diff --git a/CHANGELOG.md b/CHANGELOG.md
index d054e580..a0b3954c 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -125,6 +125,22 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   [ADR 0012](docs/adr/0012-local-server-connectivity-tauri-http.md) and
   [`docs/LOCAL-AI.md`](docs/LOCAL-AI.md).
 
+- **Desktop Ollama/LM Studio/vLLM discovery still broken in packaged builds after #269 (#266).**
+  Root cause: `vite.config.ts`'s `rollupOptions.external` unconditionally externalized every
+  `@tauri-apps/*` package from **every** `vite build`, including the exact build Tauri's
+  `beforeBuildCommand` invokes to produce the `.deb`/`.msi`. Since `services/localServerHttp.ts`'s
+  `@tauri-apps/plugin-http` import is dynamic, this left an unresolvable bare module specifier in
+  the shipped desktop bundle — every caller's catch classified the resulting load failure
+  identically to a genuinely-down server (no CORS noise, no discovery, even with Ollama/LM Studio
+  running). `tauri dev` was unaffected (Vite's dev server doesn't apply `rollupOptions`), so the
+  regression only surfaced in packaged builds. Fixed by extracting the existing Tauri-build
+  detection (`resolveViteBase.ts`'s `TAURI_ENV_PLATFORM`/`TAURI_PLATFORM` check) into a shared
+  `isTauriBuild()` export and making the external array conditional on it; the web/PWA build is
+  unaffected. `localServerHttp.ts`'s `resolveFetch()` also now classifies a plugin load failure as
+  a distinct `LocalServerError('plugin_unavailable')` (logged), instead of folding it into
+  `'unreachable'`. See the 2026-07-28 update in
+  [ADR 0012](docs/adr/0012-local-server-connectivity-tauri-http.md).
+
 - **Feature-catalog / slice default drift made structurally impossible.** `features/featureCatalog.ts`
   now covers all **22** flags (was 16) and **derives** each entry's `defaultOn` from the slice's
   `defaultFeatureFlagsState` instead of hand-keying it — the class of bug where the catalog said
diff --git a/CLAUDE.md b/CLAUDE.md
index 22d34899..c5ba180f 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -6,15 +6,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 
 ### Low-End Local Hardware (Current Environment)
 
-**ONE Bash tool call per turn.** Concurrent shells (vitest, biome, tsc, vite, pnpm build) cause OOM and pool-worker timeouts.
-- ONE Bash tool call per turn. Wait for the result. Then proceed.
-- NO `run_in_background` for vitest, biome, tsc, vite, or any pnpm build command.
-- NO parallel Agent tool calls that each issue shell commands.
-- NO multiple Bash tool calls in the same response block.
-- Chain sequential steps inside ONE Bash call using `&&` if needed.
-- DevContainer / Codespaces configuration has been removed; this is a local-only development environment.
-
-**Plan-mode exception — parallel exploration IS allowed.** The OOM risk is from concurrent *heavy shells* (vitest/biome/tsc/vite/build), not from read-only exploration. In **plan mode** (read-only research, no builds), you SHOULD launch parallel `Explore`/`Plan` subagents to map the codebase faster — these read files and run light `grep`/`ls`, which is safe. The ONE-Bash-per-turn rule still governs the **main loop's own heavy commands** at all times, and parallel agents must not each kick off vitest/biome/tsc/vite/build. Outside plan mode (implementation), keep agent spawns sequential when they issue heavy shells.
+Shell-execution rules (one Bash call per turn, no parallel heavy shells, plan-mode exception) live in the user-global `~/.claude/CLAUDE.md` and apply here. Project-specific addition: DevContainer / Codespaces configuration has been removed; this is a local-only development environment.
 
 ## Commands
 
@@ -78,8 +70,6 @@ Conventional Commits format: `feat:`, `fix:`, `docs:`, `refactor:`, `test:`, `ch
 
 WorldScript Studio is an offline-first PWA — a React 19 SPA with Google Gemini AI, IndexedDB persistence, and optional Tauri desktop packaging. No backend; API keys are entered in the UI and encrypted at rest.
 
-**Stack:** React 19, TypeScript (strict), Vite 8, Tailwind CSS 4.x, Redux Toolkit 2.x, pnpm 11, Node ≥ 22. Four internal workspace packages (`@domain/ai-core`, `@domain/ui`, `collab-transport`, `@domain/worker-bus` in `packages/`) are consumed as `workspace:*` deps.
-
 **Live:** `https://worldscript-studio.vercel.app/` (Vercel, primary) · GitHub Pages: `https://qnbs.github.io/WorldScript-Studio/` · Cloudflare Pages: `wrangler.toml` · Vercel: `vercel.json`.
 
 ### Directory map
@@ -194,6 +184,8 @@ Wrap each major view root with `components/ui/ViewErrorBoundary.tsx` — provide
 
 **Prompt assembly:** `services/ragPromptAssembly.ts` — `assembleRAGPrompt(opts)`. Templates from `services/promptLibrary.ts`.
 
+**Heuristic-fallback layer:** `services/ai/heuristicFallback/` — when every provider in the chain fails terminally (offline, quota, Eco/Heuristics-only mode), `aiProviderService` calls `applyHeuristicFallback(task, ctx)` (`seam.ts`) before falling through to the generic local stub. It looks up a per-feature generator in `registry.ts` (keyed by task id: `outline`, `character.profile`, `world.profile`, `plotBoard.beat`) and builds a `HeuristicFallbackResult` from existing project data — no network call. Returns `null` when no generator is registered for a task, so wiring a new call site is always non-breaking. Generators self-register via `registerHeuristicGenerator(task, fn)` at module load (mirrors the `services/copilot/heuristicEngine.ts` pluggable-rule pattern). `useHeuristicFallback()` + `<AssistedModeBadge>` surface an "Assisted (offline)" badge wherever a result came from this layer; events also feed `telemetryService` as `backend: 'heuristic'`.
+
 ### DuckDB Analytics
 
 `workers/duckdbWorker.ts` off main thread (OPFS → in-memory fallback). `duckdbClient.ts`: singleton, init retry 3× backoff. Schema: 10 tables + 5 views incl. `rag_chunks` (FLOAT[384]). Gate all paths behind `enableDuckDbAnalytics`. Dual-write via `duckdbListenerLoader.ts` (dynamically imported). `ragVectorMigration.ts`: FLOAT[64]→FLOAT[384] upgrade. `useDuckDb.ts` 30s timeout; `useAnalytics.ts` parallelizes 4 queries.
@@ -248,7 +240,7 @@ Production uses **rolldown** (not esbuild/rollup); CI E2E runs `vite dev` — pr
 
 Experimental features are gated behind `features/featureFlags/featureFlagsSlice.ts` (**22 flags**). New installs get the **full feature set**: all flags default **on** except six user-opt-in flags. `enableCodexAutoTracking` + `enableCrossProjectSearch` were promoted to permanent core behaviour (v1.20 / v1.8); `enablePlotBoardV2` and `enableCloudSync` were retired — none of those four remain in the slice. UI: Settings → Experimental flags (`FeatureFlagsSection.tsx`). Do not use scattered `if (true)` hacks.
 
-**Default on (17):** `enableStoryBibleAdvanced`, `enableBinderResearch`, `enableCompileWizard`, `enableProjectHealthScore`, `enableAppHealthPanel`, `enableDuckDbAnalytics`, `enableObjectsGroups`, `enableMindMaps`, `enableCharacterInterviews`, `enableLoraAdapters`, `enablePluginSystem`, `enableIdbAtRestEncryption` (B-1, passphrase UX complete — Settings › Privacy), `enableAdaptiveAiEngine`, `enableWebnnInference`, `enableComputeShaders`, `enableWorkerBusV2`, `enableRustCompute`. **User opt-in — default off (6):** `enableProForge` (experimental, token-heavy 8-stage agentic pipeline — flipped to opt-in in v1.24 post-release), `enableVoiceSupport` (requires browser mic permission), `enableVoiceWasm` (B-2, ~57 MB Whisper download), `enableGlobalCopilot` (ambient AI), `enableRtlLayout` (B-5, ar/he stubs only), `enableLocalFirstSync` (shadow Yjs projection, ADR-0008; Redux stays SoT). The Settings UI groups these by catalog tier; `features/featureCatalog.ts` **derives** each flag's `defaultOn` from the slice (no hand-keyed drift). Note: `enableCloudSync` was **retired** in v1.20 (no UI shipped; `CloudSyncBackend.create()` requires explicit-consent boolean instead).
+**Default on (16):** `enableStoryBibleAdvanced`, `enableBinderResearch`, `enableCompileWizard`, `enableProjectHealthScore`, `enableAppHealthPanel`, `enableDuckDbAnalytics`, `enableObjectsGroups`, `enableMindMaps`, `enableCharacterInterviews`, `enableLoraAdapters`, `enablePluginSystem`, `enableIdbAtRestEncryption` (B-1, passphrase UX complete — Settings › Privacy), `enableAdaptiveAiEngine`, `enableComputeShaders`, `enableWorkerBusV2`, `enableRustCompute`. **User opt-in — default off (6):** `enableProForge` (experimental, token-heavy 8-stage agentic pipeline — flipped to opt-in in v1.24 post-release), `enableVoiceSupport` (requires browser mic permission), `enableVoiceWasm` (B-2, ~57 MB Whisper download), `enableGlobalCopilot` (ambient AI), `enableRtlLayout` (B-5, ar/he stubs only), `enableLocalFirstSync` (shadow Yjs projection, ADR-0008; Redux stays SoT). The Settings UI groups these by catalog tier; `features/featureCatalog.ts` **derives** each flag's `defaultOn` from the slice (no hand-keyed drift). Note: `enableCloudSync` was **retired** in v1.20 (no UI shipped; `CloudSyncBackend.create()` requires explicit-consent boolean instead).
 
 ### Command Center & shortcuts
 
@@ -407,7 +399,7 @@ Reference plugins: `wordCountOverlay.plugin.ts`, `sceneAppender.plugin.ts`. Gate
 
 ### Virtual scrolling
 
-`NavigatorPanel.tsx` uses `useVirtualizer` (`@tanstack/react-virtual`): scrollable `<ul ref={scrollRef}>` + `position: relative`; sentinel `<li>` sets `height: totalSize`; items `position: absolute, transform: translateY(start)`. Items need `data-index` + `ref={measureElement}`. Use `estimateSize: () => 40, overscan: 3`. Never lift `overflow-y: auto` into a parent.
+Feature-specific implementation patterns (Plot Board, ProForge Pipeline, scene-level services, LanguageTool, test mock patterns, Settings Navigation, cross-project & backup, Global AI Copilot, Voice Full Support, local inference, Plugin System, Cloud Sync, LoRA Adapter Inference, virtual scrolling) now live in nested `CLAUDE.md` files, loaded automatically only when working under that directory: `features/plotBoard/`, `services/proForge/`, `services/`, `tests/`, `components/`, `services/copilot/`, `services/voice/`, `services/cloudSync/`, `features/lora/`.
 
 ## Known Technical Debt
 
diff --git a/components/CLAUDE.md b/components/CLAUDE.md
new file mode 100644
index 00000000..503caaa3
--- /dev/null
+++ b/components/CLAUDE.md
@@ -0,0 +1,7 @@
+# Settings Navigation
+
+`components/SettingsView.tsx` uses `NAV_GROUPS` — typed array of `{ key: string; ids: readonly string[] }` — for semantic sidebar sections (Writing, AI Models, Appearance & Accessibility, Privacy & Data, Connections, System). When adding a new settings tab: add its `id` to the correct group in `NAV_GROUPS`; do not create a flat ungrouped entry.
+
+# Virtual scrolling
+
+`NavigatorPanel.tsx` uses `useVirtualizer` (`@tanstack/react-virtual`): scrollable `<ul ref={scrollRef}>` + `position: relative`; sentinel `<li>` sets `height: totalSize`; items `position: absolute, transform: translateY(start)`. Items need `data-index` + `ref={measureElement}`. Use `estimateSize: () => 40, overscan: 3`. Never lift `overflow-y: auto` into a parent.
diff --git a/components/settings/AiProviderCard.tsx b/components/settings/AiProviderCard.tsx
index dc91c822..1d866798 100644
--- a/components/settings/AiProviderCard.tsx
+++ b/components/settings/AiProviderCard.tsx
@@ -101,7 +101,14 @@ export const AiProviderCard: FC<AiProviderCardProps> = ({
         setTestStatus('ok');
       } else {
         setTestStatus('error');
-        setTestError(result.error ?? t('settings.ai.connectionFailed'));
+        // QNBS-v3: `kind` is a stable, i18n-mappable classification — prefer it over the raw
+        // `error` string (which is a technical/English message kept for logs) so every failure
+        // path across all providers renders localized text, not leaked English.
+        setTestError(
+          result.kind
+            ? t(`settings.ai.testError.${result.kind}`, result.params)
+            : (result.error ?? t('settings.ai.connectionFailed')),
+        );
       }
     } catch (e) {
       setTestStatus('error');
diff --git a/config/resolveViteBase.ts b/config/resolveViteBase.ts
index ed7e5326..d04bdf64 100644
--- a/config/resolveViteBase.ts
+++ b/config/resolveViteBase.ts
@@ -5,6 +5,23 @@
 
 export const GITHUB_PAGES_BASE = '/WorldScript-Studio/';
 
