diff --git a/apps/stage-pocket/vite.config.ts b/apps/stage-pocket/vite.config.ts
index 1dcc975a1d..4ba7f06c5e 100644
--- a/apps/stage-pocket/vite.config.ts
+++ b/apps/stage-pocket/vite.config.ts
@@ -34,6 +34,7 @@ function isEnvTruthy(value: string | undefined | null): boolean {
 
 const stageUIAssetsRoot = resolve(join(import.meta.dirname, '..', '..', 'packages', 'stage-ui', 'src', 'assets'))
 const sharedCacheDir = resolve(join(import.meta.dirname, '..', '..', '.cache'))
+const FISH_AUDIO_PROXY_RE = /^\/fish-audio-api/
 
 export default defineConfig({
   optimizeDeps: {
@@ -83,6 +84,18 @@ export default defineConfig({
       // See: https://vite.dev/config/server-options#server-fs-strict
       strict: false,
     },
+    // NOTICE: Fish Audio's API is server-to-server only and doesn't send
+    // Access-Control-Allow-Origin headers for browser origins, so direct
+    // fetch() calls from the browser are blocked by CORS. We proxy them
+    // through the local Vite dev server so they appear same-origin.
+    // See packages/stage-ui/src/stores/providers.ts for the matching client-side usage.
+    proxy: {
+      '/fish-audio-api': {
+        target: 'https://api.fish.audio',
+        changeOrigin: true,
+        rewrite: (path: string) => path.replace(FISH_AUDIO_PROXY_RE, ''),
+      },
+    },
     warmup: {
       clientFiles: [
         `${resolve(join(import.meta.dirname, '..', '..', 'packages', 'stage-ui', 'src'))}/*.vue`,
diff --git a/apps/stage-tamagotchi/src/main/services/electron/fish-audio.ts b/apps/stage-tamagotchi/src/main/services/electron/fish-audio.ts
new file mode 100644
index 0000000000..1cb72d1013
--- /dev/null
+++ b/apps/stage-tamagotchi/src/main/services/electron/fish-audio.ts
@@ -0,0 +1,43 @@
+import type { createContext } from '@moeru/eventa/adapters/electron/main'
+
+import { defineInvokeHandler } from '@moeru/eventa'
+import { electronFishAudioTTS } from '@proj-airi/stage-ui/stores/providers/fish-audio/ipc'
+import { net } from 'electron'
+
+/**
+ * Register the Fish Audio TTS IPC handler.
+ *
+ * The renderer cannot call https://api.fish.audio directly because the CDN
+ * does not send CORS headers for browser origins. By routing the request
+ * through the main process with net.fetch, we bypass the renderer's CORS
+ * enforcement entirely.
+ */
+export function createFishAudioService(params: {
+  context: ReturnType<typeof createContext>['context']
+}): void {
+  defineInvokeHandler(params.context, electronFishAudioTTS, async (payload) => {
+    const response = await net.fetch(`${payload.baseUrl}/v1/tts`, {
+      method: 'POST',
+      headers: {
+        'Authorization': `Bearer ${payload.apiKey}`,
+        'Content-Type': 'application/json',
+      },
+      body: JSON.stringify({
+        text: payload.text,
+        reference_id: payload.referenceId ?? null,
+        format: 'mp3',
+      }),
+    })
+
+    if (!response.ok) {
+      throw new Error(`Fish Audio TTS failed: ${response.status} ${response.statusText}`)
+    }
+
+    const buffer = await response.arrayBuffer()
+    return {
+      data: new Uint8Array(buffer),
+      contentType: response.headers.get('content-type') ?? 'audio/mpeg',
+      status: response.status,
+    }
+  })
+}
diff --git a/apps/stage-tamagotchi/src/main/services/electron/index.ts b/apps/stage-tamagotchi/src/main/services/electron/index.ts
index 6b58098a74..b3a244e342 100644
--- a/apps/stage-tamagotchi/src/main/services/electron/index.ts
+++ b/apps/stage-tamagotchi/src/main/services/electron/index.ts
@@ -1,5 +1,6 @@
 export * from './app'
 export * from './auto-updater'
+export * from './fish-audio'
 export * from './powerMonitor'
 export * from './screen'
 export * from './window'
diff --git a/apps/stage-tamagotchi/src/main/windows/shared/window.ts b/apps/stage-tamagotchi/src/main/windows/shared/window.ts
index e8b92058a8..aa20485497 100644
