Skip to content

armgit / cindy

Public Imported from GitHub
0

Consider it done. The open-source AI agent that works out of the box · 想到,就能做到。开源、开箱即用的 AI Agent。

fix(providers): scope speech recommendations to endpoint region

JiangJiahao31ef395d6f87
diff --git a/apps/desktop/scripts/fixtures/provider-speech-setup/provider-speech-setup-fixture.tsx b/apps/desktop/scripts/fixtures/provider-speech-setup/provider-speech-setup-fixture.tsx
index b4933e2f00..43ec949399 100644
--- a/apps/desktop/scripts/fixtures/provider-speech-setup/provider-speech-setup-fixture.tsx
+++ b/apps/desktop/scripts/fixtures/provider-speech-setup/provider-speech-setup-fixture.tsx
@@ -114,60 +114,76 @@ function Fixture({ scenario }: { scenario: string }) {
);
}
void mountPage({
- host: (ctx) => ({
- safeStorageRead: async () => 'fixture-key',
- maker: {
- listProviderPresets: async () => ({
- presets: BUNDLED_CATALOG.presets!.filter((p) => p.id === 'minimax-global'),
- }),
- localModelList: async () => ({
- status: { runtime: 'ollama', kind: 'absent', appInstalled: false },
- models: [],
- }),
- scanLocalCli: async () => ({ detections: [] }),
- llamaCppStatus: async () => ({
- installed: false,
- supported: true,
- running: false,
- models: [],
- }),
- onProviderOAuthProgress: subscription('oauthProgress'),
- onProvidersChanged: subscription('providersChanged'),
- fetchProviderModels: async () =>
- ctx.case === 'elevenlabs'
- ? {
- ok: true,
- models: [
- {
- id: 'eleven_flash_v2_5',
- name: 'Eleven Flash v2.5',
- discoveredMetadata: { mode: 'audio_speech' },
- },
- ],
- }
- : ctx.case === 'bailian'
+ host: (ctx) => {
+ if (ctx.case === 'bailian-existing')
+ stored = {
+ id: 'bailian-workspace',
+ name: 'Alibaba Cloud · Workspace',
+ runtimes: {
+ codex: {
+ baseUrl: 'https://ws-test.cn-beijing.maas.aliyuncs.com/api/v1',
+ wireProtocol: 'openai-chat',
+ models: [{ id: 'qwen-plus', name: 'Qwen Plus' }],
+ },
+ },
+ };
+ return {
+ safeStorageRead: async () => 'fixture-key',
+ maker: {
+ listProviderPresets: async () => ({
+ presets: BUNDLED_CATALOG.presets!.filter((p) => p.id === 'minimax-global'),
+ }),
+ localModelList: async () => ({
+ status: { runtime: 'ollama', kind: 'absent', appInstalled: false },
+ models: [],
+ }),
+ scanLocalCli: async () => ({ detections: [] }),
+ llamaCppStatus: async () => ({
+ installed: false,
+ supported: true,
+ running: false,
+ models: [],
+ }),
+ onProviderOAuthProgress: subscription('oauthProgress'),
+ onProvidersChanged: subscription('providersChanged'),
+ fetchProviderModels: async () =>
+ ctx.case === 'elevenlabs'
? {
ok: true,
- models: Array.from({ length: 100 }, (_, index) => ({
- id: `chat-${index}`, name: `Chat ${index}`, mode: 'chat',
- })),
+ models: [
+ {
+ id: 'eleven_flash_v2_5',
+ name: 'Eleven Flash v2.5',
+ discoveredMetadata: { mode: 'audio_speech' },
+ },
+ ],
}
- : { ok: false, code: 'NETWORK' },
- createCustomProvider: async (config: CustomProviderConfig) => {
- stored = config;
- return { ok: true };
- },
- updateCustomProvider: async (
- config: CustomProviderConfig,
- _keys: unknown,
- options: { appendSpeechModelsOnly?: true },
- ) => {
- if (!options.appendSpeechModelsOnly) throw new Error('Expected additions only');
