diff --git a/apps/web/src/components/settings/VoiceSettingsPanel.browser.test.tsx b/apps/web/src/components/settings/VoiceSettingsPanel.browser.test.tsx
index ab1c0187c2806..c4311915a700d 100644
--- a/apps/web/src/components/settings/VoiceSettingsPanel.browser.test.tsx
+++ b/apps/web/src/components/settings/VoiceSettingsPanel.browser.test.tsx
@@ -20,7 +20,8 @@ import { VoiceSettingsPanel } from "./VoiceSettingsPanel";
 
 vi.mock(import("../../speech/useEnvironmentSpeechInput"), { spy: true });
 
-it("preserves the model test transcript when its model is filtered out", async () => {
+it("keeps setup available with a non-English locale and preserves a filtered test transcript", async () => {
+  using _languages = vi.spyOn(navigator, "languages", "get").mockReturnValue(["fr-FR"]);
   const target = new PrimaryConnectionTarget({
     environmentId: EnvironmentId.make("voice-filter"),
     label: "Test environment",
@@ -40,12 +41,12 @@ it("preserves the model test transcript when its model is filtered out", async (
     supportsStreaming: false,
     supportsTranslation: false,
     supportsLanguageDetection: false,
-    active: true,
-    state: "installed",
+    active: false,
+    state: "downloadable",
   };
   const status = {
     supported: true,
-    state: "ready",
+    state: "missing-model",
     modelId: model.id,
     model: model.name,
     size: model.size,
@@ -103,6 +104,7 @@ it("preserves the model test transcript when its model is filtered out", async (
             model: null,
             downloading: false,
             error: null,
+            show: async () => {},
             setOpen: () => {},
             download: async () => {},
             cancelDownload: async () => {},
@@ -128,6 +130,13 @@ it("preserves the model test transcript when its model is filtered out", async (
               <VoiceSettingsPanel />
             </SettingsScopeProvider>,
           );
+          await expect
+            .element(screen.getByRole("button", { name: "Set up voice input", exact: true }))
+            .toBeVisible();
+          await expect.element(screen.getByText("Bilingual", { exact: true })).toBeVisible();
+          await userEvent.click(
+            screen.getByRole("button", { name: "Set up voice input", exact: true }),
+          );
           const testModel = screen.getByRole("button", { name: "Test model", exact: true });
           await expect.element(testModel).toBeVisible();
           await userEvent.click(testModel);
@@ -137,13 +146,22 @@ it("preserves the model test transcript when its model is filtered out", async (
           const search = screen.getByRole("textbox", { name: "Search transcription models" });
           await userEvent.fill(search, "no matching model");
           await expect.element(screen.getByText("No models match.")).toBeVisible();
+          await expect
+            .element(screen.getByRole("button", { name: "Set up voice input", exact: true }))
+            .toBeVisible();
           await userEvent.clear(search);
           await userEvent.fill(screen.getByRole("textbox", { name: "Search languages" }), "fr");
           const french = screen.getByRole("button", { name: "French", exact: true });
           await userEvent.click(french);
           await expect.element(screen.getByText("Bilingual", { exact: true })).toBeVisible();
           await expect.element(screen.getByText("B+", { exact: true })).toBeVisible();
-          expect(screen.container.querySelector("#transcription-test")).toHaveAttribute("hidden");
+          await expect
+            .element(
+              screen.container.querySelector<HTMLParagraphElement>(
+                '[aria-label="Test transcript"]',
+              ),
+            )
+            .not.toBeVisible();
           await userEvent.click(french);
           await expect.element(result).toBeVisible();
           await vi.waitFor(() =>
diff --git a/apps/web/src/components/settings/VoiceSettingsPanel.tsx b/apps/web/src/components/settings/VoiceSettingsPanel.tsx
index c08844bafe09b..a7bdfccc8c5d7 100644
--- a/apps/web/src/components/settings/VoiceSettingsPanel.tsx
+++ b/apps/web/src/components/settings/VoiceSettingsPanel.tsx
@@ -36,6 +36,9 @@ import {
 } from "lucide-react";
 import { useEffect, useRef, useState, type ReactNode } from "react";
 