--- a/apps/stage-tamagotchi/src/main/windows/shared/window.ts
+++ b/apps/stage-tamagotchi/src/main/windows/shared/window.ts
@@ -10,7 +10,7 @@ import { isMacOS } from 'std-env'
 
 import { createServerChannelService } from '../../services/airi/channel-server'
 import { createI18nService } from '../../services/airi/i18n'
-import { createAppService, createPowerMonitorService, createScreenService, createWindowService } from '../../services/electron'
+import { createAppService, createFishAudioService, createPowerMonitorService, createScreenService, createWindowService } from '../../services/electron'
 
 export function toggleWindowShow(window?: BrowserWindow | null): void {
   if (!window) {
@@ -100,6 +100,7 @@ export async function setupBaseWindowElectronInvokes(params: {
   createWindowService({ context: params.context, window: params.window })
   createAppService({ context: params.context, window: params.window })
   createPowerMonitorService({ context: params.context, window: params.window })
+  createFishAudioService({ context: params.context })
 
   await createI18nService({ context: params.context, window: params.window, i18n: params.i18n })
 
diff --git a/apps/stage-tamagotchi/src/shared/eventa.ts b/apps/stage-tamagotchi/src/shared/eventa.ts
index 97a4668dd2..f47ac127e5 100644
--- a/apps/stage-tamagotchi/src/shared/eventa.ts
+++ b/apps/stage-tamagotchi/src/shared/eventa.ts
@@ -283,6 +283,9 @@ export interface ElectronAuthTokens {
   expiresIn: number
 }
 export const electronAuthStartLogin = defineInvokeEventa<void>('eventa:invoke:electron:auth:start-login')
+
+export { electron } from '@proj-airi/electron-eventa'
+export * from '@proj-airi/electron-eventa/electron-updater'
 export const electronAuthCallback = defineEventa<ElectronAuthTokens>('eventa:event:electron:auth:callback')
 export const electronAuthCallbackError = defineEventa<{ error: string }>('eventa:event:electron:auth:callback-error')
 export const electronAuthLogout = defineInvokeEventa<void>('eventa:invoke:electron:auth:logout')
@@ -290,5 +293,7 @@ export const electronAuthLogout = defineInvokeEventa<void>('eventa:invoke:electr
 export const i18nSetLocale = defineInvokeEventa<void, Locale>('eventa:invoke:electron:i18n:set-locale')
 export const i18nGetLocale = defineInvokeEventa<Locale>('eventa:invoke:electron:i18n:get-locale')
 
-export { electron } from '@proj-airi/electron-eventa'
-export * from '@proj-airi/electron-eventa/electron-updater'
+// Fish Audio TTS IPC contract lives in packages/stage-ui so it can be imported
+// by the renderer without creating an app → package dependency inversion.
+export { electronFishAudioTTS } from '@proj-airi/stage-ui/stores/providers/fish-audio/ipc'
+export type { ElectronFishAudioTTSPayload, ElectronFishAudioTTSResult } from '@proj-airi/stage-ui/stores/providers/fish-audio/ipc'
diff --git a/apps/stage-web/vite.config.ts b/apps/stage-web/vite.config.ts
index 9ef172faad..5836b789fe 100644
--- a/apps/stage-web/vite.config.ts
+++ b/apps/stage-web/vite.config.ts
@@ -25,6 +25,7 @@ import { VitePWA } from 'vite-plugin-pwa'
 
 const stageUIAssetsRoot = resolve(join(import.meta.dirname, '..', '..', 'packages', 'stage-ui', 'src', 'assets'))
 const sharedCacheDir = resolve(join(import.meta.dirname, '..', '..', '.cache'))
+const FISH_AUDIO_PROXY_RE = /^\/fish-audio-api/
 
 function hasFlagEnableMkcert(): boolean {
   if (process.argv.includes('--mkcert')) {
@@ -83,6 +84,18 @@ export default defineConfig({
       // See: https://vite.dev/config/server-options#server-fs-strict
       strict: false,
     },
+    // NOTICE: Fish Audio's API is server-to-server only and doesn't send
+    // Access-Control-Allow-Origin headers for browser origins, so direct
+    // fetch() calls from the browser are blocked by CORS. We proxy them
+    // through the local Vite dev server so they appear same-origin.