- stored = appendSpeechModels(stored, config);
- return { ok: true };
+ : ctx.case === 'bailian'
+ ? {
+ ok: true,
+ models: Array.from({ length: 100 }, (_, index) => ({
+ id: `chat-${index}`,
+ name: `Chat ${index}`,
+ mode: 'chat',
+ })),
+ }
+ : { ok: false, code: 'NETWORK' },
+ createCustomProvider: async (config: CustomProviderConfig) => {
+ stored = config;
+ return { ok: true };
+ },
+ updateCustomProvider: async (
+ config: CustomProviderConfig,
+ _keys: unknown,
+ options: { appendSpeechModelsOnly?: true },
+ ) => {
+ if (!options.appendSpeechModelsOnly) throw new Error('Expected additions only');
+ stored = appendSpeechModels(stored, config);
+ return { ok: true };
+ },
+ setModelDisable: forbid('setModelDisable'),
},
- setModelDisable: forbid('setModelDisable'),
- },
- }),
+ };
+ },
render: (ctx) => <Fixture scenario={ctx.case} />,
});
diff --git a/apps/desktop/scripts/fixtures/provider-speech-setup/ui-page.mjs b/apps/desktop/scripts/fixtures/provider-speech-setup/ui-page.mjs
index 12f9c93c4b..b10dd02572 100644
--- a/apps/desktop/scripts/fixtures/provider-speech-setup/ui-page.mjs
+++ b/apps/desktop/scripts/fixtures/provider-speech-setup/ui-page.mjs
@@ -13,14 +13,20 @@ export default {
'Single add-provider flow: MiniMax API key, grouped chat/speech, recommended speech default',
},
elevenlabs: { about: 'ElevenLabs catalog discovery, TTS-only connection' },
- bailian: { about: 'Alibaba Cloud speech recommendations visible above 100 discovered chat models' },
+ bailian: {
+ about: 'Alibaba Cloud speech recommendations visible above 100 discovered chat models',
+ },
+ 'bailian-existing': {
+ about: 'Existing Beijing workspace recommends CosyVoice and preserves its chat model',
+ },
subscription: { about: 'Existing OpenAI subscription creates a separate API-key connection' },
},
async ready(page) {
await page.getByTestId('provider-model-toolbar').waitFor();
},
async check(page, ui) {
- if (['default', 'subscription'].includes(ui.case)) {
+ const existing = ['default', 'bailian-existing'].includes(ui.case);
+ if (existing || ui.case === 'subscription') {
await page
.getByRole('button', {
name: await ui.t('settings.providers.speechSetup.add'),
@@ -39,7 +45,7 @@ export default {
.getByRole('button', { name: await ui.t('settings.providers.wizard.useApiKey') })
.click();
}
- if (ui.case !== 'default') {
+ if (!existing) {
if (ui.case === 'subscription')
await page.getByText(await ui.t('settings.providers.speechSetup.separateKey')).waitFor();
await page.getByPlaceholder('sk-…').fill('fixture-key');
@@ -52,30 +58,33 @@ export default {
ui.case === 'elevenlabs'
? 'eleven_flash_v2_5'
: ui.case === 'bailian'
- ? 'cosyvoice-v3-flash'
- : ui.case === 'subscription'
- ? 'gpt-4o-mini-tts'
- : 'speech-2.8-turbo';
+ ? 'qwen3-tts-flash'
+ : ui.case === 'bailian-existing'
+ ? 'cosyvoice-v3-flash'
+ : ui.case === 'subscription'
+ ? 'gpt-4o-mini-tts'
+ : 'speech-2.8-turbo';
const recommended = page.getByRole('checkbox', { name: new RegExp(recommendedId) });
await recommended.waitFor();
assert.equal(await recommended.getAttribute('aria-checked'), 'true');
if (ui.case === 'bailian') {
// PR #6768: first-screen discoverability, before any test-induced scrolling.
