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): honor speech selections and surface recommendations

JiangJiahaoafdb9da6c80b
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 721ec4569f..b4933e2f00 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
@@ -145,7 +145,14 @@ void mountPage({
},
],
}
- : { ok: false, code: 'NETWORK' },
+ : 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 };
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 ec33163fb7..12f9c93c4b 100644
--- a/apps/desktop/scripts/fixtures/provider-speech-setup/ui-page.mjs
+++ b/apps/desktop/scripts/fixtures/provider-speech-setup/ui-page.mjs
@@ -13,7 +13,7 @@ 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 API-key connection with Qwen/CosyVoice references' },
+ bailian: { about: 'Alibaba Cloud speech recommendations visible above 100 discovered chat models' },
subscription: { about: 'Existing OpenAI subscription creates a separate API-key connection' },
},
async ready(page) {
@@ -59,6 +59,12 @@ export default {
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);
+ assert.ok(box && box.y >= 0 && box.y + box.height < height);
+ }
await recommended.scrollIntoViewIfNeeded();
await ui.capture('models');
const finish =
diff --git a/apps/desktop/src/main/voice-conversation/__tests__/ipc.test.ts b/apps/desktop/src/main/voice-conversation/__tests__/ipc.test.ts
index da67f9620a..e077141999 100644
--- a/apps/desktop/src/main/voice-conversation/__tests__/ipc.test.ts
+++ b/apps/desktop/src/main/voice-conversation/__tests__/ipc.test.ts
@@ -65,6 +65,7 @@ vi.mock('../dialogue.js', () => ({ CompanionCallDialogue: class {
import { normalizeVoiceInputSettings } from '../../../shared/voiceInputData.js';
import { registerVoiceConversationIpc } from '../index.js';
import { outboundFetch } from '../../maker-host/outbound-fetch.js';
+import { clearModelVisibilityMirror, setModelVisibilityMirror } from '../../maker-host/model-visibility-mirror.js';
const sender = (id = 1) => ({ sender: { id, once: vi.fn() } });
const call = (name: string, ...args: unknown[]) =>
h.handlers.get(`voice-conversation:${name}`)!(sender(), ...args);
@@ -74,6 +75,7 @@ beforeEach(() => {
h.trusted.mockReset();
h.append.mockReset();
h.boundary = false;
+ setModelVisibilityMirror({});
vi.mocked(outboundFetch).mockReset();
let settings = normalizeVoiceInputSettings({});
registerVoiceConversationIpc({
@@ -95,6 +97,7 @@ beforeEach(() => {
});
});
afterEach(async () => {
+ clearModelVisibilityMirror();
h.trusted.mockReset();
await call('end', 'call');
await call('stop-preview', 'preview');
@@ -108,6 +111,19 @@ const start = () =>
voice: 'alloy',
});
+// PR #6768 unit regression: model discovery must wait for current-owner display
+// preferences. Existing IPC tests start with ready state and cannot catch this race.
+it('waits for display preferences before returning the speech picker catalog', async () => {
+ clearModelVisibilityMirror();
+ const ready = vi.fn();
+ const pending = call('models') as Promise<unknown>;
+ void pending.then(ready);
+ await Promise.resolve();
+ expect(ready).not.toHaveBeenCalled();
+ setModelVisibilityMirror({});
+ await expect(pending).resolves.toEqual([expect.objectContaining({ id: 'voice-test' })]);
+});
+
it('patches playback without replacing the saved route and validates boolean overrides', async () => {
const choice = { modelId: 'voice-test', voice: 'alloy', speed: 1.15 };
await call('set-preference', 'bot-a', choice);
diff --git a/apps/desktop/src/main/voice-conversation/__tests__/models.test.ts b/apps/desktop/src/main/voice-conversation/__tests__/models.test.ts
index c23fc17301..8e970913bf 100644
--- a/apps/desktop/src/main/voice-conversation/__tests__/models.test.ts
+++ b/apps/desktop/src/main/voice-conversation/__tests__/models.test.ts
@@ -39,6 +39,31 @@ function custom(
);
}
+// PR #6768 unit regression: speechModels must honor the wizard's saved deselection
+// and later Settings switches. Existing route tests only covered the disable gate.