+    // See packages/stage-ui/src/stores/providers.ts for the matching client-side usage.
+    proxy: {
+      '/fish-audio-api': {
+        target: 'https://api.fish.audio',
+        changeOrigin: true,
+        rewrite: (path: string) => path.replace(FISH_AUDIO_PROXY_RE, ''),
+      },
+    },
     warmup: {
       clientFiles: [
         `${resolve(join(import.meta.dirname, '..', '..', 'packages', 'stage-ui', 'src'))}/*.vue`,
diff --git a/packages/i18n/src/locales/en/settings.yaml b/packages/i18n/src/locales/en/settings.yaml
index d847c34ac8..2758990a45 100644
--- a/packages/i18n/src/locales/en/settings.yaml
+++ b/packages/i18n/src/locales/en/settings.yaml
@@ -827,6 +827,9 @@ pages:
           default-text: Hello! This is a test of the Kokoro text-to-speech system.
           title: Voice Playground
         title: Kokoro TTS (Local)
+      fish-audio:
+        description: fish.audio
+        title: Fish Audio
       fireworks:
         description: fireworks.ai
         title: Fireworks.ai
diff --git a/packages/i18n/src/locales/es/settings.yaml b/packages/i18n/src/locales/es/settings.yaml
index 3b375ea0ee..de66a3217c 100644
--- a/packages/i18n/src/locales/es/settings.yaml
+++ b/packages/i18n/src/locales/es/settings.yaml
@@ -774,6 +774,9 @@ pages:
           default-text: '¡Hola! Esta es una prueba del sistema de texto a voz de Kokoro.'
           title: Campo de voz
         title: TTS de Kokoro (local)
+      fish-audio:
+        description: fish.audio
+        title: Fish Audio
       fireworks:
         description: fireworks.ai
         title: Fireworks.ai
diff --git a/packages/i18n/src/locales/fr/settings.yaml b/packages/i18n/src/locales/fr/settings.yaml
index a43b7b703e..11caaea59b 100644
--- a/packages/i18n/src/locales/fr/settings.yaml
+++ b/packages/i18n/src/locales/fr/settings.yaml
@@ -774,6 +774,9 @@ pages:
           default-text: Bonjour ! Ceci est un test du système de synthèse vocale Kokoro.
           title: Zone d'essai vocale
         title: Kokoro TTS (Local)
+      fish-audio:
+        description: fish.audio
+        title: Fish Audio
       fireworks:
         description: fireworks.ai
         title: Fireworks.ai
diff --git a/packages/i18n/src/locales/ja/settings.yaml b/packages/i18n/src/locales/ja/settings.yaml
index 98eacfc09b..7839a859af 100644
--- a/packages/i18n/src/locales/ja/settings.yaml
+++ b/packages/i18n/src/locales/ja/settings.yaml
@@ -774,6 +774,9 @@ pages:
           default-text: こんにちは！これは音声合成システム Kokoro のテストです。
           title: 音声合成実験場
         title: Kokoro TTS (ローカル)
+      fish-audio:
+        description: fish.audio
+        title: Fish Audio
       fireworks:
         description: fireworks.ai
         title: Fireworks.ai
diff --git a/packages/i18n/src/locales/ko/settings.yaml b/packages/i18n/src/locales/ko/settings.yaml
index 87a0c023e3..a3ba0906d6 100644
--- a/packages/i18n/src/locales/ko/settings.yaml
+++ b/packages/i18n/src/locales/ko/settings.yaml
@@ -774,6 +774,9 @@ pages:
           default-text: 안녕하세요! 이것은 Kokoro 문자 음성 변환 시스템의 테스트입니다.
           title: 음성 플레이그라운드
         title: Kokoro TTS (로컬)
+      fish-audio:
+        description: fish.audio
+        title: Fish Audio
       fireworks:
         description: fireworks.ai
         title: Fireworks.ai
diff --git a/packages/i18n/src/locales/ru/settings.yaml b/packages/i18n/src/locales/ru/settings.yaml
index 973ecec9f4..9d665bc17f 100644
--- a/packages/i18n/src/locales/ru/settings.yaml
+++ b/packages/i18n/src/locales/ru/settings.yaml
@@ -774,6 +774,9 @@ pages:
           default-text: Привет! Это тест-системы синтеза речи.
