iobroker.assistant 1.4.5 → 1.4.7

package/build/lib/media.js

@@ -0,0 +1,226 @@
+"use strict";
+/**
+ * Pictures and videos that come with a question (`ask` with `images` / `video`, e.g. from Telegram).
+ *
+ * Pictures go to the cloud model as they are (base64, the vendors take JPEG/PNG/GIF/WebP). Videos are not looked at
+ * yet: the sender keeps them in the ioBroker file storage and only says where, so a video never travels through the
+ * message bus as base64.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.MAX_IMAGE_BASE64 = exports.MAX_IMAGES = void 0;
+exports.sniffImageType = sniffImageType;
+exports.parseAskImages = parseAskImages;
+exports.parseAskVideo = parseAskVideo;
+exports.isVideoError = isVideoError;
+exports.redactImages = redactImages;
+exports.imageOnlyQuestion = imageOnlyQuestion;
+exports.videoNotSupportedText = videoNotSupportedText;
+exports.noVisionText = noVisionText;
+exports.attachmentNote = attachmentNote;
+exports.historyText = historyText;
+/** At most this many pictures per question (an album of Telegram has up to 10) */
+exports.MAX_IMAGES = 4;
+/** Largest picture as base64 text: Anthropic refuses images over 5 MB */
+exports.MAX_IMAGE_BASE64 = 5 * 1024 * 1024;
+const IMAGE_TYPES = new Set(['image/jpeg', 'image/png', 'image/gif', 'image/webp']);
+/**
+ * The type of a picture from its first bytes, `null` if it is none of the four the vendors take.
+ *
+ * @param buf the start of the file (16 bytes are enough)
+ */
+function sniffImageType(buf) {
+    if (buf.length >= 3 && buf[0] === 0xff && buf[1] === 0xd8 && buf[2] === 0xff) {
+        return 'image/jpeg';
+    }
+    if (buf.length >= 8 && buf.subarray(0, 8).equals(Buffer.from([0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a]))) {
+        return 'image/png';
+    }
+    if (buf.length >= 6 && /^GIF8[79]a$/.test(buf.subarray(0, 6).toString('latin1'))) {
+        return 'image/gif';
+    }
+    if (buf.length >= 12 && buf.toString('latin1', 0, 4) === 'RIFF' && buf.toString('latin1', 8, 12) === 'WEBP') {
+        return 'image/webp';
+    }
+    return null;
+}
+/**
+ * Check the `images` of an `ask`: base64 of JPEG/PNG/GIF/WebP, at most {@link MAX_IMAGES} and
+ * {@link MAX_IMAGE_BASE64} each. The type is taken from the content, not from what the sender says.
+ *
+ * @param raw the `images` field: a list of `{ data, mimeType? }` (one object or a bare base64 string work too)
+ * @returns the pictures, or the reason they are refused
+ */
+function parseAskImages(raw) {
+    if (raw === undefined || raw === null || raw === '') {
+        return { images: [] };
+    }
+    const list = Array.isArray(raw) ? raw : [raw];
+    if (list.length > exports.MAX_IMAGES) {
+        return { images: [], error: `too many pictures (${list.length}, at most ${exports.MAX_IMAGES})` };
+    }
+    const images = [];
+    for (const item of list) {
+        const data = typeof item === 'string' ? item : item?.data;
+        if (typeof data !== 'string' || !data) {
+            return { images: [], error: 'a picture without data' };
+        }
+        // a data URL is fine as well
+        const base64 = data.replace(/^data:[^;,]*;base64,/, '').replace(/\s+/g, '');
+        if (base64.length > exports.MAX_IMAGE_BASE64) {
+            return {
+                images: [],
+                error: `picture too big (${Math.round((base64.length * 3) / 4 / 1024)} KB, at most ${(exports.MAX_IMAGE_BASE64 * 3) / 4 / 1024} KB)`,
+            };
+        }
+        const mimeType = sniffImageType(Buffer.from(base64.slice(0, 32), 'base64'));
+        if (!mimeType || !IMAGE_TYPES.has(mimeType)) {
+            return { images: [], error: 'not a JPEG, PNG, GIF or WebP picture' };
+        }
+        images.push({ data: base64, mimeType });
+    }
+    return { images };
+}
+/**
+ * Check the `video` of an `ask`: only a reference into the ioBroker file storage, never the content.