+import { useAtomValue } from "@effect/atom-react";
+import { shortcutLabelForCommand } from "../../keybindings";
+import { primaryServerKeybindingsAtom } from "../../state/server";
 import { useStableCallback } from "~/hooks/useStableCallback";
 import { listBrowserMicrophones, type BrowserMicrophone } from "../../speech/browserMicrophones";
 import {
@@ -126,6 +129,7 @@ type ModelCardProps = {
   readonly onSelect: () => void;
   readonly onCancel: () => void;
   readonly onDelete: () => void;
+  readonly hidden?: boolean;
   readonly testControls?: ReactNode;
   readonly children?: ReactNode;
 };
@@ -147,7 +151,7 @@ function ModelCard(props: ModelCardProps) {
   const downloading = model.state === "downloading" || model.state === "verifying";
   const progress = model.downloaded === undefined ? 0 : (model.downloaded / model.size) * 100;
   return (
-    <div className="flex flex-wrap items-start gap-3 px-3 py-2.5 sm:px-4">
+    <div hidden={props.hidden} className="flex flex-wrap items-start gap-3 px-3 py-2.5 sm:px-4">
       <div className="min-w-0 flex-1">
         <div className="flex flex-wrap items-center gap-1.5">
           <span className="text-sm font-medium leading-5">{model.name}</span>
@@ -737,7 +741,9 @@ function SpeechModelBrowser(props: {
         (!query || (model.name + " " + model.description).toLocaleLowerCase().includes(query)),
     )
     .toSorted((left, right) => modelSortOrder(left) - modelSortOrder(right));
-  const activeModel = models.find((model) => model.active);
+  const activeModel = models.find(
+    (model) => model.active || (status?.supported && model.id === status.modelId),
+  );
   const renderedModels =
     activeModel && !sortedModels.includes(activeModel)
       ? [activeModel, ...sortedModels]
@@ -856,11 +862,15 @@ function SpeechModelBrowser(props: {
                         runtime.runPromise(removeEnvironmentSpeechModel(prepared, model.id)),
                       ),
                   };
-                  return model.active ? (
+                  return model === activeModel ? (
                     <TranscriptionTest
                       key={model.id}
                       card={card}
                       hidden={!sortedModels.includes(model)}
+                      onSetup={() => {
+                        setSearch("");
+                        setSelection([]);
+                      }}
                       environmentId={props.testEnvironmentId}
                       projectId={props.projectId}
                       projectName={props.projectName}
@@ -881,6 +891,7 @@ function SpeechModelBrowser(props: {
 function TranscriptionTest(props: {
   readonly card: ModelCardProps;
   readonly hidden: boolean;
+  readonly onSetup: () => void;
   readonly environmentId: EnvironmentId;
   readonly projectId: ProjectId | null;
   readonly projectName: string | null;
@@ -907,9 +918,24 @@ function TranscriptionTest(props: {
     commitDraft: (text) => setTranscript(text),
   });
   return (
-    <SettingsSearchTarget {...searchableSetting("transcription-test")} hidden={props.hidden}>
+    <SettingsSearchTarget {...searchableSetting("transcription-test")}>
+      {speech.available && speech.state.phase === "idle" ? (
+        <div className="flex justify-end px-3 py-2 sm:px-4">
+          <Button
+            size="sm"
+            onClick={() => {
+              props.onSetup();
+              setShowTranscript(true);
+              void speech.setup.show();
+            }}
+          >
+            Set up voice input
+          </Button>
+        </div>
+      ) : null}
       <ModelCard
         {...props.card}
+        hidden={props.hidden}
         testControls={
           <div className="flex min-w-0 max-w-full items-center gap-1">
             <ComposerSpeechRecordingPill
@@ -1001,6 +1027,10 @@ function TranscriptionTest(props: {
 }
 