const box = await recommended.boundingBox();
- const height = await page.evaluate(() => window.innerHeight);
+ const height = await page.evaluate(() => globalThis.innerHeight);
assert.ok(box && box.y >= 0 && box.y + box.height < height);
}
await recommended.scrollIntoViewIfNeeded();
await ui.capture('models');
- const finish =
- ui.case === 'default'
- ? 'settings.providers.wizard.addManualModel'
- : 'settings.providers.wizard.finish';
+ const finish = existing
+ ? 'settings.providers.wizard.addManualModel'
+ : 'settings.providers.wizard.finish';
await page.getByRole('button', { name: await ui.t(finish), exact: true }).click();
await page.getByTestId('saved').waitFor();
assert.ok((await page.getByTestId('saved').textContent()).includes(recommendedId));
if (ui.case === 'default')
assert.ok((await page.getByTestId('saved').textContent()).includes('MiniMax-M3'));
+ if (ui.case === 'bailian-existing')
+ assert.ok((await page.getByTestId('saved').textContent()).includes('qwen-plus'));
await ui.capture('saved');
},
};
diff --git a/apps/desktop/src/renderer/__tests__/addProviderWizardPresetEntry.test.tsx b/apps/desktop/src/renderer/__tests__/addProviderWizardPresetEntry.test.tsx
index de593fd83d..5d296eb550 100644
--- a/apps/desktop/src/renderer/__tests__/addProviderWizardPresetEntry.test.tsx
+++ b/apps/desktop/src/renderer/__tests__/addProviderWizardPresetEntry.test.tsx
@@ -1504,11 +1504,11 @@ describe('speech models in the existing provider wizard', () => {
Array.from({ length: 100 }, (_, index) => ({ id: `chat-${index}`, name: `Chat ${index}`, mode: 'chat' })),
});
await enter('aliyun-bailian-api-cn');
- const speech = await screen.findByRole('checkbox', { name: /cosyvoice-v3-flash/ });
+ const speech = await screen.findByRole('checkbox', { name: /qwen3-tts-flash/ });
const firstChat = screen.getByText('Chat 0');
expect(speech.compareDocumentPosition(firstChat) & Node.DOCUMENT_POSITION_FOLLOWING).toBeTruthy();
expect(speech.getAttribute('aria-checked')).toBe('true');
- expect(screen.getByRole('checkbox', { name: /qwen3-tts-flash/ }).getAttribute('aria-checked')).toBe('false');
+ expect(screen.queryByRole('checkbox', { name: /cosyvoice/ })).toBeNull();
});
it('reuses server API presets instead of duplicating the same speech service entry', async () => {
window.electronAPI.maker.listProviderPresets = vi.fn(async () => ({ presets: [{
@@ -1523,7 +1523,7 @@ describe('speech models in the existing provider wizard', () => {
fireEvent.click(serverPreset);
fireEvent.change(await screen.findByPlaceholderText('sk-…'), { target: { value: 'fake-test-key' } });
fireEvent.click(screen.getByText('settings.providers.wizard.next'));
- expect((await screen.findByRole('checkbox', { name: /cosyvoice-v3-flash/ })).getAttribute('aria-checked')).toBe('true');
+ expect((await screen.findByRole('checkbox', { name: /qwen3-tts-flash/ })).getAttribute('aria-checked')).toBe('true');
});
async function enter(presetId: string, speechOnly = false) {
render(<AddProviderWizard providers={[]} entry={{ kind: 'preset', presetId, speechOnly }} onClose={vi.fn()} onDone={vi.fn()} onOpenCustomForm={vi.fn()} />);
@@ -1574,6 +1574,114 @@ describe('speech models in the existing provider wizard', () => {
});
});
+// PR #6768 / #6660 integration regression (Renderer unit): the real wizard and
+// additions dialog must save defaults supported at the configured endpoint.
+// Previous tests asserted one brand-wide CosyVoice default; a public DashScope
+// URL is the smallest failing input. No live credentials or synthesis required.
