhuman/config/schema/local_ai.rs
+++ b/src/openhuman/config/schema/local_ai.rs
@@ -49,6 +49,8 @@ pub struct LocalAiConfig {
     pub context_compaction_threshold_tokens: usize,
     #[serde(default = "default_max_suggestions")]
     pub max_suggestions: usize,
+    #[serde(default)]
+    pub selected_tier: Option<String>,
 }
 
 fn default_enabled() -> bool {
@@ -169,6 +171,7 @@ impl Default for LocalAiConfig {
             autosummary_debounce_ms: default_autosummary_debounce_ms(),
             context_compaction_threshold_tokens: default_context_compaction_threshold_tokens(),
             max_suggestions: default_max_suggestions(),
+            selected_tier: None,
         }
     }
 }
diff --git a/src/openhuman/local_ai/device.rs b/src/openhuman/local_ai/device.rs
new file mode 100644
index 000000000..ca7cc0bfb
--- /dev/null
+++ b/src/openhuman/local_ai/device.rs
@@ -0,0 +1,111 @@
+//! Device profile detection for guided model selection.
+
+use serde::{Deserialize, Serialize};
+use sysinfo::System;
+
+/// Summary of local hardware relevant for model tier selection.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct DeviceProfile {
+    pub total_ram_bytes: u64,
+    pub cpu_count: usize,
+    pub cpu_brand: String,
+    pub os_name: String,
+    pub os_version: String,
+    pub has_gpu: bool,
+    pub gpu_description: Option<String>,
+}
+
+impl DeviceProfile {
+    /// Total RAM expressed in whole gigabytes (rounded down).
+    pub fn total_ram_gb(&self) -> u64 {
+        self.total_ram_bytes / (1024 * 1024 * 1024)
+    }
+}
+
+/// Probe the current machine and return a [`DeviceProfile`].
+///
+/// GPU detection is best-effort: Apple Silicon is assumed to have a GPU (Metal);
+/// on other platforms we report "unknown" unless more specific probing is added later.
+pub fn detect_device_profile() -> DeviceProfile {
+    let mut sys = System::new_all();
+    sys.refresh_all();
+
+    let total_ram_bytes = sys.total_memory();
+    let cpu_count = sys.cpus().len();
+    let cpu_brand = sys
+        .cpus()
+        .first()
+        .map(|c| c.brand().trim().to_string())
+        .unwrap_or_default();
+
+    let os_name = System::name().unwrap_or_else(|| "unknown".to_string());
+    let os_version = System::os_version().unwrap_or_else(|| "unknown".to_string());
+
+    let (has_gpu, gpu_description) = detect_gpu(&cpu_brand, &os_name);
+
+    tracing::debug!(
+        total_ram_bytes,
+        cpu_count,
+        cpu_brand = %cpu_brand,
+        os_name = %os_name,
+        os_version = %os_version,
+        has_gpu,
+        gpu_description = ?gpu_description,
+        "device profile detected"
+    );
+
+    DeviceProfile {
+        total_ram_bytes,
+        cpu_count,
+        cpu_brand,
+        os_name,
+        os_version,
+        has_gpu,
+        gpu_description,
+    }
+}
+
+/// Best-effort GPU detection.
+///
+/// Apple Silicon always has a unified GPU (Metal). On other systems we cannot
+/// reliably detect discrete GPUs without heavy platform-specific dependencies,
+/// so we conservatively report unknown.
+fn detect_gpu(cpu_brand: &str, os_name: &str) -> (bool, Option<String>) {
+    let brand_lower = cpu_brand.to_ascii_lowercase();
+    let os_lower = os_name.to_ascii_lowercase();
+
+    // Apple Silicon detection: brand contains "apple" or we're on macOS with an ARM chip
+    if brand_lower.contains("apple") || (os_lower.contains("mac") && brand_lower.contains("arm")) {
+        return (true, Some("Apple Silicon (Metal)".to_string()));
+    }
+
+    // Fallback: cannot reliably detect discrete GPU without platform libraries.