           title: Голосовая платформа
         title: Kokoro TTS (Local)
+      fish-audio:
+        description: fish.audio
+        title: Fish Audio
       fireworks:
         description: Fireworks.ai
         title: Fireworks.ai
diff --git a/packages/i18n/src/locales/vi/settings.yaml b/packages/i18n/src/locales/vi/settings.yaml
index fdc131f002..6885a50fa2 100644
--- a/packages/i18n/src/locales/vi/settings.yaml
+++ b/packages/i18n/src/locales/vi/settings.yaml
@@ -774,6 +774,9 @@ pages:
           default-text: Xin chào! Đây là bản thử nghiệm hệ thống giọng nói Kokoro.
           title: Thử nghiệm giọng nói
         title: Kokoro TTS (Cục bộ)
+      fish-audio:
+        description: fish.audio
+        title: Fish Audio
       fireworks:
         description: fireworks.ai
         title: Fireworks.ai
diff --git a/packages/i18n/src/locales/zh-Hans/settings.yaml b/packages/i18n/src/locales/zh-Hans/settings.yaml
index 40dc75dd95..112cc3e6f9 100644
--- a/packages/i18n/src/locales/zh-Hans/settings.yaml
+++ b/packages/i18n/src/locales/zh-Hans/settings.yaml
@@ -776,6 +776,9 @@ pages:
           default-text: 您好！这是 Kokoro 文本转语音（TTS）系统的测试。
           title: 实验平台
         title: Kokoro TTS (本地)
+      fish-audio:
+        description: fish.audio
+        title: Fish Audio
       fireworks:
         description: Fireworks.ai
         title: Fireworks.ai
diff --git a/packages/i18n/src/locales/zh-Hant/settings.yaml b/packages/i18n/src/locales/zh-Hant/settings.yaml
index 0c430d72df..15599339c9 100644
--- a/packages/i18n/src/locales/zh-Hant/settings.yaml
+++ b/packages/i18n/src/locales/zh-Hant/settings.yaml
@@ -774,6 +774,9 @@ pages:
           default-text: 您好！這是 Kokoro 文字轉語音系統的測試。
           title: 語音測試場
         title: Kokoro TTS (本地)
+      fish-audio:
+        description: fish.audio
+        title: Fish Audio
       fireworks:
         description: Fireworks.ai
         title: Fireworks.ai
diff --git a/packages/stage-pages/src/pages/settings/providers/speech/fish-audio.vue b/packages/stage-pages/src/pages/settings/providers/speech/fish-audio.vue
new file mode 100644
index 0000000000..9151358163
--- /dev/null
+++ b/packages/stage-pages/src/pages/settings/providers/speech/fish-audio.vue
@@ -0,0 +1,84 @@
+<script setup lang="ts">
+import type { SpeechProviderWithExtraOptions } from '@xsai-ext/providers/utils'
+
+import {
+  SpeechPlayground,
+  SpeechProviderSettings,
+} from '@proj-airi/stage-ui/components'
+import { useSpeechStore } from '@proj-airi/stage-ui/stores/modules/speech'
+import { useProvidersStore } from '@proj-airi/stage-ui/stores/providers'
+import { useDebounceFn } from '@vueuse/core'
+import { storeToRefs } from 'pinia'
+import { computed, onMounted, watch } from 'vue'
+
+const providerId = 'fish-audio'
+const defaultModel = 's2-pro'
+
+const speechStore = useSpeechStore()
+const providersStore = useProvidersStore()
+const { providers } = storeToRefs(providersStore)
+
+const apiKeyConfigured = computed(() => !!providers.value[providerId]?.apiKey)
+
+const availableVoices = computed(() => speechStore.availableVoices[providerId] || [])
+
+async function handleGenerateSpeech(input: string, voiceId: string, _useSSML: boolean) {
+  const provider = await providersStore.getProviderInstance(providerId) as SpeechProviderWithExtraOptions<string>
+  if (!provider) {
+    throw new Error('Failed to initialize speech provider')
+  }
+
+  const providerConfig = providersStore.getProviderConfig(providerId)
+  const model = providerConfig.model as string | undefined || defaultModel
+
+  return await speechStore.speech(
+    provider,
+    model,
+    input,
+    voiceId,
+    { ...providerConfig },
+  )
+}
+
+async function tryLoadVoices() {
+  if (apiKeyConfigured.value) {
+    await speechStore.loadVoicesForProvider(providerId)
+  }
+}
+
+// Debounced so rapid keystrokes while editing API key / base URL don't fire
+// repeated requests with partial/invalid credentials.