+describe('Alibaba speech recommendations follow the actual endpoint', () => {
+ const endpoints = [
+ ['https://dashscope.aliyuncs.com/compatible-mode/v1', 'qwen3-tts-flash'],
+ ['https://dashscope-intl.aliyuncs.com/api/v1/', 'qwen3-tts-flash'],
+ ['https://ws-test.cn-beijing.maas.aliyuncs.com/api/v1', 'cosyvoice-v3-flash'],
+ ['https://work-unit.cn-beijing.maas.aliyuncs.com/compatible-mode/v1', 'cosyvoice-v3-flash'],
+ ['https://ws-test.ap-southeast-1.maas.aliyuncs.com/api/v1', 'qwen-audio-3.0-tts-flash'],
+ ['https://dashscope-us.aliyuncs.com/compatible-mode/v1', null],
+ ] as const;
+ const publicUrl = 'https://dashscope-intl.aliyuncs.com/compatible-mode/v1';
+ async function openWizard(baseUrl: string) {
+ renderWizard('aliyun-bailian-api-global');
+ fireEvent.change(await screen.findByDisplayValue(publicUrl), { target: { value: baseUrl } });
+ fireEvent.change(screen.getByPlaceholderText('sk-…'), { target: { value: 'fake-test-key' } });
+ fireEvent.click(screen.getByText('settings.providers.wizard.next'));
+ await waitFor(() => expect(screen.queryByText('settings.providers.wizard.fetching')).toBeNull());
+ }
+ function expectRecommendation(id: string | null) {
+ const section = screen.getByRole('region', { name: 'settings.providers.speechSetup.title' });
+ const checked = within(section).queryAllByRole('checkbox').filter(row => row.getAttribute('aria-checked') === 'true');
+ expect(checked).toHaveLength(id ? 1 : 0);
+ if (id) {
+ expect(checked[0].textContent).toContain(id);
+ expect(within(checked[0]).getByText('settings.providers.wizard.recommended')).toBeTruthy();
+ } else {
+ expect(within(section).queryByText('settings.providers.wizard.recommended')).toBeNull();
+ }
+ }
+ it.each(endpoints)('new provider uses %s instead of the preset region', async (baseUrl, expected) => {
+ await openWizard(baseUrl);
+ expectRecommendation(expected);
+ if (!expected) return;
+ fireEvent.click(screen.getByText('settings.providers.wizard.finish'));
+ await waitFor(() => expect(createCustomProvider).toHaveBeenCalledTimes(1));
+ const runtime = vi.mocked(createCustomProvider).mock.calls[0][0].runtimes.codex!;
+ expect(runtime.baseUrl).toBe(baseUrl);
+ const speech = runtime.models.filter(model => /tts|cosyvoice/.test(model.id));
+ expect(speech.map(model => model.id)).toEqual([expected]);
+ expect(speech[0].defaultEnabled).not.toBe(false);
+ });
+ it.each(endpoints)('existing provider uses the same recommendation for %s', async (baseUrl, expected) => {
+ const provider = { ...buildUserProvider({ id: 'my-ali', name: 'Alibaba connection', runtimes: {
+ codex: { baseUrl, wireProtocol: 'openai-chat', models: [{ id: 'qwen-plus', name: 'My chat', defaultEnabled: false }] },
+ } }), connected: true } as ProviderView;
+ vi.mocked(updateCustomProvider).mockResolvedValue({ ok: true });
+ render(<AddSpeechModelsDialog provider={provider} onClose={vi.fn()} onSaved={vi.fn()} />);
+ await waitFor(() => expect(screen.queryByText('settings.providers.wizard.fetching')).toBeNull());
+ expectRecommendation(expected);
+ if (!expected) return;
+ fireEvent.click(screen.getByText('settings.providers.wizard.addManualModel'));
+ await waitFor(() => expect(updateCustomProvider).toHaveBeenCalledTimes(1));
+ expect(updateCustomProvider).toHaveBeenCalledWith(expect.objectContaining({
+ runtimes: { codex: expect.objectContaining({ baseUrl, models: [expect.objectContaining({ id: expected, defaultEnabled: true })] }) },
+ }), {}, { appendSpeechModelsOnly: true });
+ });
+ it('recomputes recommendations after going back and editing the endpoint', async () => {
+ await openWizard(publicUrl);
+ expectRecommendation('qwen3-tts-flash');
+ fireEvent.click(screen.getByText('settings.providers.wizard.back'));
+ const workspace = 'https://ws-test.cn-beijing.maas.aliyuncs.com/api/v1';
+ fireEvent.change(await screen.findByDisplayValue(publicUrl), { target: { value: workspace } });
+ fireEvent.click(screen.getByText('settings.providers.wizard.next'));
