From bc59bb9307e9f141465362bc33a710bedc790fff Mon Sep 17 00:00:00 2001
From: Heiko Liebscher <hliebscher@idn.de>
Date: Mon, 22 Jun 2026 13:58:24 +0200
Subject: [PATCH] i18n(ai-routing): localize workload labels, descriptions, and
 hints (#3870)

Co-authored-by: OpenHuman <package@openhuman.org>
Co-authored-by: Steven Enamakel <enamakel@tinyhumans.ai>
---
 .../components/settings/panels/AIPanel.tsx    | 137 ++++++++++--------
 .../panels/__tests__/AIPanel.test.tsx         |  44 +++++-
 app/src/lib/i18n/ar.ts                        |  40 +++++
 app/src/lib/i18n/bn.ts                        |  40 +++++
 app/src/lib/i18n/de.ts                        |  43 ++++++
 app/src/lib/i18n/en.ts                        |  40 +++++
 app/src/lib/i18n/es.ts                        |  40 +++++
 app/src/lib/i18n/fr.ts                        |  40 +++++
 app/src/lib/i18n/hi.ts                        |  40 +++++
 app/src/lib/i18n/id.ts                        |  40 +++++
 app/src/lib/i18n/it.ts                        |  40 +++++
 app/src/lib/i18n/ko.ts                        |  40 +++++
 app/src/lib/i18n/pl.ts                        |  40 +++++
 app/src/lib/i18n/pt.ts                        |  40 +++++
 app/src/lib/i18n/ru.ts                        |  40 +++++
 app/src/lib/i18n/zh-CN.ts                     |  40 +++++
 16 files changed, 682 insertions(+), 62 deletions(-)

diff --git a/app/src/components/settings/panels/AIPanel.tsx b/app/src/components/settings/panels/AIPanel.tsx
index 5cd7fb2d2..a14d20dde 100644
--- a/app/src/components/settings/panels/AIPanel.tsx
+++ b/app/src/components/settings/panels/AIPanel.tsx
@@ -109,7 +109,14 @@ export type ProviderRef =
   | { kind: 'local'; model: string; temperature?: number | null }
   | { kind: 'claude-code'; model: string; temperature?: number | null };
 
-type Workload = { id: WorkloadId; group: WorkloadGroup; label: string; description: string };
+type Workload = {
+  id: WorkloadId;
+  group: WorkloadGroup;
+  // i18n keys (resolved with `t()` at render) rather than literal English, so the
+  // workload labels/descriptions translate like the rest of the panel.
+  labelKey: string;
+  descriptionKey: string;
+};
 
 export type RoutingMap = Record<WorkloadId, ProviderRef>;
 type RoutingMode = 'managed' | 'own' | 'custom';
@@ -161,79 +168,99 @@ const WORKLOADS: Workload[] = [
   {
     id: 'chat',
     group: 'chat',
-    label: 'Chat',
-    description: 'Direct conversational back-and-forth — “Quick” mode in Conversations',
+    labelKey: 'settings.ai.routing.workload.chat.label',
+    descriptionKey: 'settings.ai.routing.workload.chat.description',
   },
   {
     id: 'reasoning',
     group: 'chat',
-    label: 'Reasoning',
-    description: 'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+    labelKey: 'settings.ai.routing.workload.reasoning.label',
+    descriptionKey: 'settings.ai.routing.workload.reasoning.description',
   },
   {
     id: 'agentic',
     group: 'chat',
-    label: 'Agentic',
-    description: 'Sub-agent runners, tool loops, GIF decisions',
+    labelKey: 'settings.ai.routing.workload.agentic.label',
+    descriptionKey: 'settings.ai.routing.workload.agentic.description',
   },
   {
     id: 'coding',
     group: 'chat',
-    label: 'Coding',
-    description: 'Code generation and refactor passes',
+    labelKey: 'settings.ai.routing.workload.coding.label',
+    descriptionKey: 'settings.ai.routing.workload.coding.description',
   },
   {
     id: 'vision',
     group: 'chat',
-    label: 'Vision',
-    description: 'Image understanding for the vision sub-agent — always multimodal',
+    labelKey: 'settings.ai.routing.workload.vision.label',