+/**
+ * True when the active `vite build` is the one Tauri's `beforeBuildCommand` hook invokes to
+ * produce the desktop bundle, as opposed to the web/PWA build.
+ *
+ * QNBS-v3: Tauri 2.x sets TAURI_ENV_PLATFORM for hook commands; the legacy Tauri 1.x name was
+ * TAURI_PLATFORM. Accept both so a CLI downgrade/upgrade can't silently reintroduce a
+ * platform-detection regression. Single source of truth — reused by both the `base` resolution
+ * below and `vite.config.ts`'s `rollupOptions.external` (a desktop build that externalized
+ * `@tauri-apps/*` left an unresolvable bare module specifier in the shipped bundle; see
+ * docs/adr/0012-local-server-connectivity-tauri-http.md).
+ */
+export function isTauriBuild(env: Record<string, string | undefined> = process.env): boolean {
+  // QNBS-v3: require a trimmed, non-empty value — an empty-string env var (e.g. a CI matrix
+  // that sets TAURI_ENV_PLATFORM="" to "unset" it) must not be read as "Tauri build".
+  return Boolean(env['TAURI_ENV_PLATFORM']?.trim()) || Boolean(env['TAURI_PLATFORM']?.trim());
+}
+
 /**
  * Resolve the Vite `base` for the active build target.
  * - Tauri desktop  → './'  (relative; assets must load from tauri://localhost/ root)
@@ -13,11 +30,7 @@ export const GITHUB_PAGES_BASE = '/WorldScript-Studio/';
  * - default (GitHub Pages project page) → '/WorldScript-Studio/'
  */
 export function resolveViteBase(env: Record<string, string | undefined> = process.env): string {
-  // QNBS-v3: Tauri 2.x sets TAURI_ENV_PLATFORM for hook commands (beforeBuildCommand); the legacy
-  // Tauri 1.x name was TAURI_PLATFORM. Accept both so a CLI downgrade/upgrade can't silently
-  // reintroduce the absolute-base blank-screen bug.
-  const isTauri = env['TAURI_ENV_PLATFORM'] !== undefined || env['TAURI_PLATFORM'] !== undefined;
-  if (isTauri) {
+  if (isTauriBuild(env)) {
     return './';
   }
   const viteBase = env['VITE_BASE']?.trim();
diff --git a/docs/adr/0012-local-server-connectivity-tauri-http.md b/docs/adr/0012-local-server-connectivity-tauri-http.md
index 13accefa..6e705345 100644
--- a/docs/adr/0012-local-server-connectivity-tauri-http.md
+++ b/docs/adr/0012-local-server-connectivity-tauri-http.md
@@ -100,3 +100,32 @@ modes were reported in #266:
 - **Allow PWA status-only probing with Private Network Access permission** — rejected (product
   decision): keeps CORS noise possible, adds a second-class UX path, and conflicts with the
   privacy-first "desktop-only" policy for localhost servers.
+
+## Update (2026-07-28): build-pipeline gap left the desktop path broken in packaged builds
+
+This ADR's decision (route local-server HTTP through `@tauri-apps/plugin-http`) was correct, but
+the packaged desktop build never actually exercised it. `vite.config.ts`'s `rollupOptions.external`
+unconditionally externalized every `@tauri-apps/*` package from **every** `vite build`, including
+the exact build Tauri's `beforeBuildCommand` invokes to produce the `.deb`/`.msi`. Since
+`services/localServerHttp.ts`'s `@tauri-apps/plugin-http` import is dynamic (`await import(...)`),
+externalizing it left an unresolvable bare module specifier in the shipped bundle — confirmed with
+a real build + real-Chromium repro, producing `TypeError: Failed to resolve module specifier
+'@tauri-apps/plugin-http'` the instant `resolveFetch()` ran, before any network request. Every
+caller's catch classified this identically to a genuinely-down server, matching the exact symptom
+reported on issue #266 after #269 merged: no CORS console noise (nothing reached the network
+layer), and no Ollama/LM Studio discovery despite both running.
+
+Root cause: `resolveViteBase.ts` already had the right Tauri-vs-web build detection (via
+`TAURI_ENV_PLATFORM`/`TAURI_PLATFORM`), used for the `base` config, but `rollupOptions.external`
+was never given the same treatment. `tauri dev` was unaffected (Vite's dev server doesn't apply
+`rollupOptions`), so the regression only surfaced in packaged builds — and the unit test suite
+mocks `@tauri-apps/plugin-http` via `vi.mock`, which bypasses real module resolution entirely and
+structurally cannot catch this class of bug.
+
+Fix: extracted the Tauri-build check into a shared `isTauriBuild()` export in `resolveViteBase.ts`
+and made `rollupOptions.external` conditional on it — the desktop build now bundles
+`@tauri-apps/plugin-http` correctly; the web/PWA build is unaffected (those code paths are gated
+by `isTauriRuntime()` and never exercised there). `services/localServerHttp.ts`'s `resolveFetch()`
+also now wraps the dynamic import in its own try/catch, classifying a load failure as a distinct
+`LocalServerError('plugin_unavailable')` and logging it, so a future regression of this class fails
+loudly and distinctly instead of silently misclassifying as "unreachable."
diff --git a/features/lora/CLAUDE.md b/features/lora/CLAUDE.md
new file mode 100644
index 00000000..d3c3d243
--- /dev/null
+++ b/features/lora/CLAUDE.md
@@ -0,0 +1,7 @@
+# LoRA Adapter Inference
+
+**Wiring (C-3):** `AIRequestOptions.loraModelPath?: string` — when set and `provider === 'ollama'`, `streamProvider()` substitutes it as the Ollama model identifier. `selectActiveLoraOllamaTag` (`features/lora/loraSelectors.ts`) returns active adapter's `ollamaModelTag` or null.
+
+**Prerequisite:** `ollama create <tag> -f Modelfile` before setting `ollamaModelTag`. Training is a Python sidecar.
+
+**Gating:** `enableLoraAdapters` flag. UI: Settings → AI → Fine-Tuning.
diff --git a/features/plotBoard/CLAUDE.md b/features/plotBoard/CLAUDE.md
new file mode 100644
index 00000000..4873b76a
--- /dev/null
+++ b/features/plotBoard/CLAUDE.md
@@ -0,0 +1,9 @@
+# Plot Board
+
+**plotBoardSlice:** `features/plotBoard/plotBoardSlice.ts` — ephemeral viewport/UI state only (zoom/pan/mode/draw). NOT undo-able; persists to `localStorage`. Story content (connections, subplots, tensionOverrides) lives in `projectSlice` — use selectors from `features/project/projectSelectors.ts` and dispatch `projectActions.add/removePlotConnection`, `add/deletePlotSubplot`, `setPlotTensionOverride`.
+
+**plotBoardService:** `services/plotBoardService.ts` — `computeTensionCurve(sections, overrides)`, `autoLayoutScenes(sections)`, `exportBoardAsSvg(svgEl)`.
+
+**Plot Board AI:** `features/project/thunks/plotBoardAiThunks.ts` — `suggestNextBeatThunk` calls `assembleRAGPrompt` then dispatches to AI. Hook: `hooks/usePlotBoardAi.ts`.
+
+**PlotMinimap:** `components/scene-board/PlotMinimap.tsx` — viewport overview overlay. `plotLayoutUtils.ts` provides grid-snap helpers.
diff --git a/locales/ar/settings.json b/locales/ar/settings.json
index 0c7e5a73..3653537d 100644
--- a/locales/ar/settings.json
+++ b/locales/ar/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – إبداعي",
   "settings.ai.temperature.precise": "0 – دقيق",
   "settings.ai.testConnection": "اختبار الاتصال",
+  "settings.ai.testError.noApiKey": "لم يتم تعيين مفتاح API لـ {{provider}}",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "انتهت مهلة الاتصال ({{url}})",
+  "settings.ai.testError.unreachable": "غير قابل للوصول ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "شبكة الخادم المحلي غير متوفرة: فشل تحميل مكوّن HTTP لسطح المكتب.",
+  "settings.ai.testError.desktopRequired": "تتوفر Ollama والخوادم المحلية المتوافقة مع OpenAI فقط في تطبيق سطح المكتب. تحظر المتصفحات الاتصالات المباشرة من صفحات الويب بـ localhost (CORS وPrivate Network Access).",
+  "settings.ai.testError.backendProxyRequired": "يتطلب Claude وكيل خلفية (قيود CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU غير متوفر في هذا المتصفح — يحتاج WebLLM إلى WebGPU (جرّب Chrome/Edge أو فعّل العلامات).",
+  "settings.ai.testError.unknownProvider": "مزوّد غير معروف",
+  "settings.ai.testError.unexpected": "حدث خطأ غير متوقع. يرجى المحاولة مرة أخرى.",
   "settings.ai.title": "إعداد الذكاء الاصطناعي",
   "settings.ai.transformers.hint": "يشغّل أي نموذج HuggingFace متوافق مع Xenova محليًا عبر WebAssembly أو WebGPU.",
   "settings.ai.transformers.modelId": "معرّف النموذج ‏(HuggingFace)",
diff --git a/locales/de/settings.json b/locales/de/settings.json
index 7ac4253b..6db15e27 100644
--- a/locales/de/settings.json
+++ b/locales/de/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Kreativ",
   "settings.ai.temperature.precise": "0 – Präzise",
   "settings.ai.testConnection": "Verbindung testen",
+  "settings.ai.testError.noApiKey": "Kein {{provider}} API-Schlüssel gesetzt",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Zeitüberschreitung bei der Verbindung ({{url}})",
+  "settings.ai.testError.unreachable": "Nicht erreichbar ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Lokale Serververnetzung nicht verfügbar: Das Desktop-HTTP-Plugin konnte nicht geladen werden.",
+  "settings.ai.testError.desktopRequired": "Ollama und lokale OpenAI-kompatible Server sind nur in der Desktop-App verfügbar. Browser blockieren direkte Verbindungen von Webseiten zu localhost (CORS und Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude benötigt einen Backend-Proxy (CORS-Einschränkung)",
+  "settings.ai.testError.noWebgpu": "WebGPU in diesem Browser nicht verfügbar — WebLLM benötigt WebGPU (Chrome/Edge verwenden oder Flags aktivieren).",
+  "settings.ai.testError.unknownProvider": "Unbekannter Anbieter",
+  "settings.ai.testError.unexpected": "Ein unerwarteter Fehler ist aufgetreten. Bitte versuchen Sie es erneut.",
   "settings.ai.title": "KI-Konfiguration",
   "settings.ai.transformers.hint": "Führt jedes Xenova-kompatible HuggingFace-Modell lokal via WebAssembly oder WebGPU aus.",
   "settings.ai.transformers.modelId": "Modell-ID (HuggingFace)",
diff --git a/locales/el/settings.json b/locales/el/settings.json
index 238735ef..c66e4d32 100644
--- a/locales/el/settings.json
+++ b/locales/el/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Δημιουργικό",
   "settings.ai.temperature.precise": "0 – Ακριβές",
   "settings.ai.testConnection": "Δοκιμαστική σύνδεση",
+  "settings.ai.testError.noApiKey": "Δεν έχει οριστεί κλειδί API {{provider}}",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Λήξη χρονικού ορίου σύνδεσης ({{url}})",
+  "settings.ai.testError.unreachable": "Μη προσβάσιμο ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Το δίκτυο τοπικού διακομιστή δεν είναι διαθέσιμο: η πρόσθετη λειτουργία HTTP επιφάνειας εργασίας απέτυχε να φορτωθεί.",
+  "settings.ai.testError.desktopRequired": "Το Ollama και οι τοπικοί διακομιστές συμβατοί με OpenAI είναι διαθέσιμοι μόνο στην εφαρμογή επιφάνειας εργασίας. Οι φυλλομετρητές αποκλείουν άμεσες συνδέσεις από ιστοσελίδες προς το localhost (CORS και Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Το Claude απαιτεί διακομιστή μεσολάβησης backend (περιορισμός CORS)",
+  "settings.ai.testError.noWebgpu": "Το WebGPU δεν είναι διαθέσιμο σε αυτό το πρόγραμμα περιήγησης — το WebLLM απαιτεί WebGPU (δοκιμάστε Chrome/Edge ή ενεργοποιήστε τις σημαίες).",
+  "settings.ai.testError.unknownProvider": "Άγνωστος πάροχος",
+  "settings.ai.testError.unexpected": "Παρουσιάστηκε ένα μη αναμενόμενο σφάλμα. Δοκιμάστε ξανά.",
   "settings.ai.title": "Διαμόρφωση AI",
   "settings.ai.transformers.hint": "Εκτελεί οποιοδήποτε μοντέλο HuggingFace συμβατό με Xenova τοπικά μέσω WebAssembly ή WebGPU.",
   "settings.ai.transformers.modelId": "Αναγνωριστικό μοντέλου (HuggingFace)",
diff --git a/locales/en/settings.json b/locales/en/settings.json
index 355b0b0f..4e8f42f9 100644
--- a/locales/en/settings.json
+++ b/locales/en/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Creative",
   "settings.ai.temperature.precise": "0 – Precise",
   "settings.ai.testConnection": "Test connection",
+  "settings.ai.testError.noApiKey": "No {{provider}} API key set",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Connection timed out ({{url}})",
+  "settings.ai.testError.unreachable": "Not reachable ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Local server networking is unavailable: the desktop HTTP plugin failed to load.",
+  "settings.ai.testError.desktopRequired": "Ollama and local OpenAI-compatible servers are only available in the desktop app. Browsers block direct connections from web pages to localhost (CORS and Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude requires a backend proxy (CORS restriction)",
+  "settings.ai.testError.noWebgpu": "WebGPU unavailable in this browser — WebLLM needs WebGPU (try Chrome/Edge or enable flags).",
+  "settings.ai.testError.unknownProvider": "Unknown provider",
+  "settings.ai.testError.unexpected": "An unexpected error occurred. Please try again.",
   "settings.ai.title": "AI Configuration",
   "settings.ai.transformers.hint": "Runs any Xenova-compatible HuggingFace model locally via WebAssembly or WebGPU.",
   "settings.ai.transformers.modelId": "Model ID (HuggingFace)",
diff --git a/locales/es/settings.json b/locales/es/settings.json
index 35f65ece..fd165e92 100644
--- a/locales/es/settings.json
+++ b/locales/es/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Creativo",
   "settings.ai.temperature.precise": "0 – Preciso",
   "settings.ai.testConnection": "Probar conexión",
+  "settings.ai.testError.noApiKey": "No se ha configurado una clave de API de {{provider}}",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Tiempo de espera agotado ({{url}})",
+  "settings.ai.testError.unreachable": "No accesible ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "La red de servidores locales no está disponible: el plugin HTTP de escritorio no se pudo cargar.",
+  "settings.ai.testError.desktopRequired": "Ollama y los servidores locales compatibles con OpenAI solo están disponibles en la aplicación de escritorio. Los navegadores bloquean las conexiones directas desde páginas web a localhost (CORS y Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude requiere un proxy de backend (restricción CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU no disponible en este navegador — WebLLM necesita WebGPU (prueba Chrome/Edge o activa las opciones).",
+  "settings.ai.testError.unknownProvider": "Proveedor desconocido",
+  "settings.ai.testError.unexpected": "Se produjo un error inesperado. Inténtalo de nuevo.",
   "settings.ai.title": "Configuración de IA",
   "settings.ai.transformers.hint": "Ejecuta cualquier modelo HuggingFace compatible con Xenova localmente vía WebAssembly o WebGPU.",
   "settings.ai.transformers.modelId": "ID del modelo (HuggingFace)",
diff --git a/locales/eu/settings.json b/locales/eu/settings.json
index aabe4db1..4fceeeff 100644
--- a/locales/eu/settings.json
+++ b/locales/eu/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Sormena",
   "settings.ai.temperature.precise": "0 – Zehatza",
   "settings.ai.testConnection": "Probatu konexioa",
+  "settings.ai.testError.noApiKey": "Ez da {{provider}} API gakorik ezarri",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Konexioak denbora-muga gainditu du ({{url}})",
+  "settings.ai.testError.unreachable": "Ezin da iritsi ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Zerbitzari lokalaren sarea ez dago erabilgarri: mahaigaineko HTTP pluginak huts egin du kargatzean.",
+  "settings.ai.testError.desktopRequired": "Ollama eta OpenAI-rekin bateragarriak diren zerbitzari lokalak mahaigaineko aplikazioan bakarrik daude erabilgarri. Nabigatzaileek web orrialdeetatik localhost-era zuzeneko konexioak blokeatzen dituzte (CORS eta Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude-k backend proxy bat behar du (CORS murrizketa)",
+  "settings.ai.testError.noWebgpu": "WebGPU ez dago erabilgarri nabigatzaile honetan — WebLLM-ek WebGPU behar du (probatu Chrome/Edge edo gaitu banderak).",
+  "settings.ai.testError.unknownProvider": "Hornitzaile ezezaguna",
+  "settings.ai.testError.unexpected": "Ustekabeko errorea gertatu da. Saiatu berriro.",
   "settings.ai.title": "AI konfigurazioa",
   "settings.ai.transformers.hint": "Xenova-rekin bateragarria den HuggingFace modelo edozein exekutatzen du lokalean WebAssembly edo WebGPU bidez.",
   "settings.ai.transformers.modelId": "Modelo ID (HuggingFace)",
diff --git a/locales/fa/settings.json b/locales/fa/settings.json
index 48e6d490..87f6e34c 100644
--- a/locales/fa/settings.json
+++ b/locales/fa/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2- خلاق",
   "settings.ai.temperature.precise": "0 - دقیق",
   "settings.ai.testConnection": "تست اتصال",
+  "settings.ai.testError.noApiKey": "کلید API {{provider}} تنظیم نشده است",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "زمان اتصال به پایان رسید ({{url}})",
+  "settings.ai.testError.unreachable": "در دسترس نیست ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "شبکه‌سازی سرور محلی در دسترس نیست: بارگذاری افزونه HTTP دسکتاپ ناموفق بود.",
+  "settings.ai.testError.desktopRequired": "Ollama و سرورهای محلی سازگار با OpenAI فقط در برنامه دسکتاپ در دسترس هستند. مرورگرها اتصالات مستقیم از صفحات وب به localhost را مسدود می‌کنند (CORS و Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude به یک پراکسی بک‌اند نیاز دارد (محدودیت CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU در این مرورگر در دسترس نیست — WebLLM به WebGPU نیاز دارد (Chrome/Edge را امتحان کنید یا پرچم‌ها را فعال کنید).",