+const debouncedLoadVoices = useDebounceFn(tryLoadVoices, 800)
+
+onMounted(tryLoadVoices)
+
+// Reload voices whenever the API key or base URL changes
+watch(
+  () => [providers.value[providerId]?.apiKey, providers.value[providerId]?.baseUrl],
+  debouncedLoadVoices,
+)
+</script>
+
+<template>
+  <SpeechProviderSettings
+    :provider-id="providerId"
+    :default-model="defaultModel"
+  >
+    <template #playground>
+      <SpeechPlayground
+        :available-voices="availableVoices"
+        :generate-speech="handleGenerateSpeech"
+        :api-key-configured="apiKeyConfigured"
+        :voices-loading="speechStore.isLoadingSpeechProviderVoices"
+        default-text="Hello! This is a test of the Fish Audio voice synthesis."
+      />
+    </template>
+  </SpeechProviderSettings>
+</template>
+
+<route lang="yaml">
+  meta:
+    layout: settings
+    stageTransition:
+      name: slide
+  </route>
diff --git a/packages/stage-ui/src/components/scenarios/providers/speech-provider-settings.vue b/packages/stage-ui/src/components/scenarios/providers/speech-provider-settings.vue
index d679c4dbd7..2eb8f9f6c1 100644
--- a/packages/stage-ui/src/components/scenarios/providers/speech-provider-settings.vue
+++ b/packages/stage-ui/src/components/scenarios/providers/speech-provider-settings.vue
@@ -33,6 +33,7 @@ defineSlots<{
   'advanced-settings': (props: any) => any
   'playground': (props: any) => any
 }>()
+
 const { t } = useI18n()
 const router = useRouter()
 const providersStore = useProvidersStore()
diff --git a/packages/stage-ui/src/components/scenes/Stage.vue b/packages/stage-ui/src/components/scenes/Stage.vue
index 0fb3e65de8..41c7b4189e 100644
--- a/packages/stage-ui/src/components/scenes/Stage.vue
+++ b/packages/stage-ui/src/components/scenes/Stage.vue
@@ -314,7 +314,13 @@ const speechPipeline = createSpeechPipeline<AudioBuffer>({
       const audioBuffer = await audioContext.decodeAudioData(res)
       return audioBuffer
     }
-    catch {
+    catch (error) {
+      console.error('[Speech Pipeline] TTS failed:', {
+        provider: activeSpeechProvider.value,
+        model: activeSpeechModel.value,
+        voice: activeSpeechVoice.value?.id,
+        error,
+      })
       return null
     }
   },
diff --git a/packages/stage-ui/src/stores/providers.ts b/packages/stage-ui/src/stores/providers.ts
index 52012fcb62..795ef04d75 100644
--- a/packages/stage-ui/src/stores/providers.ts
+++ b/packages/stage-ui/src/stores/providers.ts
@@ -20,7 +20,7 @@ import type {
 
 import type { AliyunRealtimeSpeechExtraOptions } from './providers/aliyun/stream-transcription'
 
-import { isStageTamagotchi, isUrl } from '@proj-airi/stage-shared'
+import { isElectronWindow, isStageCapacitor, isStageTamagotchi, isStageWeb, isUrl } from '@proj-airi/stage-shared'
 import { computedAsync, useIntervalFn, useLocalStorage } from '@vueuse/core'
 import {
   createOpenAI,
@@ -58,6 +58,8 @@ import { buildOpenAICompatibleProvider } from './providers/openai-compatible-bui
 import { buildOpenRouterAudioSpeechProvider } from './providers/openrouter/audio-speech'
 import { createWebSpeechAPIProvider } from './providers/web-speech-api'
 
+const TRAILING_SLASH_RE = /\/$/
+
 const ALIYUN_NLS_REGIONS = [
   'cn-shanghai',
   'cn-shanghai-internal',
@@ -1540,6 +1542,182 @@ export const useProvidersStore = defineStore('providers', () => {
         },
       },
     },
+    'fish-audio': {
+      id: 'fish-audio',
+      category: 'speech',
+      tasks: ['text-to-speech'],
+      nameKey: 'settings.pages.providers.provider.fish-audio.title',
+      name: 'Fish Audio',
+      descriptionKey: 'settings.pages.providers.provider.fish-audio.description',
+      description: 'fish.audio',
+      icon: 'i-lobe-icons:fishaudio',
+      defaultOptions: () => ({
+        baseUrl: 'https://api.fish.audio',
+      }),
+      createProvider: async (config) => {
+        const apiKey = (config.apiKey as string ?? '').trim()
+        const baseUrl = ((config.baseUrl as string) || 'https://api.fish.audio').replace(TRAILING_SLASH_RE, '')
+        // NOTICE: Fish Audio's API is server-to-server only and does not send CORS
+        // headers for browser origins. Depending on the runtime, we choose one of
+        // three strategies to avoid CORS failures:
+        //
+        //   1. Electron — delegate to the main process via Eventa IPC. The main
+        //      process uses net.fetch which runs in Node and is never subject to
+        //      browser CORS enforcement. The IPC contract is defined in
+        //      apps/stage-tamagotchi/src/shared/eventa.ts (electronFishAudioTTS).