+// Minimal input: two discovered TTS models, one unchecked, on two separate connections.
+it('lists only selected speech models and applies connection-scoped display overrides', () => {
+ const models: ProviderRuntimeModelConfig[] = [
+ { id: 'eleven_flash_v2_5', name: 'Flash', discoveredMetadata: { mode: 'audio_speech' } },
+ { id: 'eleven_v3', name: 'V3', discoveredMetadata: { mode: 'audio_speech' }, defaultEnabled: false },
+ ];
+ const d = deps([
+ custom('a', 'https://api.elevenlabs.io/v1', models),
+ custom('b', 'https://api.elevenlabs.io/v1', models),
+ ]);
+ const choices = () => speechModels(d).map((item) => [item.providerId, item.id]);
+ expect(choices()).toEqual([['a', 'eleven_flash_v2_5'], ['b', 'eleven_flash_v2_5']]);
+ d.visibilityOverride = (agent, provider, id) => agent === 'codex' && provider === 'a'
+ ? id === 'eleven_v3' : undefined;
+ expect(choices()).toEqual([['a', 'eleven_v3'], ['b', 'eleven_flash_v2_5']]);
+ d.disabled = (provider, id) => provider === 'a' && id === 'eleven_v3';
+ expect(choices()).toEqual([['b', 'eleven_flash_v2_5']]);
+ // Hiding remains a display preference, not a replacement for route admission.
+ expect(resolveSpeechConnection(d, { providerId: 'a', agent: 'codex', modelId: 'eleven_flash_v2_5' }).protocol).toBe('elevenlabs');
+ d.visibilityOverride = () => undefined;
+ expect(choices()).toEqual([['a', 'eleven_flash_v2_5'], ['b', 'eleven_flash_v2_5']]);
+});
+
// Protocol authority comes from the real endpoint, not the UI's override.
it('rejects unsupported speed on an official binding before opening a call', () => {
const d = deps([
diff --git a/apps/desktop/src/main/voice-conversation/index.ts b/apps/desktop/src/main/voice-conversation/index.ts
index f67496bfe7..180febd382 100644
--- a/apps/desktop/src/main/voice-conversation/index.ts
+++ b/apps/desktop/src/main/voice-conversation/index.ts
@@ -7,6 +7,7 @@ import {
type VoiceConversationPreference,
} from '../../shared/voiceConversation.js';
import { getActiveCatalog } from '../maker-host/active-catalog.js';
+import { getModelVisibilityOverride, waitForModelVisibilityMirror } from '../maker-host/model-visibility-mirror.js';
import { getProviderSecretStore, readCustomProviderKey } from '../secrets/providerSecretStore.js';
import { effectiveXdGatewayBaseUrl } from '../model-access/effectiveEndpoint.js';
import { outboundFetch } from '../maker-host/outbound-fetch.js';
@@ -50,6 +51,7 @@ export function registerVoiceConversationIpc(deps: {
catalog: getActiveCatalog(),
gatewayAllowed: getAppCapabilities().canUseCindyGateway,
disabled: isProviderModelRouteDisabled,
+ visibilityOverride: getModelVisibilityOverride,
readKey: readCustomProviderKey,
gatewayConnection: () => ({
baseUrl: effectiveXdGatewayBaseUrl(),
@@ -102,7 +104,10 @@ export function registerVoiceConversationIpc(deps: {
}
});
};
- handle('models', () => service.models());
+ handle('models', async () => {
+ await waitForModelVisibilityMirror();
+ return service.models();
+ });
handle('voices', (_event, raw) => {
const selection = normalizeSpeechModelSelection(raw);
if (!selection) throwIpcError('INVALID_PARAMS', 'Invalid speech model');
diff --git a/apps/desktop/src/main/voice-conversation/models.ts b/apps/desktop/src/main/voice-conversation/models.ts
index 39e2f791e6..865abc3de1 100644
--- a/apps/desktop/src/main/voice-conversation/models.ts
+++ b/apps/desktop/src/main/voice-conversation/models.ts
@@ -1,6 +1,7 @@
import {
BUNDLED_CATALOG,
isLoopbackProviderUrl,
+ isModelVisible,
resolveModelNativeApi,
type AgentKind,
type Catalog,
@@ -21,6 +22,7 @@ export type SpeechCatalogDeps = {
catalog: Catalog;
gatewayAllowed: boolean;
disabled(providerId: string, modelId: string): boolean;
+ visibilityOverride?(agent: AgentKind, providerId: string, modelId: string): boolean | undefined;
gatewayConnection(): { baseUrl: string; apiKey: string };
readKey(providerId: string, agent: AgentKind): string | null;
};
@@ -92,6 +94,12 @@ function bindings(deps: SpeechCatalogDeps, selection?: SpeechModelSelection): Bi
(media.availability !== undefined && media.availability !== 'available')
)
continue;
+ // Display preferences affect the picker, independently of route admission.