+    (false, None)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn detect_device_profile_returns_nonzero_hardware() {
+        let profile = detect_device_profile();
+        assert!(profile.total_ram_bytes > 0, "RAM should be > 0");
+        assert!(profile.cpu_count > 0, "CPU count should be > 0");
+        assert!(!profile.os_name.is_empty(), "OS name should be non-empty");
+    }
+
+    #[test]
+    fn total_ram_gb_rounds_down() {
+        let profile = DeviceProfile {
+            total_ram_bytes: 17_179_869_184, // 16 GiB exactly
+            cpu_count: 8,
+            cpu_brand: "test".to_string(),
+            os_name: "test".to_string(),
+            os_version: "1.0".to_string(),
+            has_gpu: false,
+            gpu_description: None,
+        };
+        assert_eq!(profile.total_ram_gb(), 16);
+    }
+}
diff --git a/src/openhuman/local_ai/mod.rs b/src/openhuman/local_ai/mod.rs
index 5982e8d83..350297813 100644
--- a/src/openhuman/local_ai/mod.rs
+++ b/src/openhuman/local_ai/mod.rs
@@ -1,7 +1,9 @@
 //! Bundled local AI stack (Ollama, whisper.cpp, Piper).
 
 mod core;
+pub mod device;
 pub mod ops;
+pub mod presets;
 mod schemas;
 
 mod install;
@@ -13,8 +15,10 @@ mod service;
 mod types;
 
 pub use core::*;
+pub use device::DeviceProfile;
 pub use ops as rpc;
 pub use ops::*;
+pub use presets::{ModelPreset, ModelTier};
 pub use schemas::{
     all_controller_schemas as all_local_ai_controller_schemas,
     all_registered_controllers as all_local_ai_registered_controllers,
diff --git a/src/openhuman/local_ai/presets.rs b/src/openhuman/local_ai/presets.rs
new file mode 100644
index 000000000..21503fd22
--- /dev/null
+++ b/src/openhuman/local_ai/presets.rs
@@ -0,0 +1,251 @@
+//! Tiered model presets and recommendation logic for local AI.
+//!
+//! # Tier → Model ID Mapping
+//!
+//! | Tier   | Chat / Vision Model       | Embedding Model            | Min RAM | ~Download |
+//! |--------|---------------------------|----------------------------|---------|-----------|
+//! | Low    | `gemma3:1b-it-q4_0`       | `nomic-embed-text:latest`  | 4 GB    | ~1 GB     |
+//! | Medium | `gemma3:4b-it-qat`        | `nomic-embed-text:latest`  | 8 GB    | ~3 GB     |
+//! | High   | `gemma3:12b-it-q4_K_M`    | `nomic-embed-text:latest`  | 16 GB   | ~8 GB     |
+//!
+//! # Changing defaults for a release
+//!
+//! Edit [`all_presets()`] below. Each `ModelPreset` defines:
+//! - `chat_model_id` / `vision_model_id` — Ollama tag for the chat and vision models.
+//! - `embedding_model_id` — Ollama tag for the embedding model.
+//! - `quantization` — quantization label shown in the UI.
+//! - `min_ram_gb` / `approx_download_gb` — user-facing guidance.
+//!
+//! After changing model tags, verify they exist in the Ollama library and update
+//! the `OPENHUMAN_LOCAL_AI_TIER` env var docs in `.env.example` if tier names change.
+
+use serde::{Deserialize, Serialize};
+
+use crate::openhuman::config::schema::LocalAiConfig;
+
+use super::device::DeviceProfile;
+
+/// Performance tier for local AI model selection.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(rename_all = "lowercase")]
+pub enum ModelTier {
+    Low,
+    Medium,
+    High,
+    Custom,
+}
+
+impl ModelTier {
+    pub fn as_str(&self) -> &'static str {
+        match self {
+            Self::Low => "low",
+            Self::Medium => "medium",
+            Self::High => "high",
+            Self::Custom => "custom",
+        }
+    }
+
+    pub fn from_str_opt(s: &str) -> Option<Self> {
+        match s.to_ascii_lowercase().as_str() {
+            "low" => Some(Self::Low),
+            "medium" => Some(Self::Medium),
+            "high" => Some(Self::High),
+            "custom" => Some(Self::Custom),
+            _ => None,
+        }
+    }
+}
+
+/// A concrete model preset tied to a performance tier.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ModelPreset {
+    pub tier: ModelTier,
+    pub label: &'static str,
+    pub description: &'static str,
+    pub chat_model_id: &'static str,
+    pub vision_model_id: &'static str,
+    pub embedding_model_id: &'static str,
+    pub quantization: &'static str,
+    pub min_ram_gb: u64,
+    pub approx_download_gb: f32,
+}
+
+/// Return all built-in presets (Low, Medium, High).