+ *
+ * @param raw the `video` field
+ * @returns the reference, `null` without one, or the reason it is refused (e.g. base64 content)
+ */
+function parseAskVideo(raw) {
+    if (raw === undefined || raw === null || raw === '') {
+        return null;
+    }
+    if (typeof raw !== 'object' || Array.isArray(raw)) {
+        return { error: 'video must be { adapter, path } of the ioBroker file storage' };
+    }
+    const v = raw;
+    if ('data' in v) {
+        return { error: 'a video is not sent as content - store it and send { adapter, path }' };
+    }
+    const str = (x) => (typeof x === 'string' ? x.trim() : '');
+    const num = (x) => typeof x === 'number' && Number.isFinite(x) && x >= 0 ? x : undefined;
+    const ref = {
+        adapter: str(v.adapter),
+        path: str(v.path).replace(/^\/+/, ''),
+        mimeType: str(v.mimeType) || undefined,
+        durationSec: num(v.durationSec),
+        size: num(v.size),
+        error: str(v.error) || undefined,
+    };
+    if ((!ref.adapter || !ref.path) && !ref.error) {
+        return { error: 'video without adapter and path' };
+    }
+    return ref;
+}
+/** Whether {@link parseAskVideo} refused the video */
+function isVideoError(v) {
+    return !!v && !('adapter' in v);
+}
+/**
+ * Copy of a request / message list with the pictures replaced by a short note — for the trace and the debug log,
+ * where hundreds of KB of base64 per round help nobody. Understands the OpenAI (`image_url` with a data URL) and the
+ * Anthropic form (`image` with a base64 `source`).
+ *
+ * @param value what to copy
+ */
+function redactImages(value) {
+    const walk = (v) => {
+        if (Array.isArray(v)) {
+            return v.map(walk);
+        }
+        if (!v || typeof v !== 'object') {
+            return v;
+        }
+        const o = v;
+        if (o.type === 'image_url' && typeof o.image_url?.url === 'string' && o.image_url.url.startsWith('data:')) {
+            return { ...o, image_url: { ...o.image_url, url: imageNote(o.image_url.url) } };
+        }
+        if (o.type === 'image' && o.source?.type === 'base64' && typeof o.source.data === 'string') {
+            return { ...o, source: { ...o.source, data: imageNote(o.source.data, o.source.media_type) } };
+        }
+        const out = {};
+        for (const [k, x] of Object.entries(o)) {
+            out[k] = walk(x);
+        }
+        return out;
+    };
+    return walk(value);
+}
+function imageNote(data, mimeType) {
+    const m = /^data:([^;,]+)/.exec(data);
+    const base64 = data.replace(/^data:[^,]*,/, '');
+    return `<${mimeType || m?.[1] || 'image'}, ${Math.round((base64.length * 3) / 4 / 1024)} KB>`;
+}
+/**
+ * The question for pictures that came without one (a photo without caption).
+ *
+ * @param lang language of the answer
+ * @param count number of pictures
+ */
+function imageOnlyQuestion(lang, count = 1) {
+    if (lang === 'ru') {
+        return count > 1 ? 'Что на этих картинках?' : 'Что на этой картинке?';
+    }
+    if (lang === 'de') {
+        return count > 1 ? 'Was ist auf diesen Bildern?' : 'Was ist auf diesem Bild?';
+    }
+    return count > 1 ? 'What is in these pictures?' : 'What is in this picture?';
+}
+/**
+ * The answer to a video that came without a question.
+ *
+ * @param lang language of the answer
+ */
+function videoNotSupportedText(lang) {
+    if (lang === 'ru') {
+        return 'Видео я пока смотреть не умею.';
+    }
+    if (lang === 'de') {
+        return 'Videos kann ich mir noch nicht ansehen.';
+    }
+    return "I can't watch videos yet.";
+}
+/**
+ * The answer to pictures when there is no cloud model to look at them.