+ // BYOK media keeps the runtime that owns both its key and display override.
+ if (!selection && !isModelVisible(
+ media.sourceAgent ? deps.visibilityOverride?.(media.sourceAgent, provider.id, media.id) : undefined,
+ media.defaultEnabled,
+ )) continue;
const declared = nativeProtocol(deps.catalog, provider.id, media);
if (provider.id === 'xd') {
if (
diff --git a/apps/desktop/src/renderer/__tests__/addProviderWizardPresetEntry.test.tsx b/apps/desktop/src/renderer/__tests__/addProviderWizardPresetEntry.test.tsx
index 8fe8390f3b..de593fd83d 100644
--- a/apps/desktop/src/renderer/__tests__/addProviderWizardPresetEntry.test.tsx
+++ b/apps/desktop/src/renderer/__tests__/addProviderWizardPresetEntry.test.tsx
@@ -1497,6 +1497,19 @@ it('sets up Sub2API from one site address and retains discovered capabilities in
});
describe('speech models in the existing provider wizard', () => {
+ // PR #6768 unit regression: speech choices must be reached before a long chat
+ // catalog. Existing setup tests checked membership/defaults, never reading order.
+ it('puts Aliyun speech choices before the discovered general catalog and keeps the recommendation selected', async () => {
+ vi.mocked(window.electronAPI.maker.fetchProviderModels).mockResolvedValue({ ok: true, models:
+ 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 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');
+ });
it('reuses server API presets instead of duplicating the same speech service entry', async () => {
window.electronAPI.maker.listProviderPresets = vi.fn(async () => ({ presets: [{
id: 'aliyun-model-studio-global', name: 'Server Model Studio', runtimes: {
diff --git a/apps/desktop/src/renderer/components/settings/AddProviderWizard.tsx b/apps/desktop/src/renderer/components/settings/AddProviderWizard.tsx
index 5e61dcd197..579989a443 100644
--- a/apps/desktop/src/renderer/components/settings/AddProviderWizard.tsx
+++ b/apps/desktop/src/renderer/components/settings/AddProviderWizard.tsx
@@ -2089,6 +2089,20 @@ export function AddProviderWizard({
{fetchState.status === 'done' && fetchState.failed && (
<InfoLine text={t('settings.providers.wizard.fetchFailed')} />
)}
+ {(speechTarget || speechPicks.length > 0) && <SpeechModelChoices
+ choices={speechPicks.map(([id, value]) => ({ id, ...value, reference: value.speechReference, manual: value.manualSpeech }))}
+ onToggle={togglePick}
+ onAdd={id => {
+ const agent = speechTarget?.agent ?? speechPicks[0]?.[1].agents[0];
+ if (!agent) return;
+ setPicks(previous => {
+ const next = new Map(previous); const existing = next.get(id);
+ next.set(id, { ...existing, name: existing?.name ?? id, checked: true,
+ recommended: existing?.recommended ?? false, agents: existing?.agents ?? [agent],
+ speechReference: false, manualSpeech: true });
+ return next;
+ });
+ }} />}
{speechPicks.length > 0 && generalPicks.length > 0 && <p className="text-13 font-medium text-[var(--text-primary)]">{t('settings.providers.speechSetup.otherModels')}</p>}
{generalPicks.length > 0 && (
<div
@@ -2162,20 +2176,6 @@ export function AddProviderWizard({
))}
</div>
)}
- {(speechTarget || speechPicks.length > 0) && <SpeechModelChoices