+pub fn all_presets() -> Vec<ModelPreset> {
+    vec![
+        ModelPreset {
+            tier: ModelTier::Low,
+            label: "Lightweight",
+            description: "Smallest footprint. Works on machines with 4 GB+ RAM.",
+            chat_model_id: "gemma3:1b-it-q4_0",
+            vision_model_id: "gemma3:1b-it-q4_0",
+            embedding_model_id: "nomic-embed-text:latest",
+            quantization: "q4_0",
+            min_ram_gb: 4,
+            approx_download_gb: 1.0,
+        },
+        ModelPreset {
+            tier: ModelTier::Medium,
+            label: "Balanced",
+            description: "Good quality with moderate resource use. Requires 8 GB+ RAM.",
+            chat_model_id: "gemma3:4b-it-qat",
+            vision_model_id: "gemma3:4b-it-qat",
+            embedding_model_id: "nomic-embed-text:latest",
+            quantization: "q4",
+            min_ram_gb: 8,
+            approx_download_gb: 3.0,
+        },
+        ModelPreset {
+            tier: ModelTier::High,
+            label: "Performance",
+            description: "Best quality. Requires 16 GB+ RAM and a capable GPU.",
+            chat_model_id: "gemma3:12b-it-q4_K_M",
+            vision_model_id: "gemma3:12b-it-q4_K_M",
+            embedding_model_id: "nomic-embed-text:latest",
+            quantization: "q4_K_M",
+            min_ram_gb: 16,
+            approx_download_gb: 8.0,
+        },
+    ]
+}
+
+/// Return the preset for a specific tier, or `None` for `Custom`.
+pub fn preset_for_tier(tier: ModelTier) -> Option<ModelPreset> {
+    all_presets().into_iter().find(|p| p.tier == tier)
+}
+
+/// Recommend a tier based on device capabilities.
+///
+/// * < 8 GB RAM  -> Low
+/// * 8 - 15 GB   -> Medium
+/// * >= 16 GB    -> High
+pub fn recommend_tier(device: &DeviceProfile) -> ModelTier {
+    let ram_gb = device.total_ram_gb();
+    let tier = if ram_gb >= 16 {
+        ModelTier::High
+    } else if ram_gb >= 8 {
+        ModelTier::Medium
+    } else {
+        ModelTier::Low
+    };
+    tracing::debug!(ram_gb, ?tier, "recommended model tier");
+    tier
+}
+
+/// Apply a preset to a [`LocalAiConfig`], overwriting model IDs, quantization,
+/// and the `selected_tier` marker.
+pub fn apply_preset_to_config(config: &mut LocalAiConfig, tier: ModelTier) {
+    if let Some(preset) = preset_for_tier(tier) {
+        tracing::debug!(
+            ?tier,
+            chat = preset.chat_model_id,
+            "applying preset to config"
+        );
+        config.model_id = preset.chat_model_id.to_string();
+        config.chat_model_id = preset.chat_model_id.to_string();
+        config.vision_model_id = preset.vision_model_id.to_string();
+        config.embedding_model_id = preset.embedding_model_id.to_string();
+        config.quantization = preset.quantization.to_string();
+        config.selected_tier = Some(tier.as_str().to_string());
+    } else {
+        tracing::debug!("apply_preset_to_config called for Custom tier; no-op");
+    }
+}
+
+/// Reverse-lookup the current tier from config. Returns `Custom` if none of the
+/// built-in presets match the current model IDs.
+pub fn current_tier_from_config(config: &LocalAiConfig) -> ModelTier {
+    // If a tier is explicitly stored, try to match it first.