+  "settings.ai.testError.unknownProvider": "ارائه‌دهنده ناشناخته",
+  "settings.ai.testError.unexpected": "خطای غیرمنتظره‌ای رخ داد. لطفاً دوباره امتحان کنید.",
   "settings.ai.title": "پیکربندی هوش مصنوعی",
   "settings.ai.transformers.hint": "هر مدل HuggingFace سازگار با Xenova را به صورت محلی از طریق WebAssembly یا WebGPU اجرا می کند.",
   "settings.ai.transformers.modelId": "شناسه مدل (HuggingFace)",
diff --git a/locales/fi/settings.json b/locales/fi/settings.json
index 62843283..f5fc1e71 100644
--- a/locales/fi/settings.json
+++ b/locales/fi/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Luova",
   "settings.ai.temperature.precise": "0 – Tarkka",
   "settings.ai.testConnection": "Testaa yhteys",
+  "settings.ai.testError.noApiKey": "{{provider}}-API-avainta ei ole asetettu",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Yhteys aikakatkaistiin ({{url}})",
+  "settings.ai.testError.unreachable": "Ei tavoiteta ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Paikallisen palvelimen verkko ei ole käytettävissä: työpöydän HTTP-lisäosan lataus epäonnistui.",
+  "settings.ai.testError.desktopRequired": "Ollama ja paikalliset OpenAI-yhteensopivat palvelimet ovat käytettävissä vain työpöytäsovelluksessa. Selaimet estävät suorat yhteydet verkkosivuilta osoitteeseen localhost (CORS ja Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude vaatii taustajärjestelmän välityspalvelimen (CORS-rajoitus)",
+  "settings.ai.testError.noWebgpu": "WebGPU ei ole käytettävissä tässä selaimessa — WebLLM tarvitsee WebGPU:n (kokeile Chromea/Edgeä tai ota liput käyttöön).",
+  "settings.ai.testError.unknownProvider": "Tuntematon palveluntarjoaja",
+  "settings.ai.testError.unexpected": "Tapahtui odottamaton virhe. Yritä uudelleen.",
   "settings.ai.title": "AI-kokoonpano",
   "settings.ai.transformers.hint": "Suorittaa mitä tahansa Xenova-yhteensopivaa HuggingFace-mallia paikallisesti WebAssemblyn tai WebGPU:n kautta.",
   "settings.ai.transformers.modelId": "Mallin tunnus (HuggingFace)",
diff --git a/locales/fr/settings.json b/locales/fr/settings.json
index c79ca387..0ce44366 100644
--- a/locales/fr/settings.json
+++ b/locales/fr/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Créatif",
   "settings.ai.temperature.precise": "0 – Précis",
   "settings.ai.testConnection": "Tester la connexion",
+  "settings.ai.testError.noApiKey": "Aucune clé API {{provider}} définie",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Délai de connexion dépassé ({{url}})",
+  "settings.ai.testError.unreachable": "Injoignable ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Le réseau des serveurs locaux est indisponible : le plugin HTTP du bureau n'a pas pu se charger.",
+  "settings.ai.testError.desktopRequired": "Ollama et les serveurs locaux compatibles OpenAI ne sont disponibles que dans l'application de bureau. Les navigateurs bloquent les connexions directes des pages web vers localhost (CORS et Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude nécessite un proxy backend (restriction CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU indisponible dans ce navigateur — WebLLM nécessite WebGPU (essayez Chrome/Edge ou activez les indicateurs).",
+  "settings.ai.testError.unknownProvider": "Fournisseur inconnu",
+  "settings.ai.testError.unexpected": "Une erreur inattendue s'est produite. Veuillez réessayer.",
   "settings.ai.title": "Configuration de l'IA",
   "settings.ai.transformers.hint": "Exécute tout modèle HuggingFace compatible Xenova localement via WebAssembly ou WebGPU.",
   "settings.ai.transformers.modelId": "ID du modèle (HuggingFace)",
diff --git a/locales/he/settings.json b/locales/he/settings.json
index 94a84010..7294a6c6 100644
--- a/locales/he/settings.json
+++ b/locales/he/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – יצירתי",
   "settings.ai.temperature.precise": "0 – מדויק",
   "settings.ai.testConnection": "בדיקת חיבור",
+  "settings.ai.testError.noApiKey": "לא הוגדר מפתח API עבור {{provider}}",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "פסק הזמן של החיבור פג ({{url}})",
+  "settings.ai.testError.unreachable": "לא ניתן להגיע ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "רשת השרת המקומי אינה זמינה: תוסף ה-HTTP של שולחן העבודה נכשל בטעינה.",
+  "settings.ai.testError.desktopRequired": "Ollama ושרתים מקומיים תואמי OpenAI זמינים רק באפליקציית שולחן העבודה. דפדפנים חוסמים חיבורים ישירים מדפי אינטרנט אל localhost (CORS ו-Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude דורש שרת proxy (הגבלת CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU אינו זמין בדפדפן זה — WebLLM דורש WebGPU (נסה Chrome/Edge או הפעל דגלים).",
+  "settings.ai.testError.unknownProvider": "ספק לא ידוע",
+  "settings.ai.testError.unexpected": "אירעה שגיאה בלתי צפויה. נסה שוב.",
   "settings.ai.title": "תצורת AI",
   "settings.ai.transformers.hint": "מריץ כל מודל HuggingFace תואם‑Xenova מקומית דרך WebAssembly או WebGPU.",
   "settings.ai.transformers.modelId": "מזהה מודל ‏(HuggingFace)",
diff --git a/locales/hu/settings.json b/locales/hu/settings.json
index 9982e37b..1dd06f5b 100644
--- a/locales/hu/settings.json
+++ b/locales/hu/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Kreatív",
   "settings.ai.temperature.precise": "0 – Pontos",
   "settings.ai.testConnection": "Kapcsolat tesztelése",
+  "settings.ai.testError.noApiKey": "Nincs beállítva {{provider}} API-kulcs",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "A kapcsolat túllépte az időkorlátot ({{url}})",
+  "settings.ai.testError.unreachable": "Nem érhető el ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "A helyi kiszolgálóhálózat nem érhető el: az asztali HTTP bővítmény betöltése sikertelen.",
+  "settings.ai.testError.desktopRequired": "Az Ollama és a helyi OpenAI-kompatibilis kiszolgálók csak az asztali alkalmazásban érhetők el. A böngészők blokkolják a weboldalak közvetlen kapcsolatait a localhosttal (CORS és Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "A Claude háttérproxyt igényel (CORS-korlátozás)",
+  "settings.ai.testError.noWebgpu": "A WebGPU nem érhető el ebben a böngészőben — a WebLLM-nek WebGPU-ra van szüksége (próbáld a Chrome-ot/Edge-et, vagy engedélyezd a jelzőket).",
+  "settings.ai.testError.unknownProvider": "Ismeretlen szolgáltató",
+  "settings.ai.testError.unexpected": "Váratlan hiba történt. Kérjük, próbálja újra.",
   "settings.ai.title": "AI konfiguráció",
   "settings.ai.transformers.hint": "Bármilyen Xenova-kompatibilis HuggingFace modellt futtat helyileg a WebAssembly vagy WebGPU segítségével.",
   "settings.ai.transformers.modelId": "Modellazonosító (HuggingFace)",
diff --git a/locales/is/settings.json b/locales/is/settings.json
index d794ae0f..c37b1270 100644
--- a/locales/is/settings.json
+++ b/locales/is/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 - Skapandi",
   "settings.ai.temperature.precise": "0 - Nákvæmt",
   "settings.ai.testConnection": "Prófaðu tengingu",
+  "settings.ai.testError.noApiKey": "Enginn {{provider}} API lykill stilltur",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Tengingin féll á tíma ({{url}})",
+  "settings.ai.testError.unreachable": "Næst ekki ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Staðbundið netþjónasamband ekki tiltækt: ekki tókst að hlaða HTTP-viðbót skjáborðs.",
+  "settings.ai.testError.desktopRequired": "Ollama og staðbundnir OpenAI-samhæfir netþjónar eru aðeins tiltækir í skjáborðsforritinu. Vafrar loka fyrir bein tengsl frá vefsíðum við localhost (CORS og Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude þarfnast bakenda-milliþjóns (CORS-takmörkun)",
+  "settings.ai.testError.noWebgpu": "WebGPU er ekki tiltækt í þessum vafra — WebLLM þarfnast WebGPU (prófaðu Chrome/Edge eða virkjaðu flögg).",
+  "settings.ai.testError.unknownProvider": "Óþekktur veitandi",
+  "settings.ai.testError.unexpected": "Óvænt villa kom upp. Reyndu aftur.",
   "settings.ai.title": "AI stillingar",
   "settings.ai.transformers.hint": "Keyrir hvaða Xenova-samhæfða HuggingFace líkan sem er á staðnum í gegnum WebAssembly eða WebGPU.",
   "settings.ai.transformers.modelId": "Auðkenni fyrirmyndar (HuggingFace)",
diff --git a/locales/it/settings.json b/locales/it/settings.json
index a2f8eb88..34c40da8 100644
--- a/locales/it/settings.json
+++ b/locales/it/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Creativo",
   "settings.ai.temperature.precise": "0 – Preciso",
   "settings.ai.testConnection": "Testa la connessione",
+  "settings.ai.testError.noApiKey": "Nessuna chiave API {{provider}} impostata",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Connessione scaduta ({{url}})",
+  "settings.ai.testError.unreachable": "Non raggiungibile ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Rete dei server locali non disponibile: il plugin HTTP desktop non è stato caricato.",
+  "settings.ai.testError.desktopRequired": "Ollama e i server locali compatibili con OpenAI sono disponibili solo nell'app desktop. I browser bloccano le connessioni dirette dalle pagine web a localhost (CORS e Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude richiede un proxy backend (restrizione CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU non disponibile in questo browser — WebLLM richiede WebGPU (prova Chrome/Edge o abilita i flag).",
+  "settings.ai.testError.unknownProvider": "Provider sconosciuto",
+  "settings.ai.testError.unexpected": "Si è verificato un errore imprevisto. Riprova.",
   "settings.ai.title": "Configurazione IA",
   "settings.ai.transformers.hint": "Esegue qualsiasi modello HuggingFace compatibile con Xenova localmente via WebAssembly o WebGPU.",
   "settings.ai.transformers.modelId": "ID modello (HuggingFace)",
diff --git a/locales/ja/settings.json b/locales/ja/settings.json
index a287595c..2f2cc921 100644
--- a/locales/ja/settings.json
+++ b/locales/ja/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – クリエイティブ",
   "settings.ai.temperature.precise": "0 – 正確",
   "settings.ai.testConnection": "テスト接続",
+  "settings.ai.testError.noApiKey": "{{provider}} の API キーが設定されていません",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "接続がタイムアウトしました ({{url}})",
+  "settings.ai.testError.unreachable": "到達できません ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "ローカルサーバーへの接続が利用できません: デスクトップ HTTP プラグインの読み込みに失敗しました。",
+  "settings.ai.testError.desktopRequired": "Ollama およびローカルの OpenAI 互換サーバーはデスクトップアプリでのみ利用できます。ブラウザは Web ページから localhost への直接接続をブロックします（CORS および Private Network Access）。",
+  "settings.ai.testError.backendProxyRequired": "Claude にはバックエンドプロキシが必要です（CORS 制限）",
+  "settings.ai.testError.noWebgpu": "このブラウザでは WebGPU が利用できません — WebLLM には WebGPU が必要です（Chrome/Edge を試すかフラグを有効にしてください）。",
+  "settings.ai.testError.unknownProvider": "不明なプロバイダー",
+  "settings.ai.testError.unexpected": "予期しないエラーが発生しました。もう一度お試しください。",
   "settings.ai.title": "AIの構成",
   "settings.ai.transformers.hint": "WebAssembly または WebGPU を介して Xenova 互換の HuggingFace モデルをローカルで実行します。",
   "settings.ai.transformers.modelId": "モデルID (HuggingFace)",
diff --git a/locales/ko/settings.json b/locales/ko/settings.json
index 648a8679..93b8d290 100644
--- a/locales/ko/settings.json
+++ b/locales/ko/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – 창의적",
   "settings.ai.temperature.precise": "0 – 정확함",
   "settings.ai.testConnection": "연결 테스트",
+  "settings.ai.testError.noApiKey": "{{provider}} API 키가 설정되지 않았습니다",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "연결 시간이 초과되었습니다 ({{url}})",
+  "settings.ai.testError.unreachable": "연결할 수 없습니다 ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "로컬 서버 네트워킹을 사용할 수 없습니다: 데스크톱 HTTP 플러그인을 불러오지 못했습니다.",
+  "settings.ai.testError.desktopRequired": "Ollama 및 로컬 OpenAI 호환 서버는 데스크톱 앱에서만 사용할 수 있습니다. 브라우저는 웹 페이지에서 localhost로의 직접 연결을 차단합니다(CORS 및 Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude에는 백엔드 프록시가 필요합니다(CORS 제한)",
+  "settings.ai.testError.noWebgpu": "이 브라우저에서는 WebGPU를 사용할 수 없습니다 — WebLLM에는 WebGPU가 필요합니다(Chrome/Edge를 사용하거나 플래그를 활성화하세요).",
+  "settings.ai.testError.unknownProvider": "알 수 없는 제공자",
+  "settings.ai.testError.unexpected": "예기치 않은 오류가 발생했습니다. 다시 시도해 주세요.",
   "settings.ai.title": "AI 구성",
   "settings.ai.transformers.hint": "WebAssembly 또는 WebGPU를 통해 Xenova 호환 HuggingFace 모델을 로컬로 실행합니다.",
   "settings.ai.transformers.modelId": "모델 ID(HuggingFace)",
diff --git a/locales/pt/settings.json b/locales/pt/settings.json
index a9f660ad..335fe588 100644
--- a/locales/pt/settings.json
+++ b/locales/pt/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Criativo",
   "settings.ai.temperature.precise": "0 – Preciso",
   "settings.ai.testConnection": "Conexão de teste",
+  "settings.ai.testError.noApiKey": "Nenhuma chave de API {{provider}} definida",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Tempo limite de conexão excedido ({{url}})",
+  "settings.ai.testError.unreachable": "Inacessível ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "A rede do servidor local não está disponível: o plugin HTTP de ambiente de trabalho falhou ao carregar.",
+  "settings.ai.testError.desktopRequired": "O Ollama e os servidores locais compatíveis com OpenAI só estão disponíveis na aplicação de ambiente de trabalho. Os navegadores bloqueiam ligações diretas de páginas web ao localhost (CORS e Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "O Claude requer um proxy de backend (restrição CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU indisponível neste navegador — o WebLLM necessita de WebGPU (experimente o Chrome/Edge ou ative as flags).",
+  "settings.ai.testError.unknownProvider": "Fornecedor desconhecido",
+  "settings.ai.testError.unexpected": "Ocorreu um erro inesperado. Tente novamente.",
   "settings.ai.title": "IA Configuration",
   "settings.ai.transformers.hint": "Executa qualquer modelo HuggingFace compatível com Xenova localmente via WebAssembly ou WebGPU.",
   "settings.ai.transformers.modelId": "ID do modelo (HuggingFace)",
diff --git a/locales/ru/settings.json b/locales/ru/settings.json
index 92ce35fe..ee30af12 100644
--- a/locales/ru/settings.json
+++ b/locales/ru/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Креатив",
   "settings.ai.temperature.precise": "0 – Точный",
   "settings.ai.testConnection": "Тестовое соединение",
+  "settings.ai.testError.noApiKey": "Не задан ключ API {{provider}}",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Тайм-аут соединения ({{url}})",
+  "settings.ai.testError.unreachable": "Недоступно ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Сеть локального сервера недоступна: не удалось загрузить настольный HTTP-плагин.",
+  "settings.ai.testError.desktopRequired": "Ollama и локальные серверы, совместимые с OpenAI, доступны только в настольном приложении. Браузеры блокируют прямые подключения с веб-страниц к localhost (CORS и Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude требует backend-прокси (ограничение CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU недоступен в этом браузере — WebLLM требует WebGPU (попробуйте Chrome/Edge или включите флаги).",
+  "settings.ai.testError.unknownProvider": "Неизвестный провайдер",
+  "settings.ai.testError.unexpected": "Произошла непредвиденная ошибка. Повторите попытку.",
   "settings.ai.title": "Конфигурация ИИ",
   "settings.ai.transformers.hint": "Запускает любую Xenova-совместимую модель HuggingFace локально через WebAssembly или WebGPU.",
   "settings.ai.transformers.modelId": "Идентификатор модели (HuggingFace)",
diff --git a/locales/sv/settings.json b/locales/sv/settings.json
index 321e5abb..1f089506 100644
--- a/locales/sv/settings.json
+++ b/locales/sv/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – Kreativt",
   "settings.ai.temperature.precise": "0 – Exakt",
   "settings.ai.testConnection": "Testa anslutningen",
+  "settings.ai.testError.noApiKey": "Ingen {{provider}}-API-nyckel angiven",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Anslutningen fick timeout ({{url}})",
+  "settings.ai.testError.unreachable": "Ej nåbar ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Lokal servernätverksanslutning ej tillgänglig: skrivbordets HTTP-plugin misslyckades med att läsas in.",
+  "settings.ai.testError.desktopRequired": "Ollama och lokala OpenAI-kompatibla servrar är endast tillgängliga i skrivbordsappen. Webbläsare blockerar direkta anslutningar från webbsidor till localhost (CORS och Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude kräver en backend-proxy (CORS-begränsning)",
+  "settings.ai.testError.noWebgpu": "WebGPU är inte tillgängligt i den här webbläsaren — WebLLM behöver WebGPU (prova Chrome/Edge eller aktivera flaggor).",
+  "settings.ai.testError.unknownProvider": "Okänd leverantör",
+  "settings.ai.testError.unexpected": "Ett oväntat fel inträffade. Försök igen.",
   "settings.ai.title": "AI-konfiguration",
   "settings.ai.transformers.hint": "Kör valfri Xenova-kompatibel HuggingFace-modell lokalt via WebAssembly eller WebGPU.",