+        //
+        //   2. Browser DEV (stage-web / stage-pocket) — route through the local
+        //      Vite dev-server proxy (/fish-audio-api → https://api.fish.audio) so
+        //      the request appears same-origin. See vite.config.ts in each app for
+        //      the matching server.proxy entry.
+        //
+        //   3. Everything else (custom base URL, Capacitor native) — use the URL
+        //      as-is. Capacitor native builds use the OS HTTP stack and are not
+        //      subject to browser CORS. A custom base URL means the user has their
+        //      own same-origin proxy and needs no rewrite.
+        const electronWindow = isStageTamagotchi() ? isElectronWindow(window) && window : false
+        const effectiveBase = (!electronWindow && import.meta.env.DEV && baseUrl === 'https://api.fish.audio')
+          ? '/fish-audio-api'
+          : baseUrl
+        const provider: SpeechProvider = {
+          speech: (model: string) => ({
+            // NOTICE: baseURL must be an absolute URL — @xsai/generate-speech calls
+            // `new URL('audio/speech', baseURL)` internally. Our custom fetch below
+            // ignores the URL argument entirely and builds its own, so the value here
+            // only needs to be valid; we always keep the original absolute baseUrl.
+            baseURL: `${baseUrl}/`,
+            model,
+            // NOTICE: Fish Audio does not use the OpenAI /audio/speech endpoint format.
+            // We intercept the xsai generateSpeech request and translate it to
+            // Fish Audio's POST /v1/tts format (text/reference_id in body).
+            //
+            // NOTICE: The Fish Audio API accepts the model via a `model` request header,
+            // but sending custom headers from the browser triggers a CORS preflight that
+            // Fish Audio's CDN does not allow (Access-Control-Allow-Headers does not
+            // include `model`). The Fish Audio JS/Python SDKs also never send `model` as
+            // a header — they rely on the server default (`s2-pro`). We do the same here.
+            fetch: async (_url: RequestInfo | URL, init?: RequestInit) => {
+              const body = JSON.parse((init?.body as string) ?? '{}') as {
+                input?: string
+                voice?: string
+              }
+              if (electronWindow) {
+                // Delegate to main process to bypass renderer CORS.
+                // defineInvoke returns a typed invoker function; call it with the payload.
+                const { defineInvoke } = await import('@moeru/eventa')
+                const { createContext } = await import('@moeru/eventa/adapters/electron/renderer')
+                const { electronFishAudioTTS } = await import('./providers/fish-audio/ipc')
+                const { context } = createContext(electronWindow.electron.ipcRenderer)
+                // NOTICE: AbortSignal is not serializable over IPC and cannot be
+                // forwarded to the main process. Cancellation of in-flight IPC
+                // requests is not yet supported for this path.
+                const invokeTTS = defineInvoke(context, electronFishAudioTTS)
+                const result = await invokeTTS({ apiKey, baseUrl, text: body.input ?? '', referenceId: body.voice || null })
+                return new Response(result.data.buffer as ArrayBuffer, {
+                  status: result.status,
+                  headers: { 'Content-Type': result.contentType },
+                })
+              }
+              return fetch(`${effectiveBase}/v1/tts`, {
+                method: 'POST',
+                // Forward the AbortSignal so the HTTP request is cancelled when
+                // the TTS pipeline is aborted (e.g. user interrupts playback).