- choices={speechPicks.map(([id, value]) => ({ id, ...value, reference: value.speechReference, manual: value.manualSpeech }))}
- onToggle={togglePick}
- onAdd={id => {
- const agent = speechTarget?.agent ?? speechPicks[0]?.[1].agents[0];
- if (!agent) return;
- setPicks(previous => {
- const next = new Map(previous); const existing = next.get(id);
- next.set(id, { ...existing, name: existing?.name ?? id, checked: true,
- recommended: existing?.recommended ?? false, agents: existing?.agents ?? [agent],
- speechReference: false, manualSpeech: true });
- return next;
- });
- }} />}
{showManualModelFallback && (
<div
className="flex flex-col gap-3 rounded-xl border p-3.5"
diff --git a/docs/product-rules/provider-settings.md b/docs/product-rules/provider-settings.md
index eac9c994ea..180d3fbf25 100644
--- a/docs/product-rules/provider-settings.md
+++ b/docs/product-rules/provider-settings.md
@@ -119,8 +119,9 @@ 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对话与其他模型、语音模型。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,默认勾选一个推荐型号,用户可以取消或仅添加语音型号。
- 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。
+- 目录返回但未勾选的语音型号保留在连接目录中,默认隐藏,不出现在伙伴朗读选择器中。媒体投影保留该显示默认值;后续显式显示偏好按连接与 runtime 生效,显示偏好与停用、账号准入独立。
- 获取目录和保存型号不等于调用成功或已获授权;不自动发起计费合成。音色、试听、逐句朗读属于伙伴语音设置,不进入供应商添加流程。Qwen / CosyVoice 的目录接入不替代各自原生执行器的交付和验证。
diff --git a/packages/model-providers/src/__tests__/speechModelDiscovery.test.ts b/packages/model-providers/src/__tests__/speechModelDiscovery.test.ts
index ad795ded87..091a4746ea 100644
--- a/packages/model-providers/src/__tests__/speechModelDiscovery.test.ts
+++ b/packages/model-providers/src/__tests__/speechModelDiscovery.test.ts
@@ -108,8 +108,13 @@ describe('BYOK speech discovery and projection', () => {
] })!);
expect(merged[0]).toMatchObject(original[0]);
const p = provider(merged);
+ // PR #6768: the saved display switch must survive the media projection too;
+ // checking only models.codex missed unchecked TTS leaking into the voice picker.
expect(p.audioModels?.[0]).toMatchObject({ id: 'gpt-private', name: 'My Voice',
- mode: 'audio_speech', nativeApi: 'openai-audio-speech' });
+ mode: 'audio_speech', nativeApi: 'openai-audio-speech', defaultEnabled: false });
+ expect(projectProviderMediaModels(p, registry, {
+ userMetadata: () => pickModelMetadata(merged[0]),
+ }).audioModels?.[0]?.defaultEnabled).toBe(false);
expect(p.models.codex![0].defaultEnabled).toBe(false);
expect(isAgentSelectableModel(p.models.codex![0], { userProvider: true })).toBe(false);
const inherited = provider([{ ...merged[0], mode: undefined, nativeApi: undefined }]);
diff --git a/packages/model-providers/src/user-provider.ts b/packages/model-providers/src/user-provider.ts
index ffa8b56301..79446ffa05 100644
--- a/packages/model-providers/src/user-provider.ts
+++ b/packages/model-providers/src/user-provider.ts
@@ -733,6 +733,7 @@ export function buildUserProvider(
id: model.id,
name: model.name,
mode: model.mode,
+ defaultEnabled: model.defaultEnabled,
discoveredMetadata: model.discoveredMetadata,
sourceAgent: agent as AgentKind,
...(model.modalities ? { modalities: model.modalities } : {}),