   "settings.ai.transformers.modelId": "Modell-ID (HuggingFace)",
diff --git a/locales/zh/settings.json b/locales/zh/settings.json
index e4be0fa3..4e2c53a8 100644
--- a/locales/zh/settings.json
+++ b/locales/zh/settings.json
@@ -247,6 +247,16 @@
   "settings.ai.temperature.creative": "2 – 创意",
   "settings.ai.temperature.precise": "0 – 精确",
   "settings.ai.testConnection": "测试连接",
+  "settings.ai.testError.noApiKey": "未设置 {{provider}} API 密钥",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "连接超时 ({{url}})",
+  "settings.ai.testError.unreachable": "无法访问 ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "本地服务器网络不可用：桌面 HTTP 插件加载失败。",
+  "settings.ai.testError.desktopRequired": "Ollama 和本地 OpenAI 兼容服务器仅在桌面应用中可用。浏览器会阻止网页直接连接到 localhost（CORS 和 Private Network Access）。",
+  "settings.ai.testError.backendProxyRequired": "Claude 需要后端代理（CORS 限制）",
+  "settings.ai.testError.noWebgpu": "此浏览器不支持 WebGPU — WebLLM 需要 WebGPU（请尝试 Chrome/Edge 或启用相关标志）。",
+  "settings.ai.testError.unknownProvider": "未知的提供商",
+  "settings.ai.testError.unexpected": "发生意外错误。请重试。",
   "settings.ai.title": "人工智能配置",
   "settings.ai.transformers.hint": "通过 WebAssembly 或 WebGPU 在本地运行任何 Xenova 兼容的 HuggingFace 模型。",
   "settings.ai.transformers.modelId": "型号 ID (HuggingFace)",
diff --git a/public/locales/ar/bundle.json b/public/locales/ar/bundle.json
index 73b7035c..46837986 100644
--- a/public/locales/ar/bundle.json
+++ b/public/locales/ar/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – إبداعي",
   "settings.ai.temperature.precise": "0 – دقيق",
   "settings.ai.testConnection": "اختبار الاتصال",
+  "settings.ai.testError.noApiKey": "لم يتم تعيين مفتاح API لـ {{provider}}",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "انتهت مهلة الاتصال ({{url}})",
+  "settings.ai.testError.unreachable": "غير قابل للوصول ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "شبكة الخادم المحلي غير متوفرة: فشل تحميل مكوّن HTTP لسطح المكتب.",
+  "settings.ai.testError.desktopRequired": "تتوفر Ollama والخوادم المحلية المتوافقة مع OpenAI فقط في تطبيق سطح المكتب. تحظر المتصفحات الاتصالات المباشرة من صفحات الويب بـ localhost (CORS وPrivate Network Access).",
+  "settings.ai.testError.backendProxyRequired": "يتطلب Claude وكيل خلفية (قيود CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU غير متوفر في هذا المتصفح — يحتاج WebLLM إلى WebGPU (جرّب Chrome/Edge أو فعّل العلامات).",
+  "settings.ai.testError.unknownProvider": "مزوّد غير معروف",
+  "settings.ai.testError.unexpected": "حدث خطأ غير متوقع. يرجى المحاولة مرة أخرى.",
   "settings.ai.title": "إعداد الذكاء الاصطناعي",
   "settings.ai.transformers.hint": "يشغّل أي نموذج HuggingFace متوافق مع Xenova محليًا عبر WebAssembly أو WebGPU.",
   "settings.ai.transformers.modelId": "معرّف النموذج ‏(HuggingFace)",
diff --git a/public/locales/de/bundle.json b/public/locales/de/bundle.json
index 0275e289..e6115b28 100644
--- a/public/locales/de/bundle.json
+++ b/public/locales/de/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Kreativ",
   "settings.ai.temperature.precise": "0 – Präzise",
   "settings.ai.testConnection": "Verbindung testen",
+  "settings.ai.testError.noApiKey": "Kein {{provider}} API-Schlüssel gesetzt",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Zeitüberschreitung bei der Verbindung ({{url}})",
+  "settings.ai.testError.unreachable": "Nicht erreichbar ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Lokale Serververnetzung nicht verfügbar: Das Desktop-HTTP-Plugin konnte nicht geladen werden.",
+  "settings.ai.testError.desktopRequired": "Ollama und lokale OpenAI-kompatible Server sind nur in der Desktop-App verfügbar. Browser blockieren direkte Verbindungen von Webseiten zu localhost (CORS und Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude benötigt einen Backend-Proxy (CORS-Einschränkung)",
+  "settings.ai.testError.noWebgpu": "WebGPU in diesem Browser nicht verfügbar — WebLLM benötigt WebGPU (Chrome/Edge verwenden oder Flags aktivieren).",
+  "settings.ai.testError.unknownProvider": "Unbekannter Anbieter",
+  "settings.ai.testError.unexpected": "Ein unerwarteter Fehler ist aufgetreten. Bitte versuchen Sie es erneut.",
   "settings.ai.title": "KI-Konfiguration",
   "settings.ai.transformers.hint": "Führt jedes Xenova-kompatible HuggingFace-Modell lokal via WebAssembly oder WebGPU aus.",
   "settings.ai.transformers.modelId": "Modell-ID (HuggingFace)",
diff --git a/public/locales/el/bundle.json b/public/locales/el/bundle.json
index 278a289b..297adeea 100644
--- a/public/locales/el/bundle.json
+++ b/public/locales/el/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Δημιουργικό",
   "settings.ai.temperature.precise": "0 – Ακριβές",
   "settings.ai.testConnection": "Δοκιμαστική σύνδεση",
+  "settings.ai.testError.noApiKey": "Δεν έχει οριστεί κλειδί API {{provider}}",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Λήξη χρονικού ορίου σύνδεσης ({{url}})",
+  "settings.ai.testError.unreachable": "Μη προσβάσιμο ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Το δίκτυο τοπικού διακομιστή δεν είναι διαθέσιμο: η πρόσθετη λειτουργία HTTP επιφάνειας εργασίας απέτυχε να φορτωθεί.",
+  "settings.ai.testError.desktopRequired": "Το Ollama και οι τοπικοί διακομιστές συμβατοί με OpenAI είναι διαθέσιμοι μόνο στην εφαρμογή επιφάνειας εργασίας. Οι φυλλομετρητές αποκλείουν άμεσες συνδέσεις από ιστοσελίδες προς το localhost (CORS και Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Το Claude απαιτεί διακομιστή μεσολάβησης backend (περιορισμός CORS)",
+  "settings.ai.testError.noWebgpu": "Το WebGPU δεν είναι διαθέσιμο σε αυτό το πρόγραμμα περιήγησης — το WebLLM απαιτεί WebGPU (δοκιμάστε Chrome/Edge ή ενεργοποιήστε τις σημαίες).",
+  "settings.ai.testError.unknownProvider": "Άγνωστος πάροχος",
+  "settings.ai.testError.unexpected": "Παρουσιάστηκε ένα μη αναμενόμενο σφάλμα. Δοκιμάστε ξανά.",
   "settings.ai.title": "Διαμόρφωση AI",
   "settings.ai.transformers.hint": "Εκτελεί οποιοδήποτε μοντέλο HuggingFace συμβατό με Xenova τοπικά μέσω WebAssembly ή WebGPU.",
   "settings.ai.transformers.modelId": "Αναγνωριστικό μοντέλου (HuggingFace)",
diff --git a/public/locales/en/bundle.json b/public/locales/en/bundle.json
index fdaf0759..19e580ff 100644
--- a/public/locales/en/bundle.json
+++ b/public/locales/en/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Creative",
   "settings.ai.temperature.precise": "0 – Precise",
   "settings.ai.testConnection": "Test connection",
+  "settings.ai.testError.noApiKey": "No {{provider}} API key set",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Connection timed out ({{url}})",
+  "settings.ai.testError.unreachable": "Not reachable ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Local server networking is unavailable: the desktop HTTP plugin failed to load.",
+  "settings.ai.testError.desktopRequired": "Ollama and local OpenAI-compatible servers are only available in the desktop app. Browsers block direct connections from web pages to localhost (CORS and Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude requires a backend proxy (CORS restriction)",
+  "settings.ai.testError.noWebgpu": "WebGPU unavailable in this browser — WebLLM needs WebGPU (try Chrome/Edge or enable flags).",
+  "settings.ai.testError.unknownProvider": "Unknown provider",
+  "settings.ai.testError.unexpected": "An unexpected error occurred. Please try again.",
   "settings.ai.title": "AI Configuration",
   "settings.ai.transformers.hint": "Runs any Xenova-compatible HuggingFace model locally via WebAssembly or WebGPU.",
   "settings.ai.transformers.modelId": "Model ID (HuggingFace)",
diff --git a/public/locales/es/bundle.json b/public/locales/es/bundle.json
index b3c300b8..3d0027ff 100644
--- a/public/locales/es/bundle.json
+++ b/public/locales/es/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Creativo",
   "settings.ai.temperature.precise": "0 – Preciso",
   "settings.ai.testConnection": "Probar conexión",
+  "settings.ai.testError.noApiKey": "No se ha configurado una clave de API de {{provider}}",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Tiempo de espera agotado ({{url}})",
+  "settings.ai.testError.unreachable": "No accesible ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "La red de servidores locales no está disponible: el plugin HTTP de escritorio no se pudo cargar.",
+  "settings.ai.testError.desktopRequired": "Ollama y los servidores locales compatibles con OpenAI solo están disponibles en la aplicación de escritorio. Los navegadores bloquean las conexiones directas desde páginas web a localhost (CORS y Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude requiere un proxy de backend (restricción CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU no disponible en este navegador — WebLLM necesita WebGPU (prueba Chrome/Edge o activa las opciones).",
+  "settings.ai.testError.unknownProvider": "Proveedor desconocido",
+  "settings.ai.testError.unexpected": "Se produjo un error inesperado. Inténtalo de nuevo.",
   "settings.ai.title": "Configuración de IA",
   "settings.ai.transformers.hint": "Ejecuta cualquier modelo HuggingFace compatible con Xenova localmente vía WebAssembly o WebGPU.",
   "settings.ai.transformers.modelId": "ID del modelo (HuggingFace)",
diff --git a/public/locales/eu/bundle.json b/public/locales/eu/bundle.json
index ad6f5f4b..bf6e485e 100644
--- a/public/locales/eu/bundle.json
+++ b/public/locales/eu/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Sormena",
   "settings.ai.temperature.precise": "0 – Zehatza",
   "settings.ai.testConnection": "Probatu konexioa",
+  "settings.ai.testError.noApiKey": "Ez da {{provider}} API gakorik ezarri",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Konexioak denbora-muga gainditu du ({{url}})",
+  "settings.ai.testError.unreachable": "Ezin da iritsi ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Zerbitzari lokalaren sarea ez dago erabilgarri: mahaigaineko HTTP pluginak huts egin du kargatzean.",
+  "settings.ai.testError.desktopRequired": "Ollama eta OpenAI-rekin bateragarriak diren zerbitzari lokalak mahaigaineko aplikazioan bakarrik daude erabilgarri. Nabigatzaileek web orrialdeetatik localhost-era zuzeneko konexioak blokeatzen dituzte (CORS eta Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude-k backend proxy bat behar du (CORS murrizketa)",
+  "settings.ai.testError.noWebgpu": "WebGPU ez dago erabilgarri nabigatzaile honetan — WebLLM-ek WebGPU behar du (probatu Chrome/Edge edo gaitu banderak).",
+  "settings.ai.testError.unknownProvider": "Hornitzaile ezezaguna",
+  "settings.ai.testError.unexpected": "Ustekabeko errorea gertatu da. Saiatu berriro.",
   "settings.ai.title": "AI konfigurazioa",
   "settings.ai.transformers.hint": "Xenova-rekin bateragarria den HuggingFace modelo edozein exekutatzen du lokalean WebAssembly edo WebGPU bidez.",
   "settings.ai.transformers.modelId": "Modelo ID (HuggingFace)",
diff --git a/public/locales/fa/bundle.json b/public/locales/fa/bundle.json
index ec6c50bf..405c580e 100644
--- a/public/locales/fa/bundle.json
+++ b/public/locales/fa/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2- خلاق",
   "settings.ai.temperature.precise": "0 - دقیق",
   "settings.ai.testConnection": "تست اتصال",
+  "settings.ai.testError.noApiKey": "کلید API {{provider}} تنظیم نشده است",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "زمان اتصال به پایان رسید ({{url}})",
+  "settings.ai.testError.unreachable": "در دسترس نیست ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "شبکه‌سازی سرور محلی در دسترس نیست: بارگذاری افزونه HTTP دسکتاپ ناموفق بود.",
+  "settings.ai.testError.desktopRequired": "Ollama و سرورهای محلی سازگار با OpenAI فقط در برنامه دسکتاپ در دسترس هستند. مرورگرها اتصالات مستقیم از صفحات وب به localhost را مسدود می‌کنند (CORS و Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude به یک پراکسی بک‌اند نیاز دارد (محدودیت CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU در این مرورگر در دسترس نیست — WebLLM به WebGPU نیاز دارد (Chrome/Edge را امتحان کنید یا پرچم‌ها را فعال کنید).",
+  "settings.ai.testError.unknownProvider": "ارائه‌دهنده ناشناخته",
+  "settings.ai.testError.unexpected": "خطای غیرمنتظره‌ای رخ داد. لطفاً دوباره امتحان کنید.",
   "settings.ai.title": "پیکربندی هوش مصنوعی",
   "settings.ai.transformers.hint": "هر مدل HuggingFace سازگار با Xenova را به صورت محلی از طریق WebAssembly یا WebGPU اجرا می کند.",
   "settings.ai.transformers.modelId": "شناسه مدل (HuggingFace)",
diff --git a/public/locales/fi/bundle.json b/public/locales/fi/bundle.json
index e99bb953..bd8f2d51 100644
--- a/public/locales/fi/bundle.json
+++ b/public/locales/fi/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Luova",
   "settings.ai.temperature.precise": "0 – Tarkka",
   "settings.ai.testConnection": "Testaa yhteys",
+  "settings.ai.testError.noApiKey": "{{provider}}-API-avainta ei ole asetettu",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Yhteys aikakatkaistiin ({{url}})",
+  "settings.ai.testError.unreachable": "Ei tavoiteta ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Paikallisen palvelimen verkko ei ole käytettävissä: työpöydän HTTP-lisäosan lataus epäonnistui.",
+  "settings.ai.testError.desktopRequired": "Ollama ja paikalliset OpenAI-yhteensopivat palvelimet ovat käytettävissä vain työpöytäsovelluksessa. Selaimet estävät suorat yhteydet verkkosivuilta osoitteeseen localhost (CORS ja Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude vaatii taustajärjestelmän välityspalvelimen (CORS-rajoitus)",
+  "settings.ai.testError.noWebgpu": "WebGPU ei ole käytettävissä tässä selaimessa — WebLLM tarvitsee WebGPU:n (kokeile Chromea/Edgeä tai ota liput käyttöön).",
+  "settings.ai.testError.unknownProvider": "Tuntematon palveluntarjoaja",
+  "settings.ai.testError.unexpected": "Tapahtui odottamaton virhe. Yritä uudelleen.",
   "settings.ai.title": "AI-kokoonpano",
   "settings.ai.transformers.hint": "Suorittaa mitä tahansa Xenova-yhteensopivaa HuggingFace-mallia paikallisesti WebAssemblyn tai WebGPU:n kautta.",
   "settings.ai.transformers.modelId": "Mallin tunnus (HuggingFace)",
diff --git a/public/locales/fr/bundle.json b/public/locales/fr/bundle.json
index ab500ac6..68d46246 100644
--- a/public/locales/fr/bundle.json
+++ b/public/locales/fr/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Créatif",
   "settings.ai.temperature.precise": "0 – Précis",
   "settings.ai.testConnection": "Tester la connexion",
+  "settings.ai.testError.noApiKey": "Aucune clé API {{provider}} définie",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Délai de connexion dépassé ({{url}})",
+  "settings.ai.testError.unreachable": "Injoignable ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Le réseau des serveurs locaux est indisponible : le plugin HTTP du bureau n'a pas pu se charger.",
+  "settings.ai.testError.desktopRequired": "Ollama et les serveurs locaux compatibles OpenAI ne sont disponibles que dans l'application de bureau. Les navigateurs bloquent les connexions directes des pages web vers localhost (CORS et Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude nécessite un proxy backend (restriction CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU indisponible dans ce navigateur — WebLLM nécessite WebGPU (essayez Chrome/Edge ou activez les indicateurs).",
+  "settings.ai.testError.unknownProvider": "Fournisseur inconnu",
+  "settings.ai.testError.unexpected": "Une erreur inattendue s'est produite. Veuillez réessayer.",
   "settings.ai.title": "Configuration de l'IA",
   "settings.ai.transformers.hint": "Exécute tout modèle HuggingFace compatible Xenova localement via WebAssembly ou WebGPU.",
   "settings.ai.transformers.modelId": "ID du modèle (HuggingFace)",
diff --git a/public/locales/he/bundle.json b/public/locales/he/bundle.json
index 859b0843..d829774d 100644
--- a/public/locales/he/bundle.json
+++ b/public/locales/he/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – יצירתי",
   "settings.ai.temperature.precise": "0 – מדויק",
   "settings.ai.testConnection": "בדיקת חיבור",
+  "settings.ai.testError.noApiKey": "לא הוגדר מפתח API עבור {{provider}}",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "פסק הזמן של החיבור פג ({{url}})",
+  "settings.ai.testError.unreachable": "לא ניתן להגיע ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "רשת השרת המקומי אינה זמינה: תוסף ה-HTTP של שולחן העבודה נכשל בטעינה.",
+  "settings.ai.testError.desktopRequired": "Ollama ושרתים מקומיים תואמי OpenAI זמינים רק באפליקציית שולחן העבודה. דפדפנים חוסמים חיבורים ישירים מדפי אינטרנט אל localhost (CORS ו-Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude דורש שרת proxy (הגבלת CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU אינו זמין בדפדפן זה — WebLLM דורש WebGPU (נסה Chrome/Edge או הפעל דגלים).",
+  "settings.ai.testError.unknownProvider": "ספק לא ידוע",
+  "settings.ai.testError.unexpected": "אירעה שגיאה בלתי צפויה. נסה שוב.",
   "settings.ai.title": "תצורת AI",
   "settings.ai.transformers.hint": "מריץ כל מודל HuggingFace תואם‑Xenova מקומית דרך WebAssembly או WebGPU.",
   "settings.ai.transformers.modelId": "מזהה מודל ‏(HuggingFace)",