 export function VoiceSettingsPanel() {
+  const keybindings = useAtomValue(primaryServerKeybindingsAtom);
+  const dictationShortcut = shortcutLabelForCommand(keybindings, "composer.dictation", {
+    context: { terminalFocus: false, terminalOpen: false },
+  });
   const cleanupEnabled = useCleanupGrant() !== null;
   const settings = useClientSettings();
   const settingsScope = useOptionalSettingsScope();
@@ -1113,7 +1143,11 @@ export function VoiceSettingsPanel() {
         <SettingsRow
           {...searchableSetting("dictation-shortcut-mode")}
           title="Dictation shortcut"
-          description="How the dictation shortcut starts and stops recording."
+          description={
+            dictationShortcut
+              ? `Press ${dictationShortcut} to dictate in the focused editor. Choose how it starts and stops recording.`
+              : "No dictation shortcut is assigned. Set one in Keybindings settings."
+          }
           control={
             <Select
               value={settings.voiceShortcutMode}
@@ -1143,8 +1177,8 @@ export function VoiceSettingsPanel() {
           title="Send microphone audio to"
           description={
             <>
-              Microphone audio is sent to this environment to transcribe, then discarded. Choose Off
-              to turn voice input off on this device. This choice is saved on this device only.
+              Audio is sent here to transcribe, then discarded. Choose Off to disable voice input.
+              Saved on this device only.
             </>
           }
           control={
@@ -1182,6 +1216,11 @@ export function VoiceSettingsPanel() {
             </Select>
           }
         >
+          {!destination.environmentId ? (
+            <p className="mt-2 text-xs text-muted-foreground">
+              Choose an environment to enable voice input, then select Set up voice input below.
+            </p>
+          ) : null}
           {destination.environmentId && chosenUnsupported ? (
             <p className="mt-2 text-xs text-destructive">
               {chosenEnvironment.label} can’t transcribe audio. Pick an environment that can to use
diff --git a/apps/web/src/speech/useEnvironmentSpeechInput.browser.test.tsx b/apps/web/src/speech/useEnvironmentSpeechInput.browser.test.tsx
index a41f359f68ef4..8f705fca4e425 100644
--- a/apps/web/src/speech/useEnvironmentSpeechInput.browser.test.tsx
+++ b/apps/web/src/speech/useEnvironmentSpeechInput.browser.test.tsx
@@ -39,10 +39,16 @@ const mocks = vi.hoisted(() => ({
   recordingOptions: [] as SpeechTranscriptionOptions[],
   projectWords: {} as Record<string, SpeechCustomWords>,
   transcripts: [] as Array<string | Error>,
+  navigate: vi.fn(),
   committed: vi.fn(),
   draft: "",
 }));
 
+vi.mock("@tanstack/react-router", async (importOriginal) => ({
+  ...(await importOriginal<typeof import("@tanstack/react-router")>()),
+  useNavigate: () => mocks.navigate,
+}));
+
 vi.mock("../state/session", () => ({
   useEnvironmentScope: (id: string) => mocks.canDictate.has(id),
   usePreparedConnection: (environmentId: string | null) => {
@@ -166,6 +172,7 @@ const THREAD_ENV = "thread-env";
 
 function setup(options: { readonly destination?: string | null } = {}) {
   mocks.settings = {};
+  mocks.navigate.mockReset();
   setDestination(options.destination ?? null);
   chooseTranscriptCleanup(false);
   mocks.primaryEnvironmentId = "primary-env";
@@ -238,7 +245,11 @@ it.each([
   await act(() => mocks.voice.start());
   expectNoAudioSent();
   expect(mocks.committed).not.toHaveBeenCalled();
-  expect(mocks.voice.available).toBe(false);
+  expect(mocks.voice.available).toBe(true);
+  expect(mocks.navigate).toHaveBeenCalledWith({
+    to: "/settings/voice",
+    hash: "voice-transcription-destination",
+  });
   expect(mocks.voice.transcriptionEnvironmentId).toBeNull();
 });
 
@@ -309,6 +320,25 @@ it("retries a failed transcription on the same chosen environment", async () =>
   expect(mocks.committed).toHaveBeenCalledExactlyOnceWith("second try", expect.anything());
 });
 
+it.each(["missing-model", "ready"] as const)(
+  "opens setup explicitly with a %s model without recording",
+  async (modelState) => {
+    setup({ destination: "env-a" });
+    mocks.modelState = modelState;
+    await using _screen = await render(<Probe />);
+    await act(() => mocks.voice.setup.show());
+    expect(mocks.voice.setup).toMatchObject({
+      open: true,
+      step: modelState === "missing-model" ? 0 : 1,
+    });
+    expect(mocks.microphoneRequests).toEqual([]);
+    expect(mocks.transcribeCalls).toEqual([]);
+    await act(() => mocks.voice.setup.startRecording());
+    expect(mocks.voice.setup.open).toBe(modelState === "missing-model");
+    expect(mocks.microphoneRequests).toEqual(modelState === "ready" ? ["env-a"] : []);
+  },
+);
+
 it("reopens setup after reconnecting during a download and ignores its stale completion", async () => {
   setup({ destination: "env-a" });
   mocks.modelState = "missing-model";
@@ -370,7 +400,7 @@ it("revokes across windows even before the storage event and survives stale pref
     storage.removeItem(key);
     expect(canSend()).toBe(false);
     writeBrowserClientSettings({ ...DEFAULT_CLIENT_SETTINGS, wordWrap: false });
-    await vi.waitFor(() => expect(mocks.voice.available).toBe(false));
+    await vi.waitFor(() => expect(mocks.voice.transcriptionEnvironmentId).toBeNull());
     expect(mocks.voice.state.phase).toBe("idle");
     await act(() => mocks.voice.stop());
     expect(mocks.transcribeCalls).toEqual([]);
diff --git a/apps/web/src/speech/useEnvironmentSpeechInput.ts b/apps/web/src/speech/useEnvironmentSpeechInput.ts
index 644edb303aef2..c79994d94af03 100644
--- a/apps/web/src/speech/useEnvironmentSpeechInput.ts
+++ b/apps/web/src/speech/useEnvironmentSpeechInput.ts
@@ -25,6 +25,7 @@ import { resolveProjectSettings } from "@t3tools/shared/projectSettings";
 import { resolveSpeechVocabulary } from "@t3tools/shared/speech";
 import * as Option from "effect/Option";
 import { useEffect, useRef, useState } from "react";
+import { useNavigate } from "@tanstack/react-router";
 