+ await waitFor(() => expect(screen.queryByText('settings.providers.wizard.fetching')).toBeNull());
+ expectRecommendation('cosyvoice-v3-flash');
+ expect(screen.queryByRole('checkbox', { name: /qwen3-tts-flash/ })).toBeNull();
+ });
+ it('keeps US catalog models unselected and lets the user explicitly add one', async () => {
+ vi.mocked(window.electronAPI.maker.fetchProviderModels).mockResolvedValue({ ok: true, models: [
+ { id: 'cosyvoice-v3-flash', name: 'CosyVoice', discoveredMetadata: { mode: 'audio_speech' } },
+ { id: 'qwen3-tts-flash', name: 'Qwen3', discoveredMetadata: { mode: 'audio_speech' } },
+ ] });
+ await openWizard('https://dashscope-us.aliyuncs.com/compatible-mode/v1');
+ expectRecommendation(null);
+ fireEvent.click(screen.getByRole('checkbox', { name: /cosyvoice-v3-flash/ }));
+ fireEvent.click(screen.getByText('settings.providers.wizard.finish'));
+ await waitFor(() => expect(createCustomProvider).toHaveBeenCalledTimes(1));
+ const models = vi.mocked(createCustomProvider).mock.calls[0][0].runtimes.codex!.models;
+ expect(models.find(model => model.id === 'cosyvoice-v3-flash')).toBeDefined();
+ expect(models.find(model => model.id === 'cosyvoice-v3-flash')?.defaultEnabled).not.toBe(false);
+ expect(models.find(model => model.id === 'qwen3-tts-flash')?.defaultEnabled).toBe(false);
+ });
+ it('keeps a previously saved CosyVoice choice when recommending Qwen3 for an existing public connection', async () => {
+ const provider = { ...buildUserProvider({ id: 'my-ali', name: 'Alibaba connection', runtimes: {
+ codex: { baseUrl: publicUrl, wireProtocol: 'openai-chat', models: [
+ { id: 'cosyvoice-v3-flash', name: 'My saved voice', mode: 'audio_speech', defaultEnabled: false },
+ ] },
+ } }), connected: true } as ProviderView;
+ vi.mocked(updateCustomProvider).mockResolvedValue({ ok: true });
+ render(<AddSpeechModelsDialog provider={provider} onClose={vi.fn()} onSaved={vi.fn()} />);
+ const saved = await screen.findByRole('checkbox', { name: /My saved voice/ });
+ expect((saved as HTMLButtonElement).disabled).toBe(true);
+ expect(within(saved).getByText('settings.providers.speechSetup.saved')).toBeTruthy();
+ expect(screen.getByRole('checkbox', { name: /qwen3-tts-flash/ }).getAttribute('aria-checked')).toBe('true');
+ fireEvent.click(screen.getByText('settings.providers.wizard.addManualModel'));
+ await waitFor(() => expect(updateCustomProvider).toHaveBeenCalledTimes(1));
+ expect(vi.mocked(updateCustomProvider).mock.calls[0][0].runtimes.codex?.models).toEqual([
+ expect.objectContaining({ id: 'qwen3-tts-flash', defaultEnabled: true }),
+ ]);
+ expect(provider.models.codex).toEqual([
+ expect.objectContaining({ id: 'cosyvoice-v3-flash', name: 'My saved voice', defaultEnabled: false }),
+ ]);
+ });
+});
+
describe('existing connection speech additions', () => {
const provider = { ...buildUserProvider({ id: 'my-minimax', name: 'My connection', runtimes: {
codex: { baseUrl: 'https://api.minimax.io/v1', wireProtocol: 'openai-chat', models: [
diff --git a/apps/desktop/src/renderer/components/settings/AddProviderWizard.tsx b/apps/desktop/src/renderer/components/settings/AddProviderWizard.tsx
index 579989a443..c019659d40 100644
--- a/apps/desktop/src/renderer/components/settings/AddProviderWizard.tsx
+++ b/apps/desktop/src/renderer/components/settings/AddProviderWizard.tsx
@@ -1032,7 +1032,7 @@ export function AddProviderWizard({
if (entry?.kind === 'preset' && entry.speechOnly) {
for (const value of initial.values()) value.checked = false;
}
- for (const model of speechReferenceModels(speechTarget.service)) {
+ for (const model of speechReferenceModels(presetRuntimeBaseUrl(preset, speechTarget.agent, presetBaseUrls))) {
if (initial.has(model.id)) continue;
initial.set(model.id, { name: model.name, checked: model.defaultEnabled !== false,
recommended: model.defaultEnabled !== false, agents: [speechTarget.agent], speechReference: true });
diff --git a/apps/desktop/src/renderer/components/settings/AddSpeechModelsDialog.tsx b/apps/desktop/src/renderer/components/settings/AddSpeechModelsDialog.tsx