diff --git a/public/locales/hu/bundle.json b/public/locales/hu/bundle.json
index 8ce0d971..8a78c116 100644
--- a/public/locales/hu/bundle.json
+++ b/public/locales/hu/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Kreatív",
   "settings.ai.temperature.precise": "0 – Pontos",
   "settings.ai.testConnection": "Kapcsolat tesztelése",
+  "settings.ai.testError.noApiKey": "Nincs beállítva {{provider}} API-kulcs",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "A kapcsolat túllépte az időkorlátot ({{url}})",
+  "settings.ai.testError.unreachable": "Nem érhető el ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "A helyi kiszolgálóhálózat nem érhető el: az asztali HTTP bővítmény betöltése sikertelen.",
+  "settings.ai.testError.desktopRequired": "Az Ollama és a helyi OpenAI-kompatibilis kiszolgálók csak az asztali alkalmazásban érhetők el. A böngészők blokkolják a weboldalak közvetlen kapcsolatait a localhosttal (CORS és Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "A Claude háttérproxyt igényel (CORS-korlátozás)",
+  "settings.ai.testError.noWebgpu": "A WebGPU nem érhető el ebben a böngészőben — a WebLLM-nek WebGPU-ra van szüksége (próbáld a Chrome-ot/Edge-et, vagy engedélyezd a jelzőket).",
+  "settings.ai.testError.unknownProvider": "Ismeretlen szolgáltató",
+  "settings.ai.testError.unexpected": "Váratlan hiba történt. Kérjük, próbálja újra.",
   "settings.ai.title": "AI konfiguráció",
   "settings.ai.transformers.hint": "Bármilyen Xenova-kompatibilis HuggingFace modellt futtat helyileg a WebAssembly vagy WebGPU segítségével.",
   "settings.ai.transformers.modelId": "Modellazonosító (HuggingFace)",
diff --git a/public/locales/is/bundle.json b/public/locales/is/bundle.json
index 089006c9..eb30df88 100644
--- a/public/locales/is/bundle.json
+++ b/public/locales/is/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 - Skapandi",
   "settings.ai.temperature.precise": "0 - Nákvæmt",
   "settings.ai.testConnection": "Prófaðu tengingu",
+  "settings.ai.testError.noApiKey": "Enginn {{provider}} API lykill stilltur",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Tengingin féll á tíma ({{url}})",
+  "settings.ai.testError.unreachable": "Næst ekki ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Staðbundið netþjónasamband ekki tiltækt: ekki tókst að hlaða HTTP-viðbót skjáborðs.",
+  "settings.ai.testError.desktopRequired": "Ollama og staðbundnir OpenAI-samhæfir netþjónar eru aðeins tiltækir í skjáborðsforritinu. Vafrar loka fyrir bein tengsl frá vefsíðum við localhost (CORS og Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude þarfnast bakenda-milliþjóns (CORS-takmörkun)",
+  "settings.ai.testError.noWebgpu": "WebGPU er ekki tiltækt í þessum vafra — WebLLM þarfnast WebGPU (prófaðu Chrome/Edge eða virkjaðu flögg).",
+  "settings.ai.testError.unknownProvider": "Óþekktur veitandi",
+  "settings.ai.testError.unexpected": "Óvænt villa kom upp. Reyndu aftur.",
   "settings.ai.title": "AI stillingar",
   "settings.ai.transformers.hint": "Keyrir hvaða Xenova-samhæfða HuggingFace líkan sem er á staðnum í gegnum WebAssembly eða WebGPU.",
   "settings.ai.transformers.modelId": "Auðkenni fyrirmyndar (HuggingFace)",
diff --git a/public/locales/it/bundle.json b/public/locales/it/bundle.json
index eb7e917d..06e39039 100644
--- a/public/locales/it/bundle.json
+++ b/public/locales/it/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Creativo",
   "settings.ai.temperature.precise": "0 – Preciso",
   "settings.ai.testConnection": "Testa la connessione",
+  "settings.ai.testError.noApiKey": "Nessuna chiave API {{provider}} impostata",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Connessione scaduta ({{url}})",
+  "settings.ai.testError.unreachable": "Non raggiungibile ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Rete dei server locali non disponibile: il plugin HTTP desktop non è stato caricato.",
+  "settings.ai.testError.desktopRequired": "Ollama e i server locali compatibili con OpenAI sono disponibili solo nell'app desktop. I browser bloccano le connessioni dirette dalle pagine web a localhost (CORS e Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude richiede un proxy backend (restrizione CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU non disponibile in questo browser — WebLLM richiede WebGPU (prova Chrome/Edge o abilita i flag).",
+  "settings.ai.testError.unknownProvider": "Provider sconosciuto",
+  "settings.ai.testError.unexpected": "Si è verificato un errore imprevisto. Riprova.",
   "settings.ai.title": "Configurazione IA",
   "settings.ai.transformers.hint": "Esegue qualsiasi modello HuggingFace compatibile con Xenova localmente via WebAssembly o WebGPU.",
   "settings.ai.transformers.modelId": "ID modello (HuggingFace)",
diff --git a/public/locales/ja/bundle.json b/public/locales/ja/bundle.json
index 70e86533..3dfc59d1 100644
--- a/public/locales/ja/bundle.json
+++ b/public/locales/ja/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – クリエイティブ",
   "settings.ai.temperature.precise": "0 – 正確",
   "settings.ai.testConnection": "テスト接続",
+  "settings.ai.testError.noApiKey": "{{provider}} の API キーが設定されていません",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "接続がタイムアウトしました ({{url}})",
+  "settings.ai.testError.unreachable": "到達できません ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "ローカルサーバーへの接続が利用できません: デスクトップ HTTP プラグインの読み込みに失敗しました。",
+  "settings.ai.testError.desktopRequired": "Ollama およびローカルの OpenAI 互換サーバーはデスクトップアプリでのみ利用できます。ブラウザは Web ページから localhost への直接接続をブロックします（CORS および Private Network Access）。",
+  "settings.ai.testError.backendProxyRequired": "Claude にはバックエンドプロキシが必要です（CORS 制限）",
+  "settings.ai.testError.noWebgpu": "このブラウザでは WebGPU が利用できません — WebLLM には WebGPU が必要です（Chrome/Edge を試すかフラグを有効にしてください）。",
+  "settings.ai.testError.unknownProvider": "不明なプロバイダー",
+  "settings.ai.testError.unexpected": "予期しないエラーが発生しました。もう一度お試しください。",
   "settings.ai.title": "AIの構成",
   "settings.ai.transformers.hint": "WebAssembly または WebGPU を介して Xenova 互換の HuggingFace モデルをローカルで実行します。",
   "settings.ai.transformers.modelId": "モデルID (HuggingFace)",
diff --git a/public/locales/ko/bundle.json b/public/locales/ko/bundle.json
index aceaf9cd..28b94a93 100644
--- a/public/locales/ko/bundle.json
+++ b/public/locales/ko/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – 창의적",
   "settings.ai.temperature.precise": "0 – 정확함",
   "settings.ai.testConnection": "연결 테스트",
+  "settings.ai.testError.noApiKey": "{{provider}} API 키가 설정되지 않았습니다",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "연결 시간이 초과되었습니다 ({{url}})",
+  "settings.ai.testError.unreachable": "연결할 수 없습니다 ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "로컬 서버 네트워킹을 사용할 수 없습니다: 데스크톱 HTTP 플러그인을 불러오지 못했습니다.",
+  "settings.ai.testError.desktopRequired": "Ollama 및 로컬 OpenAI 호환 서버는 데스크톱 앱에서만 사용할 수 있습니다. 브라우저는 웹 페이지에서 localhost로의 직접 연결을 차단합니다(CORS 및 Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude에는 백엔드 프록시가 필요합니다(CORS 제한)",
+  "settings.ai.testError.noWebgpu": "이 브라우저에서는 WebGPU를 사용할 수 없습니다 — WebLLM에는 WebGPU가 필요합니다(Chrome/Edge를 사용하거나 플래그를 활성화하세요).",
+  "settings.ai.testError.unknownProvider": "알 수 없는 제공자",
+  "settings.ai.testError.unexpected": "예기치 않은 오류가 발생했습니다. 다시 시도해 주세요.",
   "settings.ai.title": "AI 구성",
   "settings.ai.transformers.hint": "WebAssembly 또는 WebGPU를 통해 Xenova 호환 HuggingFace 모델을 로컬로 실행합니다.",
   "settings.ai.transformers.modelId": "모델 ID(HuggingFace)",
diff --git a/public/locales/pt/bundle.json b/public/locales/pt/bundle.json
index 85e9a83b..f647dc67 100644
--- a/public/locales/pt/bundle.json
+++ b/public/locales/pt/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Criativo",
   "settings.ai.temperature.precise": "0 – Preciso",
   "settings.ai.testConnection": "Conexão de teste",
+  "settings.ai.testError.noApiKey": "Nenhuma chave de API {{provider}} definida",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Tempo limite de conexão excedido ({{url}})",
+  "settings.ai.testError.unreachable": "Inacessível ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "A rede do servidor local não está disponível: o plugin HTTP de ambiente de trabalho falhou ao carregar.",
+  "settings.ai.testError.desktopRequired": "O Ollama e os servidores locais compatíveis com OpenAI só estão disponíveis na aplicação de ambiente de trabalho. Os navegadores bloqueiam ligações diretas de páginas web ao localhost (CORS e Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "O Claude requer um proxy de backend (restrição CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU indisponível neste navegador — o WebLLM necessita de WebGPU (experimente o Chrome/Edge ou ative as flags).",
+  "settings.ai.testError.unknownProvider": "Fornecedor desconhecido",
+  "settings.ai.testError.unexpected": "Ocorreu um erro inesperado. Tente novamente.",
   "settings.ai.title": "IA Configuration",
   "settings.ai.transformers.hint": "Executa qualquer modelo HuggingFace compatível com Xenova localmente via WebAssembly ou WebGPU.",
   "settings.ai.transformers.modelId": "ID do modelo (HuggingFace)",
diff --git a/public/locales/ru/bundle.json b/public/locales/ru/bundle.json
index c8e03220..245aa4c0 100644
--- a/public/locales/ru/bundle.json
+++ b/public/locales/ru/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Креатив",
   "settings.ai.temperature.precise": "0 – Точный",
   "settings.ai.testConnection": "Тестовое соединение",
+  "settings.ai.testError.noApiKey": "Не задан ключ API {{provider}}",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Тайм-аут соединения ({{url}})",
+  "settings.ai.testError.unreachable": "Недоступно ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Сеть локального сервера недоступна: не удалось загрузить настольный HTTP-плагин.",
+  "settings.ai.testError.desktopRequired": "Ollama и локальные серверы, совместимые с OpenAI, доступны только в настольном приложении. Браузеры блокируют прямые подключения с веб-страниц к localhost (CORS и Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude требует backend-прокси (ограничение CORS)",
+  "settings.ai.testError.noWebgpu": "WebGPU недоступен в этом браузере — WebLLM требует WebGPU (попробуйте Chrome/Edge или включите флаги).",
+  "settings.ai.testError.unknownProvider": "Неизвестный провайдер",
+  "settings.ai.testError.unexpected": "Произошла непредвиденная ошибка. Повторите попытку.",
   "settings.ai.title": "Конфигурация ИИ",
   "settings.ai.transformers.hint": "Запускает любую Xenova-совместимую модель HuggingFace локально через WebAssembly или WebGPU.",
   "settings.ai.transformers.modelId": "Идентификатор модели (HuggingFace)",
diff --git a/public/locales/sv/bundle.json b/public/locales/sv/bundle.json
index 93aba005..1cb7f639 100644
--- a/public/locales/sv/bundle.json
+++ b/public/locales/sv/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – Kreativt",
   "settings.ai.temperature.precise": "0 – Exakt",
   "settings.ai.testConnection": "Testa anslutningen",
+  "settings.ai.testError.noApiKey": "Ingen {{provider}}-API-nyckel angiven",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "Anslutningen fick timeout ({{url}})",
+  "settings.ai.testError.unreachable": "Ej nåbar ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "Lokal servernätverksanslutning ej tillgänglig: skrivbordets HTTP-plugin misslyckades med att läsas in.",
+  "settings.ai.testError.desktopRequired": "Ollama och lokala OpenAI-kompatibla servrar är endast tillgängliga i skrivbordsappen. Webbläsare blockerar direkta anslutningar från webbsidor till localhost (CORS och Private Network Access).",
+  "settings.ai.testError.backendProxyRequired": "Claude kräver en backend-proxy (CORS-begränsning)",
+  "settings.ai.testError.noWebgpu": "WebGPU är inte tillgängligt i den här webbläsaren — WebLLM behöver WebGPU (prova Chrome/Edge eller aktivera flaggor).",
+  "settings.ai.testError.unknownProvider": "Okänd leverantör",
+  "settings.ai.testError.unexpected": "Ett oväntat fel inträffade. Försök igen.",
   "settings.ai.title": "AI-konfiguration",
   "settings.ai.transformers.hint": "Kör valfri Xenova-kompatibel HuggingFace-modell lokalt via WebAssembly eller WebGPU.",
   "settings.ai.transformers.modelId": "Modell-ID (HuggingFace)",
diff --git a/public/locales/zh/bundle.json b/public/locales/zh/bundle.json
index 4190f2a1..ab6e8835 100644
--- a/public/locales/zh/bundle.json
+++ b/public/locales/zh/bundle.json
@@ -1855,6 +1855,16 @@
   "settings.ai.temperature.creative": "2 – 创意",
   "settings.ai.temperature.precise": "0 – 精确",
   "settings.ai.testConnection": "测试连接",
+  "settings.ai.testError.noApiKey": "未设置 {{provider}} API 密钥",
+  "settings.ai.testError.httpError": "HTTP {{status}}",
+  "settings.ai.testError.timeout": "连接超时 ({{url}})",
+  "settings.ai.testError.unreachable": "无法访问 ({{url}})",
+  "settings.ai.testError.pluginUnavailable": "本地服务器网络不可用：桌面 HTTP 插件加载失败。",
+  "settings.ai.testError.desktopRequired": "Ollama 和本地 OpenAI 兼容服务器仅在桌面应用中可用。浏览器会阻止网页直接连接到 localhost（CORS 和 Private Network Access）。",
+  "settings.ai.testError.backendProxyRequired": "Claude 需要后端代理（CORS 限制）",
+  "settings.ai.testError.noWebgpu": "此浏览器不支持 WebGPU — WebLLM 需要 WebGPU（请尝试 Chrome/Edge 或启用相关标志）。",
+  "settings.ai.testError.unknownProvider": "未知的提供商",
+  "settings.ai.testError.unexpected": "发生意外错误。请重试。",
   "settings.ai.title": "人工智能配置",
   "settings.ai.transformers.hint": "通过 WebAssembly 或 WebGPU 在本地运行任何 Xenova 兼容的 HuggingFace 模型。",
   "settings.ai.transformers.modelId": "型号 ID (HuggingFace)",
diff --git a/services/CLAUDE.md b/services/CLAUDE.md
new file mode 100644
index 00000000..d3311c92
--- /dev/null
+++ b/services/CLAUDE.md
@@ -0,0 +1,31 @@
+# Scene-level services
+
+**sceneRevisionService:** `services/sceneRevisionService.ts` — IDB `scene-revisions`; `saveRevision(sectionId, snapshot, label?)`, `listRevisions(sectionId)`, `deleteRevision(id)`.
+
+**sceneCommentsSlice:** `features/sceneComments/sceneCommentsSlice.ts` — EntityAdapter; `selectCommentsBySection(sectionId)`, `selectUnresolvedCount`, `selectUnresolvedCountBySection(sectionId)`. Root key: `sceneComments`.
+
+**progressTrackerSlice:** `features/progressTracker/progressTrackerSlice.ts` — `startSession`, `endSession`, `setDailyGoal`, `setWeeklyGoal`, `syncStreak`. Exported: `computeStreak(history)`.
+
+**deepLinkService:** `services/deepLinkService.ts` — `parseHash(hash)`, `pushHash(view, sectionId?)`, `readCurrentView()`. Views: `'board' | 'preview' | 'progress' | 'project'`.
+
+# LanguageTool Integration
+
+Privacy-first, **self-hosted-only** grammar/spell checking wired into the manuscript editor (`docs/LANGUAGETOOL.md`, `docs/adr/0010-languagetool-self-hosted.md`). Off by default via `settings.integrations.languageToolEnabled`. `services/languageToolClient.ts#assertLanguageToolAllowed` blocks any non-`localhost`/`127.0.0.1` base URL whenever `settings.privacy.localStorageOnly` is on, so manuscript text can never reach a cloud LanguageTool instance. `services/languageToolService.ts` never logs manuscript text — only match counts and a `status: 'ok' | 'offline' | 'error'` via `services/logger.ts#log.warn` — and returns an empty match list on network/HTTP failure rather than throwing, so a failed check never blocks typing. Corrections apply offset-safe through `hooks/useLanguageToolCheck.ts` → `applyMatchReplacement` (redux-undo-wrapped); user dictionary persists in `settings.advancedEditor.customDictionary`.
+
+# Cross-project & backup
+
+**crossProjectIndexService / crossProjectSearchService:** IDB `projects-index-store` (DB_VERSION 8); `searchAcrossProjects()` via fuzzyScore. Indexing triggered on save. UI: `CrossProjectSearchPanel`; Zustand key: `isCrossProjectSearchOpen`.
+
+**libraryBackupService:** one-click encrypted ZIP export (AES-GCM, `META.json` + `vault.bin`). Settings → Data.
+
+# Local inference
+
+`services/localAiFacade.ts` wraps WebLLM with the same provider interface as `aiProviderService.ts`. Model download progress via `onProgress`; mount-guard via `useRef`.
+
+# Plugin System
+
+`services/pluginRegistry.ts` — `PluginRegistry` + singleton. Plugins declare `PluginDescriptor` (Zod-validated: `id`, `version`, `type`, `entrypoint`, `permissions`). `PluginSandboxedApi` gates every method behind declared permissions.
+
+**Execution API:** `execute(id, fn, rawApi)` (sync) · `executeAsync` (async) · `loadPlugin(descriptor, rawApi)` (dynamic import + `run(api)`).
+
+Reference plugins: `wordCountOverlay.plugin.ts`, `sceneAppender.plugin.ts`. Gate: `enablePluginSystem`.
diff --git a/services/aiProviderService.ts b/services/aiProviderService.ts
index 6c3183fd..afb8190f 100644
--- a/services/aiProviderService.ts
+++ b/services/aiProviderService.ts
@@ -38,6 +38,7 @@ import {
 } from './geminiService';
 import { generateLocalText } from './localAiFacade';
 import { LocalServerError, localServerFetch } from './localServerHttp';
+import { createLogger } from './logger';
 import {
   listOllamaModels as listOllamaModelsFromService,
   streamOllama,
@@ -46,6 +47,8 @@ import {
 import { storageService } from './storageService';
 import { isTauriRuntime } from './tauriRuntime';
 