 import { useStableCallback } from "~/hooks/useStableCallback";
 import { toastManager } from "../components/ui/toast";
@@ -119,6 +120,7 @@ export function useVoiceTranscriptionDestination() {
 }
 
 export function useEnvironmentSpeechInput(input: HookInput) {
+  const navigate = useNavigate();
   const destination = useVoiceTranscriptionDestination();
   const prepared = destination.prepared;
   const grantId = destination.grantId;
@@ -423,7 +425,12 @@ export function useEnvironmentSpeechInput(input: HookInput) {
     if (controller.currentState.phase === "recording") void controller.stop();
   });
 
-  const start = useStableCallback(async () => {
+  const start = useStableCallback(async (setupOnly = false) => {
+    if (latestRef.current.input.disabled) return;
+    if (!grantId) {
+      await navigate({ to: "/settings/voice", hash: "voice-transcription-destination" });
+      return;
+    }
     const request = ++startRequestRef.current;
     const controller = controllerRef.current;
     const owner = latestRef.current.input.ownerKey;
@@ -458,8 +465,12 @@ export function useEnvironmentSpeechInput(input: HookInput) {
       });
       return;
     }
-    if (latestStatus.state === "missing-model") {
-      setSetup({ ...INITIAL_SETUP, open: true });
+    if (setupOnly || latestStatus.state === "missing-model") {
+      setSetup({
+        ...INITIAL_SETUP,
+        open: true,
+        step: latestStatus.state === "missing-model" ? 0 : 1,
+      });
       return;
     }
     setLevel(0);
@@ -513,9 +524,10 @@ export function useEnvironmentSpeechInput(input: HookInput) {
 
   return {
     available:
-      prepared !== null &&
-      status?.prepared === prepared &&
-      status.value?.supported !== false &&
+      (!grantId ||
+        (prepared !== null &&
+          status?.prepared === prepared &&
+          status.value?.supported !== false)) &&
       typeof navigator !== "undefined" &&
       Boolean(navigator.mediaDevices?.getUserMedia) &&
       typeof AudioWorkletNode !== "undefined",
@@ -524,6 +536,7 @@ export function useEnvironmentSpeechInput(input: HookInput) {
     transcriptionEnvironmentLabel: destination.label,
     setup: {
       ...setup,
+      show: () => start(true),
       setOpen: (open: boolean) => setSetup((current) => ({ ...current, open })),
       download: downloadSetupModel,
       cancelDownload: cancelSetupDownload,
diff --git a/docs/user/voice-input.md b/docs/user/voice-input.md
index e204b28660626..212308ed911da 100644
--- a/docs/user/voice-input.md
+++ b/docs/user/voice-input.md
@@ -15,6 +15,8 @@ to a named environment on that phone, open **Settings → Voice → Transcribe o
 On web and desktop, transcription runs on one of your T3 Code environments.
 Nothing is sent until you choose one: open **Settings → Voice** and set
 **Send microphone audio to** the environment that should transcribe for you.
+The microphone and dictation shortcut open these settings when no destination
+has been chosen.
 Microphone audio recorded on this device is sent to that environment over your
 authenticated connection, transcribed there, and discarded. The choice is saved
 on this device only, so every computer and browser you use asks for it once.
@@ -25,8 +27,10 @@ substitutes another environment.
 
 ## Set up a model
 
-The first time you dictate, T3 Code offers to download the recommended English
-model to the chosen environment. In **Settings → Voice → Transcription**, select
+Select **Set up voice input** in Voice settings to open
+the setup wizard. It offers a model download if needed, then lets you choose a
+microphone and start dictating. The first time you dictate with a missing model,
+the same wizard opens. In **Settings → Voice → Transcription**, select
 the languages you dictate in to compare models that support all of them.
 Accuracy shows how often the model got words or characters wrong in each
 language; speed is the time to transcribe 30 seconds on a reference GPU. Choose