feat(cap_im_tg): inbound voice/audio messages → STT pipeline

.gitignore

@@ -45,3 +45,8 @@ docs/public/d2
 docs/dist
 docs/.astro
 docs/**/*.bin
+
+claude/
+.claude/
+CONTEXT.MD
+eim_config.toml

application/edge_agent/components/app_config/app_config.c

@@ -45,6 +45,37 @@ typedef struct {
 #define APP_DEFAULT_LLM_SUPPORTS_TOOLS       "false"
 #define APP_DEFAULT_LLM_SUPPORTS_VISION      "false"
 #define APP_DEFAULT_LLM_IMAGE_REMOTE_URL_ONLY "false"
+#ifdef CONFIG_APP_STT_ENABLED
+#define APP_DEFAULT_STT_ENABLED              "true"
+#else
+#define APP_DEFAULT_STT_ENABLED              "false"
+#endif
+#ifdef CONFIG_APP_STT_BACKEND_TYPE
+#define APP_DEFAULT_STT_BACKEND_TYPE         CONFIG_APP_STT_BACKEND_TYPE
+#else
+#define APP_DEFAULT_STT_BACKEND_TYPE         "openai"
+#endif
+#ifdef CONFIG_APP_STT_API_KEY
+#define APP_DEFAULT_STT_API_KEY              CONFIG_APP_STT_API_KEY
+#else
+#define APP_DEFAULT_STT_API_KEY              ""
+#endif
+#ifdef CONFIG_APP_STT_BASE_URL
+#define APP_DEFAULT_STT_BASE_URL             CONFIG_APP_STT_BASE_URL
+#else
+#define APP_DEFAULT_STT_BASE_URL             "https://api.openai.com/v1"
+#endif
+#ifdef CONFIG_APP_STT_MODEL
+#define APP_DEFAULT_STT_MODEL                CONFIG_APP_STT_MODEL
+#else
+#define APP_DEFAULT_STT_MODEL                "whisper-1"
+#endif
+#ifdef CONFIG_APP_STT_LANGUAGE
+#define APP_DEFAULT_STT_LANGUAGE             CONFIG_APP_STT_LANGUAGE
+#else
+#define APP_DEFAULT_STT_LANGUAGE             ""
+#endif
+#define APP_DEFAULT_STT_KEEP_AUDIO           "false"
 #define APP_DEFAULT_QQ_APP_ID                ""
 #define APP_DEFAULT_QQ_APP_SECRET            ""
 #define APP_DEFAULT_FEISHU_APP_ID            ""
@@ -79,6 +110,13 @@ static const app_config_field_t s_fields[] = {
     APP_CONFIG_FIELD(llm_supports_tools, "llm_sup_tools", APP_DEFAULT_LLM_SUPPORTS_TOOLS),
     APP_CONFIG_FIELD(llm_supports_vision, "llm_sup_vis", APP_DEFAULT_LLM_SUPPORTS_VISION),
     APP_CONFIG_FIELD(llm_image_remote_url_only, "llm_img_url_o", APP_DEFAULT_LLM_IMAGE_REMOTE_URL_ONLY),
+    APP_CONFIG_FIELD(stt_enabled, "stt_enabled", APP_DEFAULT_STT_ENABLED),
+    APP_CONFIG_FIELD(stt_backend_type, "stt_backend", APP_DEFAULT_STT_BACKEND_TYPE),
+    APP_CONFIG_FIELD(stt_api_key, "stt_api_key", APP_DEFAULT_STT_API_KEY),
+    APP_CONFIG_FIELD(stt_base_url, "stt_base_url", APP_DEFAULT_STT_BASE_URL),
+    APP_CONFIG_FIELD(stt_model, "stt_model", APP_DEFAULT_STT_MODEL),
+    APP_CONFIG_FIELD(stt_language, "stt_lang", APP_DEFAULT_STT_LANGUAGE),
+    APP_CONFIG_FIELD(stt_keep_audio, "stt_keep_aud", APP_DEFAULT_STT_KEEP_AUDIO),
     APP_CONFIG_FIELD(qq_app_id, "qq_app_id", APP_DEFAULT_QQ_APP_ID),
     APP_CONFIG_FIELD(qq_app_secret, "qq_app_secret", APP_DEFAULT_QQ_APP_SECRET),
     APP_CONFIG_FIELD(feishu_app_id, "feishu_app_id", APP_DEFAULT_FEISHU_APP_ID),
@@ -552,6 +590,13 @@ void app_config_to_claw(const app_config_t *config, app_claw_config_t *out)
     strlcpy(out->llm_image_remote_url_only,
             config->llm_image_remote_url_only,
             sizeof(out->llm_image_remote_url_only));
+    strlcpy(out->stt_enabled, config->stt_enabled, sizeof(out->stt_enabled));
+    strlcpy(out->stt_backend_type, config->stt_backend_type, sizeof(out->stt_backend_type));
+    strlcpy(out->stt_api_key, config->stt_api_key, sizeof(out->stt_api_key));
+    strlcpy(out->stt_base_url, config->stt_base_url, sizeof(out->stt_base_url));
+    strlcpy(out->stt_model, config->stt_model, sizeof(out->stt_model));
+    strlcpy(out->stt_language, config->stt_language, sizeof(out->stt_language));
+    strlcpy(out->stt_keep_audio, config->stt_keep_audio, sizeof(out->stt_keep_audio));
     strlcpy(out->qq_app_id, config->qq_app_id, sizeof(out->qq_app_id));
     strlcpy(out->qq_app_secret, config->qq_app_secret, sizeof(out->qq_app_secret));
     strlcpy(out->feishu_app_id, config->feishu_app_id, sizeof(out->feishu_app_id));