+    descriptionKey: 'settings.ai.routing.workload.vision.description',
   },
   {
     id: 'memory',
     group: 'background',
-    label: 'Memory summarization',
-    description: 'Tree-extracts and consolidations',
+    labelKey: 'settings.ai.routing.workload.memory.label',
+    descriptionKey: 'settings.ai.routing.workload.memory.description',
   },
   {
     id: 'heartbeat',
     group: 'background',
-    label: 'Heartbeat',
-    description: 'Background reasoning between user turns',
+    labelKey: 'settings.ai.routing.workload.heartbeat.label',
+    descriptionKey: 'settings.ai.routing.workload.heartbeat.description',
   },
   {
     id: 'learning',
     group: 'background',
-    label: 'Learning · Reflections',
-    description: 'Periodic reflection over recent history',
+    labelKey: 'settings.ai.routing.workload.learning.label',
+    descriptionKey: 'settings.ai.routing.workload.learning.description',
   },
   {
     id: 'subconscious',
     group: 'background',
-    label: 'Subconscious',
-    description: 'Eventfulness scoring + drift checks',
+    labelKey: 'settings.ai.routing.workload.subconscious.label',
+    descriptionKey: 'settings.ai.routing.workload.subconscious.description',
   },
 ];
 
-const WORKLOAD_MODEL_HINTS: Record<WorkloadId, string> = {
-  chat: 'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
-  reasoning:
-    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
-  agentic:
-    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
-  coding:
-    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
-  vision:
-    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
-  memory:
-    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
-  heartbeat:
-    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
-  learning:
-    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
-  subconscious:
-    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
+// i18n keys for the per-workload "Recommended: …" hints (resolved with `t()`).
+const WORKLOAD_MODEL_HINT_KEYS: Record<WorkloadId, string> = {
+  chat: 'settings.ai.routing.workload.chat.hint',
+  reasoning: 'settings.ai.routing.workload.reasoning.hint',
+  agentic: 'settings.ai.routing.workload.agentic.hint',
+  coding: 'settings.ai.routing.workload.coding.hint',
+  vision: 'settings.ai.routing.workload.vision.hint',
+  memory: 'settings.ai.routing.workload.memory.hint',
+  heartbeat: 'settings.ai.routing.workload.heartbeat.hint',
+  learning: 'settings.ai.routing.workload.learning.hint',
+  subconscious: 'settings.ai.routing.workload.subconscious.hint',
 };
 
+// Build the "pending routing changes" summary: one `"<label> → <target>"`
+// entry per workload whose draft route differs from the saved route. Extracted
+// as a pure, exported function so the (translated) formatting is unit-testable
+// without rendering the whole panel.
+export function buildRoutingDiffSummary(
+  saved: RoutingMap,
+  draft: RoutingMap,
+  t: (key: string) => string
+): string[] {
+  const describe = (r: ProviderRef): string => {
+    if (r.kind === 'openhuman') return 'openhuman';
+    if (r.kind === 'default') return 'cloud';
+    const tempSuffix = r.temperature != null ? `@${r.temperature.toFixed(2)}` : '';
+    if (r.kind === 'cloud') return `${r.providerSlug}:${r.model}${tempSuffix}`;
+    return `local:${r.model}${tempSuffix}`;
+  };
+  const out: string[] = [];
+  for (const w of WORKLOADS) {
+    const a = saved[w.id];
+    const b = draft[w.id];
+    if (JSON.stringify(a) !== JSON.stringify(b)) {
+      out.push(`${t(w.labelKey)} → ${describe(b)}`);
+    }
+  }
+  return out;
+}
+
 // TIER_PRESETS removed alongside the Local provider section.