+    if let Some(ref stored) = config.selected_tier {
+        if let Some(tier) = ModelTier::from_str_opt(stored) {
+            if tier == ModelTier::Custom {
+                return ModelTier::Custom;
+            }
+            // Verify the stored tier still matches the actual model IDs.
+            if let Some(preset) = preset_for_tier(tier) {
+                if config.chat_model_id == preset.chat_model_id
+                    && config.vision_model_id == preset.vision_model_id
+                    && config.embedding_model_id == preset.embedding_model_id
+                {
+                    return tier;
+                }
+            }
+        }
+    }
+
+    // Fallback: match by model IDs.
+    for preset in all_presets() {
+        if config.chat_model_id == preset.chat_model_id
+            && config.vision_model_id == preset.vision_model_id
+            && config.embedding_model_id == preset.embedding_model_id
+        {
+            return preset.tier;
+        }
+    }
+
+    ModelTier::Custom
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn recommend_tier_by_ram() {
+        let low_device = DeviceProfile {
+            total_ram_bytes: 4 * 1024 * 1024 * 1024, // 4 GB
+            cpu_count: 4,
+            cpu_brand: String::new(),
+            os_name: String::new(),
+            os_version: String::new(),
+            has_gpu: false,
+            gpu_description: None,
+        };
+        assert_eq!(recommend_tier(&low_device), ModelTier::Low);
+
+        let medium_device = DeviceProfile {
+            total_ram_bytes: 8 * 1024 * 1024 * 1024, // 8 GB
+            ..low_device.clone()
+        };
+        assert_eq!(recommend_tier(&medium_device), ModelTier::Medium);
+
+        let high_device = DeviceProfile {
+            total_ram_bytes: 32 * 1024 * 1024 * 1024_u64, // 32 GB
+            ..low_device.clone()
+        };
+        assert_eq!(recommend_tier(&high_device), ModelTier::High);
+    }
+
+    #[test]
+    fn preset_application_and_round_trip() {
+        let mut config = LocalAiConfig::default();
+        apply_preset_to_config(&mut config, ModelTier::Low);
+        assert_eq!(config.chat_model_id, "gemma3:1b-it-q4_0");
+        assert_eq!(config.selected_tier, Some("low".to_string()));
+        assert_eq!(current_tier_from_config(&config), ModelTier::Low);
+    }
+
+    #[test]
+    fn custom_detection_when_models_dont_match() {
+        let mut config = LocalAiConfig::default();
+        config.chat_model_id = "some-other-model:latest".to_string();
+        config.selected_tier = None;
+        assert_eq!(current_tier_from_config(&config), ModelTier::Custom);
+    }
+
+    #[test]
+    fn all_presets_returns_three_tiers() {
+        let presets = all_presets();
+        assert_eq!(presets.len(), 3);
+        assert_eq!(presets[0].tier, ModelTier::Low);
+        assert_eq!(presets[1].tier, ModelTier::Medium);
+        assert_eq!(presets[2].tier, ModelTier::High);
+    }
+
+    #[test]
+    fn default_config_maps_to_medium() {
+        let config = LocalAiConfig::default();
+        // Default config uses gemma3:4b-it-qat which is the Medium preset.