application/edge_agent/components/app_config/include/app_config.h

@@ -37,6 +37,13 @@ typedef struct {
     char llm_supports_tools[8];
     char llm_supports_vision[8];
     char llm_image_remote_url_only[8];
+    char stt_enabled[8];
+    char stt_backend_type[32];
+    char stt_api_key[APP_CONFIG_STR_LEN];
+    char stt_base_url[APP_CONFIG_STR_LEN];
+    char stt_model[64];
+    char stt_language[32];
+    char stt_keep_audio[8];
     char qq_app_id[32];
     char qq_app_secret[APP_CONFIG_STR_LEN];
     char feishu_app_id[64];

application/edge_agent/components/http_server/frontend_source/src/App.tsx

@@ -20,6 +20,7 @@ const BasicPage = lazy(() => import('./pages/BasicPage').then((mod) => ({ defaul
 const SearchPage = lazy(() => import('./pages/SearchPage').then((mod) => ({ default: mod.SearchPage })));
 const MemoryPage = lazy(() => import('./pages/MemoryPage').then((mod) => ({ default: mod.MemoryPage })));
 const LlmPage = lazy(() => import('./pages/LlmPage').then((mod) => ({ default: mod.LlmPage })));
+const SttPage = lazy(() => import('./pages/SttPage').then((mod) => ({ default: mod.SttPage })));
 const ImPage = lazy(() => import('./pages/ImPage').then((mod) => ({ default: mod.ImPage })));
 const CapabilitiesPage = lazy(() =>
   import('./pages/CapabilitiesPage').then((mod) => ({ default: mod.CapabilitiesPage })),
@@ -232,6 +233,9 @@ const App: Component = () => {
               <Show when={currentTab() === 'llm'}>
                 <LlmPage />
               </Show>
+              <Show when={currentTab() === 'stt'}>
+                <SttPage onRestartRequest={() => void handleRestartRequest({ reloadOnSuccess: true })} />
+              </Show>
               <Show when={currentTab() === 'im'}>
                 <ImPage />
               </Show>

application/edge_agent/components/http_server/frontend_source/src/api/client.ts

@@ -16,6 +16,13 @@ export type AppConfig = {
   llm_supports_tools: string;
   llm_supports_vision: string;
   llm_image_remote_url_only: string;
+  stt_enabled: string;
+  stt_backend_type: string;
+  stt_api_key: string;
+  stt_base_url: string;
+  stt_model: string;
+  stt_language: string;
+  stt_keep_audio: string;
   qq_app_id: string;
   qq_app_secret: string;
   feishu_app_id: string;
@@ -39,6 +46,7 @@ export type AppConfig = {
 export type ConfigGroup =
   | 'wifi'
   | 'llm'
+  | 'stt'
   | 'im'
   | 'search'
   | 'capabilities'
@@ -61,6 +69,15 @@ export const GROUP_FIELDS: Record<ConfigGroup, (keyof AppConfig)[]> = {
     'llm_supports_vision',
     'llm_image_remote_url_only',
   ],
+  stt: [
+    'stt_enabled',
+    'stt_backend_type',
+    'stt_api_key',
+    'stt_base_url',
+    'stt_model',
+    'stt_language',
+    'stt_keep_audio',
+  ],
   im: [
     'qq_app_id',
     'qq_app_secret',

application/edge_agent/components/http_server/frontend_source/src/components/layout/Sidebar.tsx