+                signal: init?.signal,
+                headers: {
+                  'Authorization': `Bearer ${apiKey}`,
+                  'Content-Type': 'application/json',
+                },
+                body: JSON.stringify({
+                  text: body.input ?? '',
+                  reference_id: body.voice || null,
+                  format: 'mp3',
+                }),
+              })
+            },
+          }),
+        }
+        return provider
+      },
+      capabilities: {
+        // NOTICE: The Fish Audio API selects the model via a `model` HTTP header.
+        // Sending that custom header from a browser causes CORS preflight failures,
+        // so we cannot forward model selection to the API. The server defaults to
+        // s2-pro when no header is present. Listing models here is informational only.
+        listModels: async () => [
+          {
+            id: 's2-pro',
+            name: 'S2 Pro',
+            provider: 'fish-audio',
+            description: 'Latest generation model (server default)',
+            contextLength: 0,
+            deprecated: false,
+          },
+        ],
+        listVoices: async (config) => {
+          const { listFishAudioVoices } = await import('./providers/fish-audio/list-voices')
+          const rawBase = ((config.baseUrl as string) || 'https://api.fish.audio').replace(TRAILING_SLASH_RE, '')
+          const isElectronRenderer = typeof navigator !== 'undefined' && navigator.userAgent.includes('Electron')
+          const effectiveVoiceBase = (import.meta.env.DEV && !isElectronRenderer && rawBase === 'https://api.fish.audio')
+            ? '/fish-audio-api'
+            : rawBase
+          return listFishAudioVoices(
+            ((config.apiKey as string) ?? '').trim(),
+            effectiveVoiceBase,
+          )
+        },
+      },
+      validators: {
+        chatPingCheckAvailable: false,
+        validateProviderConfig: (config) => {
+          const errors: Error[] = []
+
+          if (!config.apiKey) {
+            errors.push(new Error('API key is required.'))
+          }
+
+          // NOTICE: We intentionally skip the shared baseUrlValidator here because
+          // it requires a trailing slash for standard URL composition. Fish Audio's
+          // custom fetch override ignores the passed URL and constructs its own, so
+          // trailing-slash strictness is irrelevant. We only verify it's a valid
+          // absolute URL when one is explicitly provided.
+          if (config.baseUrl) {
+            try {
+              const parsed = new URL(config.baseUrl as string)
+              if (!parsed.host) {
+                errors.push(new Error('Base URL must have a valid host.'))
+              }
+            }
+            catch {
+              errors.push(new Error('Base URL is not a valid absolute URL.'))
+            }
+          }
+
+          // Gate the default Fish Audio URL in contexts where browser CORS will
+          // block the request and no proxy/IPC path exists.
+          //
+          // - Electron: handled via IPC (electronFishAudioTTS) — no gate needed.
+          // - Capacitor native (iOS/Android): OS HTTP stack bypasses CORS — no gate.
+          // - Web/Capacitor production: Vite dev-server proxy is absent; users must
+          //   point baseUrl at a same-origin CORS proxy they control.
+          // Normalize before comparison: strip trailing slashes the same way
+          // createProvider does, so 'https://api.fish.audio/' is caught too.
+          const normalizedConfigUrl = ((config.baseUrl as string) || '').replace(TRAILING_SLASH_RE, '')
+          const usingDefaultUrl = !config.baseUrl || normalizedConfigUrl === 'https://api.fish.audio'
+          if (usingDefaultUrl && (isStageWeb() || isStageCapacitor()) && !import.meta.env.DEV) {
+            errors.push(new Error(
+              'Fish Audio requires a custom base URL (a CORS proxy) in production builds. '
+              + 'The dev-server proxy is only available during local development.',
+            ))
+          }
+
+          return {
+            errors,
+            reason: errors.map(e => e.message).join(', '),
+            valid: errors.length === 0,
+          }
+        },
+      },
+    },
     'kokoro-local': {
       id: 'kokoro-local',
       category: 'speech',
diff --git a/packages/stage-ui/src/stores/providers/fish-audio/ipc.ts b/packages/stage-ui/src/stores/providers/fish-audio/ipc.ts
new file mode 100644
index 0000000000..8a94de43a1
--- /dev/null
+++ b/packages/stage-ui/src/stores/providers/fish-audio/ipc.ts
@@ -0,0 +1,32 @@
+import { defineInvokeEventa } from '@moeru/eventa'
+
+/**
+ * Payload sent from the Electron renderer to the main process for a Fish Audio
+ * TTS request. AbortSignal is intentionally omitted — it is not serializable
+ * across IPC and cancellation must be handled at a higher level if needed.