 
 // ─────────────────────────────────────────────────────────────────────────────
@@ -1734,13 +1761,13 @@ const WorkloadRow = ({
     <div className="flex items-center justify-between gap-3 py-3 transition-colors">
       <div className="min-w-0 flex-1 space-y-1">
         <div className="text-sm font-medium text-neutral-900 dark:text-neutral-100">
-          {workload.label}
+          {t(workload.labelKey)}
         </div>
         <div className="text-xs leading-5 text-neutral-500 dark:text-neutral-400">
-          {workload.description}
+          {t(workload.descriptionKey)}
         </div>
         <div className="text-[11px] leading-5 text-neutral-500 dark:text-neutral-400">
-          {WORKLOAD_MODEL_HINTS[workload.id]}
+          {t(WORKLOAD_MODEL_HINT_KEYS[workload.id])}
         </div>
         {resolved ? (
           <div
@@ -2089,7 +2116,9 @@ const CustomRoutingDialog = ({
     <div
       role="dialog"
       aria-modal="true"
-      aria-label={formatI18n(t('settings.ai.customRoutingForWorkload'), { label: workload.label })}
+      aria-label={formatI18n(t('settings.ai.customRoutingForWorkload'), {
+        label: t(workload.labelKey),
+      })}
       className="fixed inset-0 z-50 flex items-center justify-center bg-black/30 p-4">
       <div className="w-full max-w-md rounded-2xl border border-neutral-200 dark:border-neutral-800 bg-white dark:bg-neutral-900 p-6 shadow-soft">
         <div className="flex items-start justify-between gap-3 mb-4">
@@ -2098,10 +2127,10 @@ const CustomRoutingDialog = ({
               {t('settings.ai.customRouting')}
             </h3>
             <p className="mt-0.5 text-xs text-neutral-500 dark:text-neutral-400">
-              {workload.label}
+              {t(workload.labelKey)}
             </p>
             <p className="mt-2 max-w-md text-xs leading-5 text-neutral-500 dark:text-neutral-400">
-              {WORKLOAD_MODEL_HINTS[workload.id]}
+              {t(WORKLOAD_MODEL_HINT_KEYS[workload.id])}
             </p>
           </div>
           <Button
@@ -2966,24 +2995,10 @@ const AIPanel = ({ embedded = false }: AIPanelProps = {}) => {
   // applyPreset removed alongside the Cloud / Local / Mixed preset pills —
   // the new Default/Custom binary toggle handles routing per workload.
 
-  const diffSummary = useMemo(() => {
-    const out: string[] = [];
-    for (const w of WORKLOADS) {
-      const a = saved.routing[w.id];
-      const b = draft.routing[w.id];
-      if (JSON.stringify(a) !== JSON.stringify(b)) {
-        const describe = (r: ProviderRef) => {
-          if (r.kind === 'openhuman') return 'openhuman';
-          if (r.kind === 'default') return 'cloud';
-          const tempSuffix = r.temperature != null ? `@${r.temperature.toFixed(2)}` : '';
-          if (r.kind === 'cloud') return `${r.providerSlug}:${r.model}${tempSuffix}`;
-          return `local:${r.model}${tempSuffix}`;
-        };
-        out.push(`${w.label} → ${describe(b)}`);
-      }
-    }
-    return out;
-  }, [saved, draft]);
+  const diffSummary = useMemo(
+    () => buildRoutingDiffSummary(saved.routing, draft.routing, t),
+    [saved, draft, t]
+  );
 
   const chatRows = WORKLOADS.filter(w => w.group === 'chat');
   const bgRows = WORKLOADS.filter(w => w.group === 'background');
diff --git a/app/src/components/settings/panels/__tests__/AIPanel.test.tsx b/app/src/components/settings/panels/__tests__/AIPanel.test.tsx
index 34a4badb8..31928ade9 100644
--- a/app/src/components/settings/panels/__tests__/AIPanel.test.tsx
+++ b/app/src/components/settings/panels/__tests__/AIPanel.test.tsx
@@ -29,7 +29,11 @@ import {
   openhumanHeartbeatSettingsSet,
   openhumanHeartbeatTickNow,
 } from '../../../../utils/tauriCommands/heartbeat';
-import AIPanel, { BackgroundLoopControls } from '../AIPanel';
+import AIPanel, {
+  BackgroundLoopControls,
+  buildRoutingDiffSummary,
+  type RoutingMap,
+} from '../AIPanel';
 
 vi.mock('../../../../services/api/aiSettingsApi', () => ({
   ALL_WORKLOADS: [
@@ -1539,3 +1543,41 @@ describe('AIPanel', () => {
     await waitFor(() => expect(screen.getByText('tick failed')).toBeInTheDocument());
   });
 });
+
+describe('buildRoutingDiffSummary', () => {
+  const allDefault = (): RoutingMap => ({
+    chat: { kind: 'default' },
+    reasoning: { kind: 'default' },
+    agentic: { kind: 'default' },
+    coding: { kind: 'default' },
+    vision: { kind: 'default' },
+    memory: { kind: 'default' },
+    heartbeat: { kind: 'default' },
+    learning: { kind: 'default' },
+    subconscious: { kind: 'default' },
+  });
+
+  it('emits one "<label> → <target>" entry per changed workload and skips unchanged ones', () => {
+    // Identity `t` so we can assert the workload's i18n label key is used.