+        assert_eq!(current_tier_from_config(&config), ModelTier::Medium);
+    }
+}
diff --git a/src/openhuman/local_ai/schemas.rs b/src/openhuman/local_ai/schemas.rs
index 7ae1478d9..89d77e90a 100644
--- a/src/openhuman/local_ai/schemas.rs
+++ b/src/openhuman/local_ai/schemas.rs
@@ -89,6 +89,11 @@ struct LocalAiDownloadAssetParams {
     capability: String,
 }
 
+#[derive(Debug, Deserialize)]
+struct LocalAiApplyPresetParams {
+    tier: String,
+}
+
 pub fn all_controller_schemas() -> Vec<ControllerSchema> {
     vec![
         schemas("agent_chat"),
@@ -110,6 +115,9 @@ pub fn all_controller_schemas() -> Vec<ControllerSchema> {
         schemas("local_ai_assets_status"),
         schemas("local_ai_downloads_progress"),
         schemas("local_ai_download_asset"),
+        schemas("local_ai_device_profile"),
+        schemas("local_ai_presets"),
+        schemas("local_ai_apply_preset"),
     ]
 }
 
@@ -191,6 +199,18 @@ pub fn all_registered_controllers() -> Vec<RegisteredController> {
             schema: schemas("local_ai_download_asset"),
             handler: handle_local_ai_download_asset,
         },
+        RegisteredController {
+            schema: schemas("local_ai_device_profile"),
+            handler: handle_local_ai_device_profile,
+        },
+        RegisteredController {
+            schema: schemas("local_ai_presets"),
+            handler: handle_local_ai_presets,
+        },
+        RegisteredController {
+            schema: schemas("local_ai_apply_preset"),
+            handler: handle_local_ai_apply_preset,
+        },
     ]
 }
 
@@ -383,6 +403,30 @@ pub fn schemas(function: &str) -> ControllerSchema {
             inputs: vec![required_string("capability", "Asset capability id.")],
             outputs: vec![json_output("status", "Assets status payload.")],
         },
+        "local_ai_device_profile" => ControllerSchema {
+            namespace: "local_ai",
+            function: "device_profile",
+            description: "Detect local device hardware profile (RAM, CPU, GPU).",
+            inputs: vec![],
+            outputs: vec![json_output("profile", "Device hardware profile.")],
+        },
+        "local_ai_presets" => ControllerSchema {
+            namespace: "local_ai",
+            function: "presets",
+            description: "List model tier presets with recommendation and current selection.",
+            inputs: vec![],
+            outputs: vec![json_output(
+                "presets",
+                "Presets, recommended tier, current tier.",
+            )],
+        },
+        "local_ai_apply_preset" => ControllerSchema {
+            namespace: "local_ai",
+            function: "apply_preset",
+            description: "Apply a model tier preset to local AI config and persist.",
+            inputs: vec![required_string("tier", "Tier to apply: low, medium, high.")],
+            outputs: vec![json_output("result", "Applied tier status.")],
+        },
         _ => ControllerSchema {
             namespace: "local_ai",
             function: "unknown",
@@ -624,6 +668,77 @@ fn handle_local_ai_download_asset(params: Map<String, Value>) -> ControllerFutur
     })
 }
 
+fn handle_local_ai_device_profile(_params: Map<String, Value>) -> ControllerFuture {
+    Box::pin(async move {
+        tracing::debug!("[local_ai] device_profile: detecting hardware");
+        let profile = crate::openhuman::local_ai::device::detect_device_profile();
+        tracing::debug!("[local_ai] device_profile: done");
+        let value = serde_json::to_value(&profile).map_err(|e| format!("serialize: {e}"))?;
+        Ok(value)
+    })
+}
+
+fn handle_local_ai_presets(_params: Map<String, Value>) -> ControllerFuture {
+    Box::pin(async move {
+        tracing::debug!("[local_ai] presets: loading config and computing tiers");
+        let config = config_rpc::load_config_with_timeout().await?;
+        let device = crate::openhuman::local_ai::device::detect_device_profile();
+        let recommended = crate::openhuman::local_ai::presets::recommend_tier(&device);