+const log = createLogger('aiProviderService');
+
 const providerTextSchema = z.object({
   text: z.string().min(1),
 });
@@ -726,21 +729,60 @@ export async function scanLocalOpenAiCompatibleEndpoints(): Promise<LocalEndpoin
   );
 }
 
+/**
+ * Stable, i18n-mappable classification of a connection-test failure across ALL providers. `error`
+ * stays a raw/technical string for logs; UI code should prefer `kind` (+ `params` for
+ * interpolation) to render a localized message via `settings.ai.testError.*`
+ * (`locales/<lang>/settings.json`) — falling back to `error` only when `kind` is absent.
+ */
+export type TestConnectionErrorKind =
+  | 'noApiKey'
+  | 'httpError'
+  | 'timeout'
+  | 'unreachable'
+  | 'pluginUnavailable'
+  | 'desktopRequired'
+  | 'backendProxyRequired'
+  | 'noWebgpu'
+  | 'unknownProvider'
+  | 'unexpected';
+
+export interface TestConnectionResult {
+  ok: boolean;
+  error?: string;
+  kind?: TestConnectionErrorKind;
+  params?: Record<string, string | number>;
+}
+
 export async function testAIConnection(
   provider: AIProvider,
   opts: Partial<AIRequestOptions>,
-): Promise<{ ok: boolean; error?: string }> {
+): Promise<TestConnectionResult> {
   try {
     switch (provider) {
       case 'openai': {
         const apiKey = await storageService.getApiKey('openai');
-        if (!apiKey) return { ok: false, error: 'Kein OpenAI API Key gesetzt' };
+        if (!apiKey) {
+          return {
+            ok: false,
+            error: 'Kein OpenAI API Key gesetzt',
+            kind: 'noApiKey',
+            params: { provider: 'OpenAI' },
+          };
+        }
         const root = resolveOpenAiCompatibleRoot(opts.openAiCompatibleBaseUrl);
         const res = await fetch(`${root}/models`, {
           headers: { Authorization: `Bearer ${apiKey}` },
           signal: AbortSignal.timeout(8000),
         });
-        if (!res.ok) return { ok: false, error: `HTTP ${res.status}` };
+        if (!res.ok) {
+          return {
+            ok: false,
+            error: `HTTP ${res.status}`,
+            kind: 'httpError',
+            params: { status: res.status },
+          };
+        }
         return { ok: true };
       }
       case 'ollama': {
@@ -748,10 +790,14 @@ export async function testAIConnection(
         // desktop Ollama (localhost) path was unreachable there.
         const isDesktop = isTauriRuntime();
         if (!isDesktop) {
+          // QNBS-v3 (ADR-0012): browsers block localhost via CORS/Private Network Access, NOT CSP
+          // — this repo's CSP already allowlists localhost (docs/adr/0004). Matches the corrected
+          // settings.ai.ollamaDesktopOnlyBody wording from #269; this hard-gate string had drifted.
           return {
             ok: false,
             error:
-              'Ollama is only available in the desktop app. The browser Content Security Policy blocks direct connections to localhost.',
+              'Ollama and local OpenAI-compatible servers are only available in the desktop app. Browsers block direct connections from web pages to localhost (CORS and Private Network Access).',
+            kind: 'desktopRequired',
           };
         }
         return testOllamaConnection(opts.ollamaBaseUrl);
@@ -760,25 +806,54 @@ export async function testAIConnection(
         return {
           ok: false,
           error: 'Claude requires a backend proxy (CORS restriction)',
+          kind: 'backendProxyRequired',
         };
       case 'grok': {
         const apiKey = await storageService.getApiKey('grok');
-        if (!apiKey) return { ok: false, error: 'Kein Grok API Key gesetzt' };
+        if (!apiKey) {
+          return {
+            ok: false,
+            error: 'Kein Grok API Key gesetzt',
+            kind: 'noApiKey',
+            params: { provider: 'Grok' },
+          };
+        }
         const res = await fetch('https://api.x.ai/v1/models', {
           headers: { Authorization: `Bearer ${apiKey}` },
           signal: AbortSignal.timeout(8000),
         });
-        if (!res.ok) return { ok: false, error: `HTTP ${res.status}` };
+        if (!res.ok) {
+          return {
+            ok: false,
+            error: `HTTP ${res.status}`,
+            kind: 'httpError',
+            params: { status: res.status },
+          };
+        }
         return { ok: true };
       }
       case 'gemini': {
         const geminiKey = await storageService.getGeminiApiKey();
-        if (!geminiKey) return { ok: false, error: 'No Gemini API key set' };
+        if (!geminiKey) {
+          return {
+            ok: false,
+            error: 'No Gemini API key set',
+            kind: 'noApiKey',
+            params: { provider: 'Gemini' },
+          };
+        }
         const res = await fetch(
           `https://generativelanguage.googleapis.com/v1beta/models?key=${geminiKey}`,
           { signal: AbortSignal.timeout(8000) },
         );
-        if (!res.ok) return { ok: false, error: `Gemini API: HTTP ${res.status}` };
+        if (!res.ok) {
+          return {
+            ok: false,
+            error: `Gemini API: HTTP ${res.status}`,
+            kind: 'httpError',
+            params: { status: res.status },
+          };
+        }
         return { ok: true };
       }
       case 'webllm':
@@ -788,6 +863,7 @@ export async function testAIConnection(
               ok: false,
               error:
                 'WebGPU unavailable in this browser — WebLLM needs WebGPU (try Chrome/Edge or enable flags).',
+              kind: 'noWebgpu',
             };
       case 'onnx':
         // QNBS-v3: ONNX Runtime Web uses WASM — always available, no GPU required.
@@ -796,12 +872,17 @@ export async function testAIConnection(
         // QNBS-v3: Transformers.js uses WASM/WebGPU — connection test is always ok; model loads on first use.
         return { ok: true };
       default:
-        return { ok: false, error: 'Unknown provider' };
+        return { ok: false, error: 'Unknown provider', kind: 'unknownProvider' };
     }
   } catch (e) {
+    const message = e instanceof Error ? e.message : String(e);
+    // QNBS-v3 (CodeAnt CWE-209): log the raw exception for diagnostics; never interpolate it into
+    // the user-facing i18n string, which could otherwise leak internal error detail to the UI.
+    log.error('testAIConnection: unexpected failure', { provider, message });
     return {
       ok: false,
-      error: e instanceof Error ? e.message : String(e),
+      error: message,
+      kind: 'unexpected',
     };
   }
 }
diff --git a/services/cloudSync/CLAUDE.md b/services/cloudSync/CLAUDE.md
new file mode 100644
index 00000000..c4ecf4a2
--- /dev/null
+++ b/services/cloudSync/CLAUDE.md
@@ -0,0 +1,3 @@
+# Cloud Sync (Cloudflare R2)
+
+`services/cloudSync/` — `cloudSyncBackend.ts` (StorageBackend, API keys never sent to cloud), `cloudSyncClient.ts` (fetch + Bearer token), `cloudSyncEncryption.ts` (AES-256-GCM E2E). The `enableCloudSync` feature flag was **retired** in v1.20; use `CloudSyncBackend.create(..., explicitConsent = true)` as the activation gate. This service is not yet wired into `storageService` (v2.0 feature).
diff --git a/services/copilot/CLAUDE.md b/services/copilot/CLAUDE.md
new file mode 100644
index 00000000..885ae7ec
--- /dev/null
+++ b/services/copilot/CLAUDE.md
@@ -0,0 +1,13 @@
+# Global AI Copilot (v2)
+
+Flag: `enableGlobalCopilot`. Redux slice: `features/copilot/copilotSlice.ts` (ephemeral, NOT persisted, NOT undo-wrapped). Hook: `hooks/useGlobalCopilot.ts`. Mount: `App.tsx` (lazy, `ErrorBoundary` + `Suspense`).
+
+**Services (`services/copilot/`):**
+- `copilotContextService.ts` — pure system-prompt builder from current view + project metadata
+- `insightGenerator.ts` — debounced heuristic analysis; result LRU-cached per section
+- `heuristicEngine.ts` — 8 built-in manuscript rules (tension-drop, underdeveloped-character, open-loop, slow-pacing, high-repetition, plot-hole, missing-world-context, overlength-scene); zero network calls
+- `actionApplier.ts` — `applyTextEdit(sectionId, newText)`: offset-safe rewrite dispatched into redux-undo; only runs when block ≥ 70 % of section length
+
+**UI (`components/copilot/`):** `CopilotPanel` (dialog/sidebar toggle, persisted to `localStorage`), `CopilotMessageList` (DOMPurify + micro-markdown renderer — no new runtime dep), `InlineAnnotationLayer` (badge inside `ManuscriptEditor`), `CopilotComposer`. ProForge: each `ReviewItemCard` has an ✦ Ask Copilot chip (pre-fills composer with review-item context).
+
+Docs: `docs/COPILOT.md` (feature guide), `docs/HEURISTIC-RULES.md` (8 rules catalogue).
diff --git a/services/localServerHttp.ts b/services/localServerHttp.ts
index 499c72cf..2aa813e3 100644
--- a/services/localServerHttp.ts
+++ b/services/localServerHttp.ts
@@ -7,9 +7,12 @@
  * on the web we keep the global fetch. The plugin is imported dynamically so the web bundle never
  * loads it. Kept deliberately thin: URL normalization, timeout composition, error classification.
  */
+import { createLogger } from './logger';
 import { isTauriRuntime } from './tauriRuntime';
 