+    const t = (key: string) => key;
+    const saved = allDefault();
+    saved.coding = { kind: 'cloud', providerSlug: 'x', model: 'y' };
+    const draft = allDefault();
+    draft.chat = { kind: 'cloud', providerSlug: 'openai', model: 'gpt-4o', temperature: 0.3 };
+    draft.reasoning = { kind: 'openhuman' };
+    draft.agentic = { kind: 'local', model: 'llama3' };
+    // coding: saved=cloud, draft=default → changed; describe(default) === 'cloud'.
+
+    expect(buildRoutingDiffSummary(saved, draft, t)).toEqual([
+      'settings.ai.routing.workload.chat.label → openai:gpt-4o@0.30',
+      'settings.ai.routing.workload.reasoning.label → openhuman',
+      'settings.ai.routing.workload.agentic.label → local:llama3',
+      'settings.ai.routing.workload.coding.label → cloud',
+    ]);
+  });
+
+  it('returns an empty list when draft matches saved', () => {
+    const routing = allDefault();
+    expect(buildRoutingDiffSummary(routing, { ...routing }, k => k)).toEqual([]);
+  });
+});
diff --git a/app/src/lib/i18n/ar.ts b/app/src/lib/i18n/ar.ts
index d869c5616..ce0a8882e 100644
--- a/app/src/lib/i18n/ar.ts
+++ b/app/src/lib/i18n/ar.ts
@@ -3541,6 +3541,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'مهام الخلفية',
   'settings.ai.routing.bgTasksDesc':
     'النماذج المستخدمة خارج التدفق الرئيسي للمحادثات للتلخيص والقلب والتعلم والتقييم اللاوعي.',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': 'إضافة موفر مخصص',
   'settings.ai.globalModel.title': 'اختر نموذجًا واحدًا لكل شيء',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/bn.ts b/app/src/lib/i18n/bn.ts
index f4f1bedd3..23b482783 100644
--- a/app/src/lib/i18n/bn.ts
+++ b/app/src/lib/i18n/bn.ts
@@ -3615,6 +3615,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'ব্যাকগ্রাউন্ড টাস্ক',
   'settings.ai.routing.bgTasksDesc':
     'তুলনা, হার্টবিট, শিক্ষা এবং অবচেতনের মূল্যায়নের মূল কথোপকথনের বাইরে যে মডেলগুলো ব্যবহার করা হয়েছে সেগুলো ব্যবহার করা হয়েছে।',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': 'কাস্টম প্রদানকারী যোগ করুন',
   'settings.ai.globalModel.title': 'সবকিছুর জন্য একটি মডেল চয়ন করুন',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/de.ts b/app/src/lib/i18n/de.ts
index 642dced3a..c222e14bd 100644
--- a/app/src/lib/i18n/de.ts
+++ b/app/src/lib/i18n/de.ts
@@ -3702,6 +3702,49 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'Hintergrundaufgaben',
   'settings.ai.routing.bgTasksDesc':
     'Modelle, die außerhalb des Hauptgesprächsflusses für Zusammenfassung, Herzschlag, Lernen und unbewusste Bewertung verwendet werden.',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direkter dialogischer Austausch hin und her – „Schnell“-Modus in Unterhaltungen',
+  'settings.ai.routing.workload.chat.hint':
+    'Empfohlen: ein günstiges oder mittelpreisiges, schnelles Chat-Modell mit hohem Tokens/Sek.-Durchsatz und geringer Latenz. Quelloffene lokale Modelle eignen sich hier gut, wenn sie sich reaktionsschnell anfühlen.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Haupt-Chat-Agent, Meeting-Zusammenfasser – „Reasoning“-Modus in Unterhaltungen',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Empfohlen: ein leistungsfähigeres Frontier- oder starkes Reasoning-Modell für tiefes Nachdenken. Wird für den Haupt-Chat-Agenten, Meeting-Zusammenfassungen und aufwändigere Antwortsynthese genutzt.',
+  'settings.ai.routing.workload.agentic.label': 'Agentisch',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-Agent-Runner, Tool-Schleifen, GIF-Entscheidungen',
+  'settings.ai.routing.workload.agentic.hint':
+    'Empfohlen: ein zuverlässiges, anweisungstreues Modell mit starker Tool-Nutzung. Mittelpreisige Frontier-Modelle sind meist am sichersten; leistungsfähige Open-Source-Modelle funktionieren, wenn das Tool-Calling stabil ist.',
+  'settings.ai.routing.workload.coding.label': 'Programmierung',
+  'settings.ai.routing.workload.coding.description': 'Code-Generierung und Refactoring-Durchläufe',
+  'settings.ai.routing.workload.coding.hint':