@@ -1,5 +1,6 @@
 import {
   Activity,
+  AudioLines,
   Blocks,
   Bot,
   DatabaseZap,
@@ -22,6 +23,7 @@ const IconStatus: Component = () => <Activity class={iconClass} />;
 const IconGear: Component = () => <Settings class={iconClass} />;
 const IconWifi: Component = () => <WifiPen class={iconClass} />;
 const IconLlm: Component = () => <Bot class={iconClass} />;
+const IconStt: Component = () => <AudioLines class={iconClass} />;
 const IconIm: Component = () => <MessageSquareCode class={iconClass} />;
 const IconSearch: Component = () => <Search class={iconClass} />;
 const IconMemory: Component = () => <DatabaseZap class={iconClass} />;
@@ -37,6 +39,7 @@ export type LeafNode = {
     | 'navStatus'
     | 'navBasic'
     | 'navLlm'
+    | 'navStt'
     | 'navIm'
     | 'navSearch'
     | 'navMemory'
@@ -65,6 +68,7 @@ export const NAV_TREE: NavNode[] = [
     children: [
       { kind: 'leaf', id: 'basic', labelKey: 'navBasic', icon: IconWifi },
       { kind: 'leaf', id: 'llm', labelKey: 'navLlm', icon: IconLlm },
+      { kind: 'leaf', id: 'stt', labelKey: 'navStt', icon: IconStt },
       { kind: 'leaf', id: 'im', labelKey: 'navIm', icon: IconIm },
       { kind: 'leaf', id: 'search', labelKey: 'navSearch', icon: IconSearch },
     ],

application/edge_agent/components/http_server/frontend_source/src/i18n/en.ts

@@ -19,6 +19,7 @@ export const en = {
   navSystemSettings: 'System Settings',
   navBasic: 'Basic Settings',
   navLlm: 'LLM',
+  navStt: 'Speech-to-Text',
   navIm: 'IM',
   navSearch: 'Web Search',
   navMemory: 'Memory',
@@ -132,6 +133,25 @@ export const en = {
   llmValidationMaxTokens: 'Max Tokens must be a positive integer.',
   llmValidationImageMaxBytes: 'Default Image Max Bytes must be a positive integer.',
 
+  sectionStt: 'Speech-to-Text',
+  sttNote:
+    'When enabled, voice and audio messages received from chat platforms (e.g. Telegram voice notes) are transcribed and fed to the agent as text.',
+  sttEnabled: 'Enable inbound transcription',
+  sttBackendType: 'Backend',
+  sttBackendOpenai: 'OpenAI / Whisper-compatible',
+  sttBackendDeepgram: 'Deepgram',
+  sttApiKey: 'API Key',
+  sttBaseUrl: 'Base URL',
+  sttBaseUrlPlaceholder: 'https://api.openai.com/v1',
+  sttModel: 'Model',
+  sttModelPlaceholderOpenai: 'whisper-1',
+  sttModelPlaceholderDeepgram: 'nova-2',
+  sttLanguage: 'Language (leave blank for multi-language auto-detect)',
+  sttLanguagePlaceholder: 'e.g. es, en, pt — blank = auto-detect',
+  sttKeepAudio: 'Keep raw audio in FATFS (debug / retry)',
+  sttKeepAudioHint:
+    'Off by default — voice notes stream straight from Telegram into the STT API (faster, no flash wear). Enable to also save the original .oga to /fatfs/inbox/ for retry or inspection.',
+
   sectionIm: 'Instant Messaging (IM)',
   imAdd: 'Add',
   imClearConf: 'Clear Config',

application/edge_agent/components/http_server/frontend_source/src/i18n/zh-cn.ts

@@ -20,6 +20,7 @@ export const zhCn: Dict = {
   navSystemSettings: '系统设置',
   navBasic: '基本设置',
   navLlm: 'LLM 设置',
+  navStt: '语音转文字',
   navIm: 'IM 设置',
   navSearch: '网络搜索设置',
   navMemory: '记忆管理',
@@ -128,6 +129,25 @@ export const zhCn: Dict = {
   llmValidationMaxTokens: '最大 Token 数必须是正整数。',
   llmValidationImageMaxBytes: '默认图片大小上限必须是正整数。',
 
+  sectionStt: '语音转文字',
+  sttNote:
+    '启用后，聊天平台收到的语音和音频消息（如 Telegram 语音条）会被转写为文本并送入 Agent。',
+  sttEnabled: '启用入站语音转写',
+  sttBackendType: '后端',
+  sttBackendOpenai: 'OpenAI / Whisper 兼容',
+  sttBackendDeepgram: 'Deepgram',
+  sttApiKey: 'API Key',
+  sttBaseUrl: 'Base URL',
+  sttBaseUrlPlaceholder: 'https://api.openai.com/v1',
+  sttModel: '模型',
+  sttModelPlaceholderOpenai: 'whisper-1',
+  sttModelPlaceholderDeepgram: 'nova-2',
+  sttLanguage: '语言（留空则自动检测多语言）',
+  sttLanguagePlaceholder: '例如 es、en、pt — 留空表示自动检测',
+  sttKeepAudio: '保留原始音频到 FATFS（调试 / 重试）',
+  sttKeepAudioHint:
+    '默认关闭——语音消息直接从 Telegram 流到 STT API（更快、不写闪存）。启用后会将原始 .oga 保存到 /fatfs/inbox/ 以便重试或检查。',
+
   sectionIm: '即时通讯 (IM)',
   imAdd: '添加',
   imClearConf: '清空配置',