-export type LocalServerErrorKind = 'unreachable' | 'timeout';
+const log = createLogger('localServerHttp');
+
+export type LocalServerErrorKind = 'unreachable' | 'timeout' | 'plugin_unavailable';
 
 /** Classified local-server failure. User aborts are NOT wrapped — they rethrow unchanged. */
 export class LocalServerError extends Error {
@@ -56,11 +59,32 @@ export interface LocalServerFetchInit {
   timeoutMs?: number;
 }
 
-/** Picks the native Tauri HTTP plugin on desktop, global fetch on the web. */
+/**
+ * Picks the native Tauri HTTP plugin on desktop, global fetch on the web.
+ *
+ * QNBS-v3: the dynamic import can fail at runtime even when `isTauriRuntime()` is true — a build
+ * config regression that externalizes `@tauri-apps/*` leaves this as an unresolvable bare
+ * specifier in the packaged app. That failure must NOT be swallowed into the generic
+ * 'unreachable' classification (indistinguishable from a genuinely-down server); it's a plugin
+ * load failure, not a network failure, and is far more actionable to surface distinctly. No
+ * `globalThis.fetch` fallback here (unlike services/ai/fetchAdapter.ts's cloud-provider path) —
+ * it wouldn't help localhost targets, which fail on CORS in the WebView regardless.
+ */
 async function resolveFetch(): Promise<typeof fetch> {
   if (isTauriRuntime()) {
-    const { fetch: tauriFetch } = await import('@tauri-apps/plugin-http');
-    return tauriFetch as unknown as typeof fetch;
+    try {
+      const { fetch: tauriFetch } = await import('@tauri-apps/plugin-http');
+      return tauriFetch as unknown as typeof fetch;
+    } catch (err) {
+      log.error('Tauri HTTP plugin failed to load; local server requests cannot proceed', {
+        error: err instanceof Error ? err.message : String(err),
+      });
+      throw new LocalServerError(
+        'plugin_unavailable',
+        'Local server networking is unavailable: the desktop HTTP plugin failed to load.',
+        { cause: err },
+      );
+    }
   }
   return globalThis.fetch.bind(globalThis);
 }
diff --git a/services/ollamaService.ts b/services/ollamaService.ts
index 7c1be071..20b8034e 100644
--- a/services/ollamaService.ts
+++ b/services/ollamaService.ts
@@ -4,10 +4,13 @@ import {
   localServerFetch,
   normalizeLocalBaseUrl,
 } from './localServerHttp';
+import { createLogger } from './logger';
 
 // QNBS-v3 (#266): canonical normalization lives in localServerHttp (shared with the scanner).
 const normalizeBaseUrl = normalizeLocalBaseUrl;
 
+const log = createLogger('ollamaService');
+
 const stripControlChars = (value: string): string => {
   let output = '';
   for (const char of String(value)) {
@@ -149,23 +152,60 @@ export async function listOllamaModels(baseUrl?: string): Promise<string[]> {
   }
 }
 
-export async function testOllamaConnection(
-  baseUrl?: string,
-): Promise<{ ok: boolean; error?: string }> {
-  const url = `${normalizeBaseUrl(baseUrl)}/api/tags`;
+/**
+ * Stable, i18n-mappable classification of a connection-test failure. `error` on the result stays
+ * a raw/technical string for logs; UI code should prefer `kind` (+ `params` for interpolation) to
+ * render a localized message, per `settings.ai.testError.*` in `locales/<lang>/settings.json`.
+ */
+export type TestConnectionErrorKind = 'httpError' | 'timeout' | 'unreachable' | 'pluginUnavailable';
+
+export interface TestConnectionResult {
+  ok: boolean;
+  error?: string;
+  kind?: TestConnectionErrorKind;
+  params?: Record<string, string | number>;
+}
+
+export async function testOllamaConnection(baseUrl?: string): Promise<TestConnectionResult> {
+  const resolvedUrl = normalizeBaseUrl(baseUrl);
+  const url = `${resolvedUrl}/api/tags`;
   try {
     const res = await localServerFetch(url, { timeoutMs: 5000 });
-    if (!res.ok) return { ok: false, error: `Ollama HTTP ${res.status}` };
+    if (!res.ok) {
+      return {
+        ok: false,
+        error: `Ollama HTTP ${res.status}`,
+        kind: 'httpError',
+        params: { status: res.status },
+      };
+    }
     return { ok: true };
   } catch (error: unknown) {
     // QNBS-v3 (#266): classified failures — distinguish a hanging server from a missing one.
     if (error instanceof LocalServerError && error.kind === 'timeout') {
-      return { ok: false, error: `Ollama timed out (${normalizeBaseUrl(baseUrl)})` };
+      return {
+        ok: false,
+        error: `Ollama timed out (${resolvedUrl})`,
+        kind: 'timeout',
+        params: { url: resolvedUrl },
+      };
+    }
+    // QNBS-v3: plugin_unavailable means the desktop HTTP transport itself failed to load — not
+    // that Ollama is down. Surfacing it distinctly (rather than folding into "not reachable")
+    // makes a future build-config regression immediately diagnosable from the UI alone.
+    if (error instanceof LocalServerError && error.kind === 'plugin_unavailable') {
+      return { ok: false, error: error.message, kind: 'pluginUnavailable' };
     }
     const message = error instanceof Error ? error.message : String(error);
+    // QNBS-v3 (CodeAnt CWE-209): the raw transport message (could be a TypeError message with
+    // internal detail) is logged for diagnostics but never interpolated into the user-facing
+    // i18n string — only the endpoint URL, which the user already configured, is safe to show.
+    log.warn('Ollama connection unreachable', { url: resolvedUrl, message });
     return {
       ok: false,
-      error: `Ollama not reachable (${normalizeBaseUrl(baseUrl)}): ${message}`,
+      error: `Ollama not reachable (${resolvedUrl}): ${message}`,
+      kind: 'unreachable',
+      params: { url: resolvedUrl },
     };
   }
 }
diff --git a/services/proForge/CLAUDE.md b/services/proForge/CLAUDE.md
new file mode 100644
index 00000000..4e3d0c45
--- /dev/null
+++ b/services/proForge/CLAUDE.md
@@ -0,0 +1,7 @@
+# ProForge Pipeline
+
+8-stage agentic editing pipeline. Flag: `enableProForge`. Full docs: `docs/PROFORGE-PIPELINE.md`. Stages: `intake` → `structural` → `lineProse` → `copyEdit` → `proof` → `production` → `publishing` → `analytics`. Pauses at `awaitingReview` — manuscript never auto-modified without user approval. On review submit, accepted `ReviewItem`s for **editing stages** are applied back into the manuscript via `applyReviewEdits.ts` (offset-safe, back-to-front, stale-match skip) before the post-stage snapshot; redux-undo makes them reversible.
+
+**Redux slice** (`features/proForge/proForgeSlice.ts`, root key `proForge`, NOT undo-wrapped, **ephemeral**). Completed/aborted runs persist per project to IDB `proforge-run-history` (`proForgeHistoryStore.ts`, cap 20) and rehydrate via `useProForgeOrchestrator`. **Types** in `features/proForge/types.ts`. **Orchestrator:** `services/proForge/proForgeOrchestrator.ts` — call via `hooks/useProForgeOrchestrator.ts` (reads live state via `appStoreRef` from `app/storeRef.ts`), never instantiate in components. Agents (all 8 extend `BaseAgent`) lazy-loaded per stage; call `this.buildAiOpts({ maxTokens: N })` — `AIRequestOptions` requires `model` + `provider`. `BaseAgent` provides `requireProject()`, `getMemoryBank()`, `selfReflect()`, `elapsed()`, `setRetryFeedback()` (supervisor reasons injected into retry prompts).
+
+**SupervisorAgent**: heuristic gate, no AI calls, gates **all 8 stages**. `evaluate(stage, result)` → `SupervisionDecision { pass, retryRecommended, qualityScore, reasons }`. Detects `isFallback: true` (intake/structural/proof). Hard gate: intake `qualityScore < 30` → fail. All `createFallback*` use 0 scores + `isFallback: true` — never fake mid-range values. **Memory Bank:** IDB `proforge-memory-bank`; `search(query, limit, mode)` honours `ragMode` (`lexical` | `semantic` | `hybrid` via MiniLM embeddings, keyword fallback). **View:** `ProForgeViewContext` + `useProForgeViewContext()`.
diff --git a/services/voice/CLAUDE.md b/services/voice/CLAUDE.md
new file mode 100644
index 00000000..6726a98d
--- /dev/null
+++ b/services/voice/CLAUDE.md
@@ -0,0 +1,7 @@
+# Voice Full Support
+
+Engines defined in `services/voice/voiceTypes.ts` (`SttEngine`, `TtsEngine`, `VadEngine`, `WakeWordEngine`, `IntentEngine`). Contract: `isAvailable()` → `initialize()` → use → `dispose()`. Web Speech API fallbacks: `WebSpeechSttEngine`, `WebSpeechTtsEngine`, `WebRtcVadEngine` (zero downloads). WASM path (B-2, `enableVoiceWasm`): `WasmSttEngine` (Whisper.cpp) + `SileroVadEngine`; model download via `VoiceModelDownloadModal` + `VoiceCommandService.preloadModel(modelType)`.
+
+**Intent engine:** `HybridIntentEngine.parse(transcript, context)` — exact match → fuzzy Jaccard + slot extraction. **Orchestrator:** `VoiceCommandService` singleton (state machine), dispatches via `runCommandById`, `appStoreRef` for Redux outside React. **Hooks:** `useVoice`, `usePushToTalk` (Ctrl+Shift+V), `useVoiceDictation`, `useVoiceAccessibility`. **Gating:** `settings.voice.enabled && featureFlags.enableVoiceSupport`. **Never log transcripts** — transcript text is PII and must never reach `services/logger.ts` (including its IDB log sink); only redacted event metadata (e.g. command id, intent match) may be logged. `startListening` has a single-flight guard (C-P1).
+
+**Voice E2E seam (P1-2):** `services/voice/voiceTestSeam.ts` — `getVoiceTestHarness()` reads `window.__voiceTestHarness` (only ever set by Playwright `addInitScript`; undefined in production). `createSttEngine`/`createVadEngine` return injected mock engines, and `downloadVoiceModels` runs a simulated download, when the harness is present. Installers: `tests/e2e/mocks/voiceMockEngines.ts`. Deterministic suite: `tests/e2e/deep/voice/whisper-stt.spec.ts` (e2e-deep); real-inference nightly: `whisper-real.spec.ts` + `voice-nightly.yml` (`RUN_REAL_VOICE_E2E=1`). Chromium fake-media flags live in `playwright.config.ts`.
diff --git a/tests/CLAUDE.md b/tests/CLAUDE.md
new file mode 100644
index 00000000..296d2dda
--- /dev/null
+++ b/tests/CLAUDE.md
@@ -0,0 +1,15 @@
+# Test mock patterns
+
+**useAppSelectorShallow with plotBoard:** Include `plotBoard: { activeMode: 'swimlane', snapToGrid: false, selectedConnectionId: null, isDrawingConnection: false, drawFromSectionId: null, activeSubplotFilter: null, zoom: 1, panX: 0, panY: 0 }` in mock state. Connections/subplots/tensionOverrides are in `project.present.data`. **Existing convention** (10 test files, e.g. `tests/unit/SceneBoardView.test.tsx`): the `useAppSelector`/`useAppSelectorShallow` mock is typed `(selector: (s: any) => unknown)` with `// biome-ignore lint/suspicious/noExplicitAny: test mock` above it, since the partial mock state doesn't satisfy the full `RootState` type. This documents that pre-existing pattern for consistency — do not add a *new* suppression elsewhere; a shared, properly-typed helper (`(selector: (s: RootState) => unknown) => selector(mockState as RootState)`, no `any` needed) that replaces all 10 call sites is a good candidate for a future dedicated refactor, not a one-off addition here.
+
+**FeatureFlagsState mocks:** Always include ALL 22 flags (TypeScript strict rejects partial). Only **six** default **off**: `enableProForge`, `enableRtlLayout`, `enableVoiceSupport`, `enableVoiceWasm`, `enableGlobalCopilot`, `enableLocalFirstSync` — every other flag (including all edge-AI flags and `enableIdbAtRestEncryption`) defaults **on**. When a test needs a non-default flag state, set it explicitly in the mock — don't assume the default.
+
+**ConnectionLayer test IDs:** `data-testid="connection-group"` — query by testid, not role.
+
+**DuckDB in tests:** Mock `services/duckdb/duckdbClient` with `{ execAsync: vi.fn(), queryAsync: vi.fn() }`. Never initialize real DuckDB-WASM.
+
+**AI thunk tests:** `settingsReducer` defaults to `privacy.localStorageOnly: true` → AI thunks throw. Fix: mock `services/ai/aiPolicy` with `assertCloudAiAllowedSync: vi.fn()` + `assertCloudAiAllowed: vi.fn().mockResolvedValue(undefined)` before all imports.
+
+**Context hooks in component tests:** Mock the context module (`vi.mock('../../../contexts/XyzContext', ...)`) rather than wrapping in the real provider tree. Apply for any `use*ViewContext` hook.
+
+**Custom Select/LanguageSelector mocks:** Mock as native `<select>` element for testing-library compatibility. See canonical pattern in existing settings test files.
diff --git a/tests/unit/aiProviderService.test.ts b/tests/unit/aiProviderService.test.ts
index 3705deeb..5b2cc25b 100644
--- a/tests/unit/aiProviderService.test.ts
+++ b/tests/unit/aiProviderService.test.ts
@@ -236,12 +236,14 @@ describe('testAIConnection', () => {
     const result = await testAIConnection('anthropic', {});
     expect(result.ok).toBe(false);
     expect(result.error).toContain('CORS');
+    expect(result.kind).toBe('backendProxyRequired');
   });
 
   it('returns ok:false for ollama in browser (no Tauri)', async () => {
     const result = await testAIConnection('ollama', {});
     expect(result.ok).toBe(false);
     expect(result.error).toContain('desktop app');
+    expect(result.kind).toBe('desktopRequired');
   });
 