+    'Empfohlen: ein auf Programmierung abgestimmtes Modell mit starker Anweisungstreue, Edit-Qualität und Long-Context-Leistung. Hier lohnt sich meist eine größere Investition.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Bildverständnis für den Vision-Sub-Agenten – immer multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Empfohlen: ein multimodales Modell, das Bildeingaben akzeptiert. Der verwaltete Standard (vision-v1) ist bildfähig; jeder hier zugewiesene Anbieter wird stets als vision-fähig behandelt.',
+  'settings.ai.routing.workload.memory.label': 'Memory-Zusammenfassung',
+  'settings.ai.routing.workload.memory.description': 'Baum-Extraktionen und Konsolidierungen',
+  'settings.ai.routing.workload.memory.hint':
+    'Empfohlen: ein günstigeres Zusammenfassungsmodell. Es sollte konsistent und kompakt sein, braucht aber keine erstklassige Frontier-Reasoning-Leistung.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description':
+    'Hintergrund-Reasoning zwischen Nutzer-Eingaben',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Empfohlen: ein günstiges, effizientes Hintergrundmodell. Läuft häufig zwischen den Eingaben, daher zählen niedrige Kosten mehr als maximale Intelligenz.',
+  'settings.ai.routing.workload.learning.label': 'Lernen · Reflexionen',
+  'settings.ai.routing.workload.learning.description':
+    'Periodische Reflexion über den jüngsten Verlauf',
+  'settings.ai.routing.workload.learning.hint':
+    'Empfohlen: ein stärkeres reflektierendes Modell. Kann mittelpreisig oder Premium sein, da es von besserer Synthese über den jüngsten Verlauf profitiert.',
+  'settings.ai.routing.workload.subconscious.label': 'Unterbewusstsein',
+  'settings.ai.routing.workload.subconscious.description':
+    'Bewertung der Ereignishaftigkeit + Drift-Prüfungen',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Empfohlen: ein sehr günstiges Überwachungsmodell, idealerweise leichtgewichtig und vorhersehbar. Für Ereignishaftigkeits-Bewertung, Drift-Prüfungen und stille Hintergrundauswertung.',
   'settings.ai.routing.addCustomProvider': 'Benutzerdefinierten Anbieter hinzufügen',
   'settings.ai.globalModel.title': 'Wählen Sie ein Modell für alles',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/en.ts b/app/src/lib/i18n/en.ts
index 2199e7c21..92eda4da6 100644
--- a/app/src/lib/i18n/en.ts
+++ b/app/src/lib/i18n/en.ts
@@ -4217,6 +4217,46 @@ const en: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'Background Tasks',
   'settings.ai.routing.bgTasksDesc':
     'Models used outside the main conversation flow for summarization, heartbeat, learning, and subconscious evaluation.',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': 'Add Custom Provider',
   'settings.ai.globalModel.title': 'Choose one model for everything',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/es.ts b/app/src/lib/i18n/es.ts
index 87374594d..9187c9039 100644
--- a/app/src/lib/i18n/es.ts
+++ b/app/src/lib/i18n/es.ts
@@ -3678,6 +3678,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'Tareas en segundo plano',
   'settings.ai.routing.bgTasksDesc':
     'Modelos utilizados fuera del flujo principal de la conversación para resumen, latido, aprendizaje y evaluación subconsciente.',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': 'Agregar proveedor personalizado',
   'settings.ai.globalModel.title': 'Elige un modelo para todo',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/fr.ts b/app/src/lib/i18n/fr.ts
index e21552ac2..aad89ddf9 100644
--- a/app/src/lib/i18n/fr.ts
+++ b/app/src/lib/i18n/fr.ts
@@ -3693,6 +3693,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'Tâches en arrière-plan',
   'settings.ai.routing.bgTasksDesc':
     "Modèles utilisés en dehors du flux principal de conversation pour la synthèse, le suivi, l'apprentissage et l'évaluation subconsciente.",
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': 'Ajouter un fournisseur personnalisé',
   'settings.ai.globalModel.title': 'Choisissez un modèle pour tout',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/hi.ts b/app/src/lib/i18n/hi.ts