+        let current =
+            crate::openhuman::local_ai::presets::current_tier_from_config(&config.local_ai);
+        let presets = crate::openhuman::local_ai::presets::all_presets();
+        tracing::debug!(
+            ?recommended,
+            ?current,
+            preset_count = presets.len(),
+            "[local_ai] presets: returning"
+        );
+        let value = serde_json::json!({
+            "presets": presets,
+            "recommended_tier": recommended,
+            "current_tier": current,
+            "device": device,
+        });
+        Ok(value)
+    })
+}
+
+fn handle_local_ai_apply_preset(params: Map<String, Value>) -> ControllerFuture {
+    Box::pin(async move {
+        let p = deserialize_params::<LocalAiApplyPresetParams>(params)?;
+        let tier_str = p.tier.trim().to_ascii_lowercase();
+        tracing::debug!(tier = %tier_str, "[local_ai] apply_preset: parsing tier");
+
+        let tier = crate::openhuman::local_ai::presets::ModelTier::from_str_opt(&tier_str)
+            .ok_or_else(|| {
+                format!(
+                    "invalid tier '{}': expected one of low, medium, high",
+                    tier_str
+                )
+            })?;
+
+        if tier == crate::openhuman::local_ai::presets::ModelTier::Custom {
+            return Err("cannot apply 'custom' tier; set model IDs directly".to_string());
+        }
+
+        let mut config = config_rpc::load_config_with_timeout().await?;
+        crate::openhuman::local_ai::presets::apply_preset_to_config(&mut config.local_ai, tier);
+        config
+            .save()
+            .await
+            .map_err(|e| format!("save config: {e}"))?;
+        tracing::debug!(tier = %tier_str, "[local_ai] apply_preset: config saved");
+
+        Ok(serde_json::json!({
+            "applied_tier": tier,
+            "chat_model_id": config.local_ai.chat_model_id,
+            "vision_model_id": config.local_ai.vision_model_id,
+            "embedding_model_id": config.local_ai.embedding_model_id,
+            "quantization": config.local_ai.quantization,
+        }))
+    })
+}
+
 fn deserialize_params<T: DeserializeOwned>(params: Map<String, Value>) -> Result<T, String> {
     serde_json::from_value(Value::Object(params)).map_err(|e| format!("invalid params: {e}"))
 }
diff --git a/src/openhuman/local_ai/service/bootstrap.rs b/src/openhuman/local_ai/service/bootstrap.rs
index 7adfd88ec..495b74e9d 100644
--- a/src/openhuman/local_ai/service/bootstrap.rs
+++ b/src/openhuman/local_ai/service/bootstrap.rs
@@ -121,14 +121,14 @@ impl LocalAiService {
         if let Err(err) = self.ensure_ollama_server(config).await {
             let mut status = self.status.lock();
             status.state = "degraded".to_string();
-            status.warning = Some(err);
+            status.warning = Some(format_degraded_warning(&err, config));
             return;
         }
 
         if let Err(err) = self.ensure_models_available(config).await {
             let mut status = self.status.lock();
             status.state = "degraded".to_string();
-            status.warning = Some(err);
+            status.warning = Some(format_degraded_warning(&err, config));
             return;
         }
 
@@ -180,6 +180,26 @@ impl LocalAiService {
     }
 }
 
+/// Append a tier step-down hint when the current tier is Medium or High.
+fn format_degraded_warning(err: &str, config: &Config) -> String {
+    let current = crate::openhuman::local_ai::presets::current_tier_from_config(&config.local_ai);
+    match current {
+        crate::openhuman::local_ai::presets::ModelTier::High => {
+            format!(
+                "{err}. Hint: your device may not support the High tier model. \
+                 Try switching to Medium or Low in Settings > Local AI Model."
+            )
+        }
+        crate::openhuman::local_ai::presets::ModelTier::Medium => {
+            format!(
+                "{err}. Hint: your device may not support the Medium tier model. \
+                 Try switching to Low in Settings > Local AI Model."