+ */
+export interface ElectronFishAudioTTSPayload {
+  apiKey: string
+  baseUrl: string
+  text: string
+  referenceId?: string | null
+}
+
+/** Response returned by the main process after proxying the Fish Audio request. */
+export interface ElectronFishAudioTTSResult {
+  /** Raw audio bytes as a structured-clone-safe Uint8Array. */
+  data: Uint8Array
+  contentType: string
+  status: number
+}
+
+/**
+ * Eventa IPC contract for proxying Fish Audio TTS through the Electron main
+ * process. The main process uses net.fetch which runs in Node and is therefore
+ * not subject to browser CORS enforcement.
+ *
+ * Handler: apps/stage-tamagotchi/src/main/services/electron/fish-audio.ts
+ */
+export const electronFishAudioTTS = defineInvokeEventa<ElectronFishAudioTTSResult, ElectronFishAudioTTSPayload>(
+  'eventa:invoke:electron:providers:fish-audio:tts',
+)
diff --git a/packages/stage-ui/src/stores/providers/fish-audio/list-voices.ts b/packages/stage-ui/src/stores/providers/fish-audio/list-voices.ts
new file mode 100644
index 0000000000..85a76aa438
--- /dev/null
+++ b/packages/stage-ui/src/stores/providers/fish-audio/list-voices.ts
@@ -0,0 +1,66 @@
+import type { VoiceInfo } from '../../providers'
+
+export interface FishAudioModelEntity {
+  _id: string
+  type: 'svc' | 'tts'
+  title: string
+  description: string
+  state: 'created' | 'training' | 'trained' | 'failed'
+  samples: Array<{ audio: string }>
+  languages: string[]
+}
+
+interface FishAudioModelListResponse {
+  items: FishAudioModelEntity[]
+}
+
+const TRAILING_SLASH_RE = /\/$/
+
+function mapModelToVoiceInfo(model: FishAudioModelEntity): VoiceInfo {
+  return {
+    id: model._id,
+    name: model.title,
+    provider: 'fish-audio',
+    description: model.description || undefined,
+    previewURL: model.samples?.[0]?.audio || undefined,
+    languages: model.languages.length > 0
+      ? model.languages.map(lang => ({ code: lang, title: lang }))
+      : [{ code: 'en', title: 'English' }],
+  }
+}
+
+async function fetchModels(url: string, headers: HeadersInit): Promise<FishAudioModelEntity[]> {
+  const response = await fetch(url, { headers })
+  if (!response.ok) {
+    throw new Error(`Failed to fetch Fish Audio voices: ${response.status} ${response.statusText}`)
+  }
+  const data = await response.json() as FishAudioModelListResponse
+  return data.items.filter(model => model.state === 'trained' && model.type === 'tts')
+}
+
+/**
+ * Fetches voice models from Fish Audio and maps them to VoiceInfo.
+ *
+ * Fetches both the user's own models (GET /model?self=true) and the top
+ * community/public models (GET /model?sort_by=task_count), merges them
+ * deduped by ID, with own models listed first.
+ */
+export async function listFishAudioVoices(apiKey: string, baseUrl: string): Promise<VoiceInfo[]> {
+  const base = baseUrl.replace(TRAILING_SLASH_RE, '')
+  const headers: HeadersInit = { Authorization: `Bearer ${apiKey}` }
+
+  // Fetch own models and popular public models in parallel
+  const [ownModels, publicModels] = await Promise.all([
+    fetchModels(`${base}/model?self=true&page_size=100`, headers),
+    fetchModels(`${base}/model?sort_by=task_count&page_size=100`, headers),
+  ])
+
+  // Merge: own models first, then public models not already in the own list
+  const ownIds = new Set(ownModels.map(m => m._id))
+  const merged = [
+    ...ownModels,
+    ...publicModels.filter(m => !ownIds.has(m._id)),
+  ]
+
+  return merged.map(mapModelToVoiceInfo)
+}