index b23a0c2c9..7711ee19d 100644
--- a/app/src/lib/i18n/hi.ts
+++ b/app/src/lib/i18n/hi.ts
@@ -3620,6 +3620,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'पृष्ठभूमि कार्य',
   'settings.ai.routing.bgTasksDesc':
     'मॉडल संक्षेपण, दिल की धड़कन, सीखने और अवचेतन मूल्यांकन के लिए मुख्य बातचीत प्रवाह के बाहर इस्तेमाल किया।',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': 'कस्टम प्रदाता जोड़ें',
   'settings.ai.globalModel.title': 'हर चीज़ के लिए एक मॉडल चुनें',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/id.ts b/app/src/lib/i18n/id.ts
index 8cc4f4f82..52b0bf6d3 100644
--- a/app/src/lib/i18n/id.ts
+++ b/app/src/lib/i18n/id.ts
@@ -3624,6 +3624,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'Tugas Latar Belakang',
   'settings.ai.routing.bgTasksDesc':
     'Model digunakan di luar aliran percakapan utama untuk summarisasi, detak jantung, pembelajaran, dan evaluasi bawah sadar.',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': 'Tambahkan Penyedia Khusus',
   'settings.ai.globalModel.title': 'Pilih satu model untuk semuanya',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/it.ts b/app/src/lib/i18n/it.ts
index e8ae55b64..4342635f5 100644
--- a/app/src/lib/i18n/it.ts
+++ b/app/src/lib/i18n/it.ts
@@ -3670,6 +3670,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'Attività in background',
   'settings.ai.routing.bgTasksDesc':
     'Modelli utilizzati al di fuori del flusso principale della conversazione per sintesi, battito, apprendimento e valutazione subconscia.',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': 'Aggiungi provider personalizzato',
   'settings.ai.globalModel.title': 'Scegli un modello per tutto',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/ko.ts b/app/src/lib/i18n/ko.ts
index c8f94c0bb..363fc29ad 100644
--- a/app/src/lib/i18n/ko.ts
+++ b/app/src/lib/i18n/ko.ts
@@ -3588,6 +3588,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': '백그라운드 작업',
   'settings.ai.routing.bgTasksDesc':
     '요약, 하트비트, 학습 및 잠재의식 평가처럼 기본 대화 흐름 밖에서 사용되는 모델입니다.',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': '사용자 정의 공급자 추가',
   'settings.ai.globalModel.title': '모든 것에 대해 하나의 모델을 선택합니다.',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/pl.ts b/app/src/lib/i18n/pl.ts
index 1b05894a2..a67fe5b49 100644
--- a/app/src/lib/i18n/pl.ts
+++ b/app/src/lib/i18n/pl.ts
@@ -3671,6 +3671,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'Zadania w tle',
   'settings.ai.routing.bgTasksDesc':
     'Modele używane poza głównym przepływem rozmowy do podsumowań, heartbeat, uczenia i oceny podświadomości.',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': 'Dodaj własnego dostawcę',
   'settings.ai.globalModel.title': 'Wybierz jeden model do wszystkiego',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/pt.ts b/app/src/lib/i18n/pt.ts
index dd02b360a..dfd381d15 100644
--- a/app/src/lib/i18n/pt.ts
+++ b/app/src/lib/i18n/pt.ts
@@ -3676,6 +3676,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'Tarefas de segundo plano',
   'settings.ai.routing.bgTasksDesc':
     'Modelos usados fora do fluxo principal de conversa para resumir, monitoramento, aprendizado e avaliação subconsciente.',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': 'Adicionar provedor personalizado',
   'settings.ai.globalModel.title': 'Escolha um modelo para tudo',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/ru.ts b/app/src/lib/i18n/ru.ts
index 4f24975eb..738b10726 100644
--- a/app/src/lib/i18n/ru.ts
+++ b/app/src/lib/i18n/ru.ts
@@ -3647,6 +3647,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.backgroundTasks': 'Фоновые задачи',
   'settings.ai.routing.bgTasksDesc':
     'Модели, используемые вне основного потока разговора для подведения итогов, сердцебиения, обучения и подсознательной оценки.',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': 'Добавить специального поставщика.',
   'settings.ai.globalModel.title': 'Выберите одну модель для всего.',
   'settings.ai.globalModel.desc':