index ad188b87ee..dd83f518ad 100644
--- a/apps/desktop/src/renderer/components/settings/AddSpeechModelsDialog.tsx
+++ b/apps/desktop/src/renderer/components/settings/AddSpeechModelsDialog.tsx
@@ -42,15 +42,17 @@ export function AddSpeechModelsDialog({
(runtime) => runtime?.models.map((model) => [model.id, model] as const) ?? [],
),
);
- const references = speechReferenceModels(target.service).map((model) => ({
- id: model.id,
- name: saved.get(model.id)?.name ?? model.name,
- checked: !saved.has(model.id) && model.defaultEnabled !== false,
- recommended: model.defaultEnabled !== false,
- reference: true,
- saved: saved.has(model.id),
- model,
- }));
+ const references = speechReferenceModels(config.runtimes[target.agent]?.baseUrl ?? '').map(
+ (model) => ({
+ id: model.id,
+ name: saved.get(model.id)?.name ?? model.name,
+ checked: !saved.has(model.id) && model.defaultEnabled !== false,
+ recommended: model.defaultEnabled !== false,
+ reference: true,
+ saved: saved.has(model.id),
+ model,
+ }),
+ );
const ids = new Set(references.map((model) => model.id));
return [
...references,
diff --git a/apps/desktop/src/shared/speechProviderSetup.ts b/apps/desktop/src/shared/speechProviderSetup.ts
index cd0433a644..89729df91d 100644
--- a/apps/desktop/src/shared/speechProviderSetup.ts
+++ b/apps/desktop/src/shared/speechProviderSetup.ts
@@ -13,13 +13,28 @@ import {
export type SpeechService = 'minimax' | 'elevenlabs' | 'openai' | 'bailian';
/** Onboarding references only: no account entitlement or synthesis probe is implied. */
-const references: Record<SpeechService, string[]> = {
+const references: Record<Exclude<SpeechService, 'bailian'>, string[]> = {
minimax: ['speech-2.8-turbo', 'speech-2.8-hd', 'speech-02-turbo'],
elevenlabs: ['eleven_flash_v2_5', 'eleven_multilingual_v2', 'eleven_v3'],
openai: ['gpt-4o-mini-tts', 'tts-1', 'tts-1-hd'],
- bailian: ['cosyvoice-v3-flash', 'qwen3-tts-flash', 'cosyvoice-v3-plus'],
};
+const bailianWorkspaceHost = /^[a-z0-9-]+\.(cn-beijing|ap-southeast-1)\.maas\.aliyuncs\.com$/;
+
+function bailianReferences(hostname: string): string[] {
+ // Match the native HTTP endpoint/model scope in #6660. Public DashScope uses
+ // Qwen3's multimodal API; CosyVoice's HTTP API requires a Beijing workspace.
+ // These suggestions neither rewrite the connection nor prove account access.
+ if (['dashscope.aliyuncs.com', 'dashscope-intl.aliyuncs.com'].includes(hostname))
+ return ['qwen3-tts-flash'];
+ const region = bailianWorkspaceHost.exec(hostname)?.[1];
+ if (region === 'cn-beijing') return ['cosyvoice-v3-flash', 'cosyvoice-v3-plus'];
+ if (region === 'ap-southeast-1') return ['qwen-audio-3.0-tts-flash', 'qwen-audio-3.0-tts-plus'];
+ // The US endpoint has no verified native TTS binding; discovery/manual entry
+ // stays available without a guessed default or recommended reference model.
+ return [];
+}
+
export function speechServiceForUrl(value: string): SpeechService | undefined {
try {
const url = new URL(value);
@@ -52,7 +67,7 @@ export function speechServiceForUrl(value: string): SpeechService | undefined {
'dashscope-intl.aliyuncs.com',
'dashscope-us.aliyuncs.com',
].includes(url.hostname) ||
- /^ws-[a-z0-9]+\.cn-beijing\.maas\.aliyuncs\.com$/.test(url.hostname)) &&
+ bailianWorkspaceHost.test(url.hostname)) &&
['/api/v1', '/compatible-mode/v1'].includes(path)
)
return 'bailian';
@@ -73,10 +88,14 @@ export function isSpeechModel(model: {
);
}
-export function speechReferenceModels(service: SpeechService): ProviderRuntimeModelConfig[] {
+export function speechReferenceModels(baseUrl: string): ProviderRuntimeModelConfig[] {
+ const service = speechServiceForUrl(baseUrl);
+ if (!service) return [];
+ const ids =
+ service === 'bailian' ? bailianReferences(new URL(baseUrl).hostname) : references[service];
// Names/capabilities remain owned by the shared Registry. This list only orders
// a few setup suggestions; adding one is an explicit user membership decision.