   it('returns ok:false when openai key is missing', async () => {
@@ -249,6 +251,8 @@ describe('testAIConnection', () => {
     const result = await testAIConnection('openai', {});
     expect(result.ok).toBe(false);
     expect(result.error).toContain('Key');
+    expect(result.kind).toBe('noApiKey');
+    expect(result.params).toEqual({ provider: 'OpenAI' });
   });
 
   it('returns ok:false when gemini key is missing', async () => {
@@ -256,11 +260,14 @@ describe('testAIConnection', () => {
     const result = await testAIConnection('gemini', {});
     expect(result.ok).toBe(false);
     expect(result.error).toContain('key');
+    expect(result.kind).toBe('noApiKey');
+    expect(result.params).toEqual({ provider: 'Gemini' });
   });
 
   it('returns ok:false for unknown provider', async () => {
     const result = await testAIConnection('unknown' as never, {});
     expect(result.ok).toBe(false);
+    expect(result.kind).toBe('unknownProvider');
   });
 
   it('returns ok:true for webllm when WebGPU is available', async () => {
@@ -274,6 +281,7 @@ describe('testAIConnection', () => {
     const result = await testAIConnection('webllm', {});
     expect(result.ok).toBe(false);
     expect(result.error).toContain('WebGPU');
+    expect(result.kind).toBe('noWebgpu');
   });
 
   it('returns ok:true for onnx (WASM always available)', async () => {
@@ -614,6 +622,8 @@ describe('testAIConnection additional branches', () => {
     const result = await testAIConnection('openai', {});
     expect(result.ok).toBe(false);
     expect(result.error).toContain('403');
+    expect(result.kind).toBe('httpError');
+    expect(result.params).toEqual({ status: 403 });
   });
 
   it('grok: returns ok:false when no API key', async () => {
@@ -621,6 +631,8 @@ describe('testAIConnection additional branches', () => {
     const result = await testAIConnection('grok', {});
     expect(result.ok).toBe(false);
     expect(result.error).toContain('Key');
+    expect(result.kind).toBe('noApiKey');
+    expect(result.params).toEqual({ provider: 'Grok' });
   });
 
   it('grok: returns ok:false on HTTP 429', async () => {
@@ -629,6 +641,8 @@ describe('testAIConnection additional branches', () => {
     const result = await testAIConnection('grok', {});
     expect(result.ok).toBe(false);
     expect(result.error).toContain('429');
+    expect(result.kind).toBe('httpError');
+    expect(result.params).toEqual({ status: 429 });
   });
 
   it('gemini: returns ok:true on HTTP 200', async () => {
@@ -644,6 +658,8 @@ describe('testAIConnection additional branches', () => {
     const result = await testAIConnection('gemini', {});
     expect(result.ok).toBe(false);
     expect(result.error).toContain('401');
+    expect(result.kind).toBe('httpError');
+    expect(result.params).toEqual({ status: 401 });
   });
 
   it('returns ok:false with error message when fetch throws (catch branch)', async () => {
@@ -652,6 +668,10 @@ describe('testAIConnection additional branches', () => {
     const result = await testAIConnection('openai', {});
     expect(result.ok).toBe(false);
     expect(result.error).toContain('network error');
+    expect(result.kind).toBe('unexpected');
+    // QNBS-v3 (CodeAnt CWE-209): the raw message stays in `error` (for logs) only -- `unexpected`
+    // has no `params`, since the i18n string for it never interpolates the raw exception text.
+    expect(result.params).toBeUndefined();
   });
 });
 
diff --git a/tests/unit/localServerHttpPluginUnavailable.test.ts b/tests/unit/localServerHttpPluginUnavailable.test.ts
new file mode 100644
index 00000000..8bdc3ed0
--- /dev/null
+++ b/tests/unit/localServerHttpPluginUnavailable.test.ts
@@ -0,0 +1,36 @@
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+
+// QNBS-v3: dedicated file — a fresh module registry is required so the mocked
+// '@tauri-apps/plugin-http' throws on its FIRST import in this process; sharing a file with tests
+// that successfully import it would cache the resolved module and mask this regression class
+// (the exact build-config bug fixed alongside this test: see the 2026-07-28 update in
+// docs/adr/0012-local-server-connectivity-tauri-http.md).
+
+vi.mock('@tauri-apps/plugin-http', () => {
+  throw new Error("Failed to resolve module specifier '@tauri-apps/plugin-http'");
+});
+
+vi.mock('../../services/tauriRuntime', () => ({
+  isTauriRuntime: () => true,
+}));
+
+import { LocalServerError, localServerFetch } from '../../services/localServerHttp';
+
+beforeEach(() => {
+  vi.stubGlobal('fetch', vi.fn());
+});
+
+afterEach(() => {
+  vi.unstubAllGlobals();
+});
+
+describe('localServerFetch (tauri plugin load failure)', () => {
+  it('classifies a failed @tauri-apps/plugin-http import as LocalServerError(plugin_unavailable), not unreachable', async () => {
+    const err = await localServerFetch('http://localhost:11434/api/tags').catch((e) => e);
+    expect(err).toBeInstanceOf(LocalServerError);
+    expect(err.kind).toBe('plugin_unavailable');
+    expect(err.message).toMatch(/desktop HTTP plugin failed to load/i);
+    // The generic web fetch must never be used as a silent fallback for a broken desktop transport.
+    expect(fetch).not.toHaveBeenCalled();
+  });
+});
diff --git a/tests/unit/ollamaService.test.ts b/tests/unit/ollamaService.test.ts
index f8e29664..2d819c93 100644
--- a/tests/unit/ollamaService.test.ts
+++ b/tests/unit/ollamaService.test.ts
@@ -1,4 +1,5 @@
 import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+import { LocalServerError, localServerFetch } from '../../services/localServerHttp';
 import {
   listOllamaModels,
   type OllamaPullProgress,
@@ -7,6 +8,15 @@ import {
   testOllamaConnection,
 } from '../../services/ollamaService';
 
+// QNBS-v3: real implementation by default (all other tests drive it via the stubbed global
+// fetch); only the plugin_unavailable test below overrides it directly, since that error
+// originates in resolveFetch() BEFORE localServerFetch's own try/catch — mocking global fetch
+// can't reproduce it (localServerFetch would just re-wrap it as a generic 'unreachable').
+vi.mock('../../services/localServerHttp', async (importOriginal) => {
+  const actual = await importOriginal<typeof import('../../services/localServerHttp')>();
+  return { ...actual, localServerFetch: vi.fn(actual.localServerFetch) };
+});
+
 /** Build a streaming NDJSON Response (one JSON object per line) for /api/pull mocks. */
 function ndjsonResponse(objects: object[], status = 200): Response {
   const enc = new TextEncoder();
@@ -76,6 +86,8 @@ describe('testOllamaConnection', () => {
     const result = await testOllamaConnection();
     expect(result.ok).toBe(false);
     expect(result.error).toContain('503');
+    expect(result.kind).toBe('httpError');
+    expect(result.params).toEqual({ status: 503 });
   });
 
   it('returns ok:false with error message on network failure', async () => {
@@ -84,6 +96,10 @@ describe('testOllamaConnection', () => {
     expect(result.ok).toBe(false);
     expect(result.error).toContain('localhost:11434');
     expect(result.error).toContain('Connection refused');
+    expect(result.kind).toBe('unreachable');
+    // QNBS-v3 (CodeAnt CWE-209): the raw technical message stays in `error` (for logs) only —
+    // `params` must never carry it, since it's interpolated into the user-facing i18n string.
+    expect(result.params).toEqual({ url: 'http://localhost:11434' });
   });
 
   it('classifies a TimeoutError-shaped rejection as a timeout (#266)', async () => {
@@ -92,6 +108,22 @@ describe('testOllamaConnection', () => {
     const result = await testOllamaConnection('http://localhost:11434');
     expect(result.ok).toBe(false);
     expect(result.error).toContain('timed out');
+    expect(result.kind).toBe('timeout');
+    expect(result.params).toEqual({ url: 'http://localhost:11434' });
+  });
+
+  it('surfaces a plugin_unavailable LocalServerError distinctly, not as generic "not reachable"', async () => {
+    const pluginErr = new LocalServerError(
+      'plugin_unavailable',
+      'Local server networking is unavailable: the desktop HTTP plugin failed to load.',
+    );
+    vi.mocked(localServerFetch).mockRejectedValueOnce(pluginErr);
+    const result = await testOllamaConnection('http://localhost:11434');
+    expect(result.ok).toBe(false);
+    expect(result.error).toBe(pluginErr.message);
+    expect(result.error).not.toContain('not reachable');
+    expect(result.kind).toBe('pluginUnavailable');
+    expect(result.params).toBeUndefined();
   });
 });
 
diff --git a/tests/unit/resolveViteBase.test.ts b/tests/unit/resolveViteBase.test.ts
index 19b32305..cecc147d 100644
--- a/tests/unit/resolveViteBase.test.ts
+++ b/tests/unit/resolveViteBase.test.ts
@@ -1,8 +1,32 @@
 import { describe, expect, it } from 'vitest';
-import { GITHUB_PAGES_BASE, resolveViteBase } from '../../config/resolveViteBase';
+import { GITHUB_PAGES_BASE, isTauriBuild, resolveViteBase } from '../../config/resolveViteBase';
+
+// QNBS-v3: Guards the desktop discovery regression — a Tauri build that externalized
+// @tauri-apps/* left an unresolvable bare module specifier in the shipped bundle (services/
+// localServerHttp.ts's dynamic `@tauri-apps/plugin-http` import). isTauriBuild is the single
+// source of truth vite.config.ts uses to keep rollupOptions.external in sync with `base`.
+describe('isTauriBuild', () => {
+  it('is true for Tauri 2.x desktop builds (TAURI_ENV_PLATFORM)', () => {
+    expect(isTauriBuild({ TAURI_ENV_PLATFORM: 'linux' })).toBe(true);
+  });
+
+  it('is true for the legacy Tauri 1.x env var (TAURI_PLATFORM)', () => {
+    expect(isTauriBuild({ TAURI_PLATFORM: 'windows' })).toBe(true);
+  });
+
+  it('is false for the web/PWA build with neither env var set', () => {
+    expect(isTauriBuild({})).toBe(false);
+    expect(isTauriBuild({ VITE_BASE: '/foo', DEPLOY_TARGET: 'edge' })).toBe(false);
+  });
+
+  it('is false for an empty-string or whitespace-only marker value (not a real platform)', () => {
+    expect(isTauriBuild({ TAURI_ENV_PLATFORM: '' })).toBe(false);
+    expect(isTauriBuild({ TAURI_ENV_PLATFORM: '   ' })).toBe(false);
+    expect(isTauriBuild({ TAURI_PLATFORM: '' })).toBe(false);
+    expect(isTauriBuild({ TAURI_PLATFORM: '   ' })).toBe(false);
+  });
+});
 
-// QNBS-v3: Guards the desktop blank-screen regression — Tauri builds MUST resolve to a relative
-// base so hashed assets load under tauri://localhost/ instead of 404ing on the GitHub Pages path.
 describe('resolveViteBase', () => {
   it('returns a relative base for Tauri 2.x desktop builds (TAURI_ENV_PLATFORM)', () => {
     expect(resolveViteBase({ TAURI_ENV_PLATFORM: 'windows' })).toBe('./');
@@ -14,6 +38,10 @@ describe('resolveViteBase', () => {
     expect(resolveViteBase({ TAURI_PLATFORM: 'windows' })).toBe('./');
   });
 
+  it('does not treat an empty-string TAURI_ENV_PLATFORM as a Tauri build', () => {
+    expect(resolveViteBase({ TAURI_ENV_PLATFORM: '' })).toBe(GITHUB_PAGES_BASE);
+  });
+
   it('lets Tauri detection win over VITE_BASE and DEPLOY_TARGET', () => {
     expect(
       resolveViteBase({ TAURI_ENV_PLATFORM: 'windows', VITE_BASE: '/x/', DEPLOY_TARGET: 'edge' }),
diff --git a/tests/unit/settings/AiProviderCard.test.tsx b/tests/unit/settings/AiProviderCard.test.tsx
index b9bba98e..1eca98cd 100644
--- a/tests/unit/settings/AiProviderCard.test.tsx
+++ b/tests/unit/settings/AiProviderCard.test.tsx
@@ -222,4 +222,47 @@ describe('AiProviderCard — ollama provider (#266)', () => {
       localBackendPreset: 'ollama_default',
     });
   });
+
+  it('desktop: a classified test failure renders the translated key, not the raw technical message', async () => {
+    setDesktopRuntime(true);
+    vi.mocked(testAIConnection).mockResolvedValue({
+      ok: false,
+      error: 'Ollama HTTP 503',
+      kind: 'httpError',
+      params: { status: 503 },
+    });
+    render(
+      <AiProviderCard
+        advancedAi={ollamaAdvancedAi}
+        onAdvancedAiPatch={mockOnAdvancedAiPatch}
+        onProviderChange={mockOnProviderChange}
+      />,
+    );
+    await waitFor(() => {
+      // QNBS-v3: the translated text renders in two places (the status-badge error line and the
+      // manual "Test connection" result span) — both share the same `testError` state.
+      expect(screen.getAllByText('settings.ai.testError.httpError').length).toBeGreaterThan(0);
+    });
+    expect(screen.queryByText('Ollama HTTP 503')).toBeNull();
+  });
+
+  it('desktop: an unexpected failure (no params) renders the translated key without crashing', async () => {
+    setDesktopRuntime(true);
+    vi.mocked(testAIConnection).mockResolvedValue({
+      ok: false,
+      error: 'TypeError: something internal broke at services/foo.ts:42',
+      kind: 'unexpected',
+    });
+    render(
+      <AiProviderCard
+        advancedAi={ollamaAdvancedAi}
+        onAdvancedAiPatch={mockOnAdvancedAiPatch}
+        onProviderChange={mockOnProviderChange}
+      />,
+    );
+    await waitFor(() => {
+      expect(screen.getAllByText('settings.ai.testError.unexpected').length).toBeGreaterThan(0);
+    });
+    expect(screen.queryByText(/something internal broke/)).toBeNull();
+  });
 });
diff --git a/vite.config.ts b/vite.config.ts
index 740f245b..1847cafd 100644
--- a/vite.config.ts
+++ b/vite.config.ts
@@ -4,11 +4,12 @@ import react from '@vitejs/plugin-react';
 import { visualizer } from 'rollup-plugin-visualizer';
 import { defineConfig } from 'vite';
 import { VitePWA } from 'vite-plugin-pwa';
-import { GITHUB_PAGES_BASE, resolveViteBase } from './config/resolveViteBase';
+import { GITHUB_PAGES_BASE, isTauriBuild, resolveViteBase } from './config/resolveViteBase';
 
 const isAnalyze = process.env['ANALYZE'] === 'true';
 
 const deployBase = resolveViteBase();
+const isTauri = isTauriBuild();
 
 export default defineConfig({
   base: deployBase,
@@ -123,10 +124,13 @@ export default defineConfig({
         ),
     },
     rollupOptions: {
-      external: [
-        // QNBS-v3: Tauri APIs are only available in the desktop shell — externalize all @tauri-apps/* so the web/Vercel build never tries to bundle them.
-        /^@tauri-apps\//,
-      ],
+      // QNBS-v3: externalizing @tauri-apps/* must NOT apply to the Tauri desktop build itself —
+      // services/localServerHttp.ts's dynamic `await import('@tauri-apps/plugin-http')` would be
+      // left as an unresolvable bare specifier in the packaged .deb/.msi, silently breaking local
+      // Ollama/LM Studio/vLLM discovery (see docs/adr/0012-local-server-connectivity-tauri-http.md).
+      // Web/Vercel builds still externalize them — those code paths are gated by isTauriRuntime()
+      // and never exercised there.
+      external: isTauri ? [] : [/^@tauri-apps\//],
       output: {
         // Asset hashing for cache busting
         entryFileNames: 'assets/[name]-[hash].js',