diff --git a/app/src/lib/i18n/zh-CN.ts b/app/src/lib/i18n/zh-CN.ts
index d2d0dca7b..2bb563374 100644
--- a/app/src/lib/i18n/zh-CN.ts
+++ b/app/src/lib/i18n/zh-CN.ts
@@ -3436,6 +3436,46 @@ const messages: TranslationMap = {
   'settings.ai.routing.chatDesc': '用于直接用户交互、回复、推理、智能体循环和编码帮助的模型。',
   'settings.ai.routing.backgroundTasks': '后台任务',
   'settings.ai.routing.bgTasksDesc': '在主对话流程之外用于总结、心跳、学习和潜意识评估的模型。',
+  'settings.ai.routing.workload.chat.label': 'Chat',
+  'settings.ai.routing.workload.chat.description':
+    'Direct conversational back-and-forth — “Quick” mode in Conversations',
+  'settings.ai.routing.workload.chat.hint':
+    'Recommended: a cheap or mid-cost fast chat model with high tokens/sec and low latency. Open-source local models can work well here if they feel responsive.',
+  'settings.ai.routing.workload.reasoning.label': 'Reasoning',
+  'settings.ai.routing.workload.reasoning.description':
+    'Main chat agent, meeting summarizer — “Reasoning” mode in Conversations',
+  'settings.ai.routing.workload.reasoning.hint':
+    'Recommended: a more expensive frontier or strong reasoning model for deep thinking. This is used for the main chat agent, meeting summaries, and heavier answer synthesis.',
+  'settings.ai.routing.workload.agentic.label': 'Agentic',
+  'settings.ai.routing.workload.agentic.description':
+    'Sub-agent runners, tool loops, GIF decisions',
+  'settings.ai.routing.workload.agentic.hint':
+    'Recommended: a reliable instruction-following model with strong tool use. Mid-cost frontier models are usually safest; capable open-source models can work if tool calling is stable.',
+  'settings.ai.routing.workload.coding.label': 'Coding',
+  'settings.ai.routing.workload.coding.description': 'Code generation and refactor passes',
+  'settings.ai.routing.workload.coding.hint':
+    'Recommended: a coding-tuned model with strong instruction following, edit quality, and long-context performance. This is usually worth spending more on.',
+  'settings.ai.routing.workload.vision.label': 'Vision',
+  'settings.ai.routing.workload.vision.description':
+    'Image understanding for the vision sub-agent — always multimodal',
+  'settings.ai.routing.workload.vision.hint':
+    'Recommended: a multimodal model that accepts image input. The managed default (vision-v1) is image-capable; any provider you route here is always treated as vision-enabled.',
+  'settings.ai.routing.workload.memory.label': 'Memory summarization',
+  'settings.ai.routing.workload.memory.description': 'Tree-extracts and consolidations',
+  'settings.ai.routing.workload.memory.hint':
+    'Recommended: a cheaper summarization model. It should be consistent and compact, but it does not need premium frontier-level reasoning.',
+  'settings.ai.routing.workload.heartbeat.label': 'Heartbeat',
+  'settings.ai.routing.workload.heartbeat.description': 'Background reasoning between user turns',
+  'settings.ai.routing.workload.heartbeat.hint':
+    'Recommended: a cheap, efficient background model. This runs often between turns, so low cost matters more than maximum intelligence.',
+  'settings.ai.routing.workload.learning.label': 'Learning · Reflections',
+  'settings.ai.routing.workload.learning.description': 'Periodic reflection over recent history',
+  'settings.ai.routing.workload.learning.hint':
+    'Recommended: a stronger reflective model. This can be mid-cost or premium because it benefits from better synthesis over recent history.',
+  'settings.ai.routing.workload.subconscious.label': 'Subconscious',
+  'settings.ai.routing.workload.subconscious.description': 'Eventfulness scoring + drift checks',
+  'settings.ai.routing.workload.subconscious.hint':
+    'Recommended: a very cheap monitoring model, ideally one that is lightweight and predictable. This is for eventfulness scoring, drift checks, and quiet background evaluation.',
   'settings.ai.routing.addCustomProvider': '添加自定义提供商',
   'settings.ai.globalModel.title': '为所有事情选择一种型号',
   'settings.ai.globalModel.desc':