- return references[service].flatMap((id, index) => {
+ return ids.flatMap((id, index) => {
const base = findBaseModel(BUNDLED_CATALOG.modelRegistry, id);
return base?.defaults.mode === 'audio_speech'
? [{ id, name: base.defaults.name ?? id, defaultEnabled: index === 0 }]
@@ -164,9 +183,10 @@ export function providerSpeechSetup(
? 'minimax-global'
: 'minimax-cn'
: service === 'bailian'
- ? upstream!.includes('dashscope-intl')
- ? 'aliyun-bailian-api-global'
- : 'aliyun-bailian-api-cn'
+ ? new URL(upstream!).hostname === 'dashscope.aliyuncs.com' ||
+ new URL(upstream!).hostname.endsWith('.cn-beijing.maas.aliyuncs.com')
+ ? 'aliyun-bailian-api-cn'
+ : 'aliyun-bailian-api-global'
: `${service}-api`,
};
}
diff --git a/docs/product-rules/provider-settings.md b/docs/product-rules/provider-settings.md
index 180d3fbf25..5f1bcad322 100644
--- a/docs/product-rules/provider-settings.md
+++ b/docs/product-rules/provider-settings.md
@@ -119,7 +119,8 @@ OpenAI、Claude 和 Grok 都允许新增独立登录账号。每次新增生成
## 语音型号接入
-- 「添加供应商」保持一个入口。API Key 连接沿用选择供应商、连接、选择模型三步;在同一模型页分组展示语音模型、对话与其他模型,语音分组位于普通模型长列表之前。MiniMax、OpenAI API、ElevenLabs、百炼普通 API 提供语音参考型号,默认勾选一个推荐型号,用户可以取消或仅添加语音型号。
+- 「添加供应商」保持一个入口。API Key 连接沿用选择供应商、连接、选择模型三步;在同一模型页分组展示语音模型、对话与其他模型,语音分组位于普通模型长列表之前。MiniMax、OpenAI API、ElevenLabs、百炼普通 API 提供语音参考型号,在已确认支持的地址默认勾选一个推荐型号,用户可以取消或仅添加语音型号。
+- 两个添加入口共用实际连接地址的推荐规则,不按预设名称或 Cindy 发行区域推断:公共 `dashscope.aliyuncs.com` / `dashscope-intl.aliyuncs.com` 推荐 `qwen3-tts-flash`;北京 workspace 推荐 `cosyvoice-v3-flash`;新加坡 workspace 推荐公共 Registry 已收录的 `qwen-audio-3.0-tts-flash`。美国公共地址及未确认组合不自动推荐或勾选参考型号,目录发现和手动添加仍可用。推荐遵循 #6660 的原生 HTTP 支持范围,不更换 origin、地域或密钥;返回连接步骤编辑地址后重新计算。已有保存型号与用户显示偏好不因新推荐改变。
- 已有 API Key 连接在模型管理工具行提供「添加语音模型」。操作绑定连接 ID,追加用户选择的新型号;不重填密钥,不覆盖已有名称、协议、模型类型、开关和其他 runtime。Main 在原有供应商修改队列内合并最新配置,连接地址已变更则要求重新打开;不使用旧快照替换整个连接。
- 订阅或设备码连接走独立 API Key 连接,不转换、不覆盖原连接。普通百炼 API 与 Coding Plan / Token Plan 入口分开,不推断套餐具有语音额度。
- 目录返回、公共参考和用户手填是不同来源。参考清单只描述添加建议,名称与能力仍继承公共 Registry;未选中的参考型号不写入账号连接。目录明确返回非 TTS 能力时,不以参考清单把它恢复成语音。