+            )
+        }
+        _ => err.to_string(),
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
diff --git a/tests/json_rpc_e2e.rs b/tests/json_rpc_e2e.rs
index 8e9fc600d..262afe56d 100644
--- a/tests/json_rpc_e2e.rs
+++ b/tests/json_rpc_e2e.rs
@@ -803,3 +803,125 @@ async fn json_rpc_skills_runtime_start_tools_call_stop() {
     mock_join.abort();
     rpc_join.abort();
 }
+
+// ---------------------------------------------------------------------------
+// Local AI device profile, presets, and apply preset
+// ---------------------------------------------------------------------------
+
+#[tokio::test]
+async fn json_rpc_local_ai_device_profile_and_presets() {
+    let _env_lock = json_rpc_e2e_env_lock();
+    let tmp = tempdir().expect("tempdir");
+    let home = tmp.path();
+    let openhuman_home = home.join(".openhuman");
+
+    let _home_guard = EnvVarGuard::set_to_path("HOME", home);
+    let _workspace_guard = EnvVarGuard::unset("OPENHUMAN_WORKSPACE");
+    let _backend_url_guard = EnvVarGuard::unset("BACKEND_URL");
+    let _vite_backend_guard = EnvVarGuard::unset("VITE_BACKEND_URL");
+    let _tier_guard = EnvVarGuard::unset("OPENHUMAN_LOCAL_AI_TIER");
+
+    let (mock_addr, mock_join) = serve_on_ephemeral(mock_upstream_router()).await;
+    let mock_origin = format!("http://{}", mock_addr);
+    write_min_config(&openhuman_home, &mock_origin);
+
+    let (rpc_addr, rpc_join) = serve_on_ephemeral(build_core_http_router(false)).await;
+    let rpc_base = format!("http://{}", rpc_addr);
+    tokio::time::sleep(Duration::from_millis(100)).await;
+
+    // --- device_profile ---
+    let profile = post_json_rpc(
+        &rpc_base,
+        30,
+        "openhuman.local_ai_device_profile",
+        json!({}),
+    )
+    .await;
+    let profile_result = assert_no_jsonrpc_error(&profile, "device_profile");
+    assert!(
+        profile_result
+            .get("total_ram_bytes")
+            .and_then(Value::as_u64)
+            .unwrap_or(0)
+            > 0,
+        "expected positive RAM: {profile_result}"
+    );
+    assert!(
+        profile_result
+            .get("cpu_count")
+            .and_then(Value::as_u64)
+            .unwrap_or(0)
+            > 0,
+        "expected positive CPU count: {profile_result}"
+    );
+
+    // --- presets ---
+    let presets = post_json_rpc(&rpc_base, 31, "openhuman.local_ai_presets", json!({})).await;
+    let presets_result = assert_no_jsonrpc_error(&presets, "presets");
+    let presets_arr = presets_result
+        .get("presets")
+        .and_then(Value::as_array)
+        .expect("presets should be an array");
+    assert_eq!(presets_arr.len(), 3, "expected 3 presets: {presets_result}");
+
+    let recommended = presets_result
+        .get("recommended_tier")
+        .and_then(Value::as_str)
+        .expect("should have recommended_tier");
+    assert!(
+        ["low", "medium", "high"].contains(&recommended),
+        "unexpected recommended_tier: {recommended}"
+    );
+
+    let current = presets_result
+        .get("current_tier")
+        .and_then(Value::as_str)
+        .expect("should have current_tier");
+    // Default config uses gemma3:4b-it-qat which matches Medium
+    assert_eq!(current, "medium", "default config should be medium tier");
+
+    // --- apply_preset (switch to Low) ---
+    let apply = post_json_rpc(
+        &rpc_base,
+        32,
+        "openhuman.local_ai_apply_preset",
+        json!({"tier": "low"}),
+    )
+    .await;
+    let apply_result = assert_no_jsonrpc_error(&apply, "apply_preset");
+    assert_eq!(
+        apply_result.get("applied_tier").and_then(Value::as_str),
+        Some("low")
+    );
+    assert_eq!(
+        apply_result.get("chat_model_id").and_then(Value::as_str),
+        Some("gemma3:1b-it-q4_0")
+    );
+
+    // --- verify presets reflects the change ---
+    let presets_after = post_json_rpc(&rpc_base, 33, "openhuman.local_ai_presets", json!({})).await;
+    let presets_after_result = assert_no_jsonrpc_error(&presets_after, "presets_after");
+    assert_eq!(
+        presets_after_result
+            .get("current_tier")
+            .and_then(Value::as_str),
+        Some("low"),
+        "current tier should now be low after apply"
+    );
+
+    // --- apply_preset with invalid tier should error ---
+    let bad_apply = post_json_rpc(
+        &rpc_base,
+        34,
+        "openhuman.local_ai_apply_preset",
+        json!({"tier": "ultra"}),
+    )
+    .await;
+    assert!(
+        bad_apply.get("error").is_some(),
+        "expected error for invalid tier: {bad_apply}"
+    );
+
+    mock_join.abort();
+    rpc_join.abort();
+}