+ *
+ * @param lang language of the answer
+ */
+function noVisionText(lang) {
+    if (lang === 'ru') {
+        return 'Картинки я могу посмотреть только с облачной моделью.';
+    }
+    if (lang === 'de') {
+        return 'Bilder kann ich mir nur mit einem Cloud-Modell ansehen.';
+    }
+    return 'I can only look at pictures with a cloud model.';
+}
+/**
+ * Note for the model in the user turn: what came along besides the text. A model that does not get the pictures
+ * (an older pi-agent) at least knows that there were some.
+ *
+ * @param images number of pictures attached
+ * @param video whether a video came along
+ */
+function attachmentNote(images, video) {
+    const parts = [];
+    if (images) {
+        parts.push(images > 1 ? `${images} pictures are attached.` : 'A picture is attached.');
+    }
+    if (video) {
+        parts.push('The user also sent a video, which you cannot watch yet - say so if the question is about it.');
+    }
+    return parts.length ? `(${parts.join(' ')})` : '';
+}
+/**
+ * How the question is kept in the conversation history: the pictures are not repeated in later turns, a marker says
+ * there were some.
+ *
+ * @param question the question
+ * @param images number of pictures
+ * @param video whether a video came along
+ */
+function historyText(question, images, video) {
+    const marks = [images ? `[${images > 1 ? `${images} pictures` : 'picture'}]` : '', video ? '[video]' : ''];
+    return [...marks, question].filter(Boolean).join(' ');
+}
+//# sourceMappingURL=media.js.map

package/build/lib/media.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"media.js","sourceRoot":"","sources":["../../src/lib/media.ts"],"names":[],"mappings":";AAAA;;;;;;GAMG;;;AAkCH,wCAcC;AASD,wCA6BC;AAQD,sCA0BC;AAGD,oCAEC;AASD,oCAsBC;AAiBD,8CAQC;AAOD,sDAQC;AAOD,oCAQC;AASD,wCASC;AAUD,kCAGC;AA7ND,kFAAkF;AACrE,QAAA,UAAU,GAAG,CAAC,CAAC;AAE5B,yEAAyE;AAC5D,QAAA,gBAAgB,GAAG,CAAC,GAAG,IAAI,GAAG,IAAI,CAAC;AAEhD,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,CAAC,YAAY,EAAE,WAAW,EAAE,WAAW,EAAE,YAAY,CAAC,CAAC,CAAC;AAEpF;;;;GAIG;AACH,SAAgB,cAAc,CAAC,GAAW;IACtC,IAAI,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,GAAG,CAAC,CAAC,CAAC,KAAK,IAAI,IAAI,GAAG,CAAC,CAAC,CAAC,KAAK,IAAI,IAAI,GAAG,CAAC,CAAC,CAAC,KAAK,IAAI,EAAE,CAAC;QAC3E,OAAO,YAAY,CAAC;IACxB,CAAC;IACD,IAAI,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,GAAG,CAAC,QAAQ,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;QAC9G,OAAO,WAAW,CAAC;IACvB,CAAC;IACD,IAAI,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,aAAa,CAAC,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,EAAE,CAAC;QAC/E,OAAO,WAAW,CAAC;IACvB,CAAC;IACD,IAAI,GAAG,CAAC,MAAM,IAAI,EAAE,IAAI,GAAG,CAAC,QAAQ,CAAC,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC,KAAK,MAAM,IAAI,GAAG,CAAC,QAAQ,CAAC,QAAQ,EAAE,CAAC,EAAE,EAAE,CAAC,KAAK,MAAM,EAAE,CAAC;QAC1G,OAAO,YAAY,CAAC;IACxB,CAAC;IACD,OAAO,IAAI,CAAC;AAChB,CAAC;AAED;;;;;;GAMG;AACH,SAAgB,cAAc,CAAC,GAAY;IACvC,IAAI,GAAG,KAAK,SAAS,IAAI,GAAG,KAAK,IAAI,IAAI,GAAG,KAAK,EAAE,EAAE,CAAC;QAClD,OAAO,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;IAC1B,CAAC;IACD,MAAM,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;IAC9C,IAAI,IAAI,CAAC,MAAM,GAAG,kBAAU,EAAE,CAAC;QAC3B,OAAO,EAAE,MAAM,EAAE,EAAE,EAAE,KAAK,EAAE,sBAAsB,IAAI,CAAC,MAAM,aAAa,kBAAU,GAAG,EAAE,CAAC;IAC9F,CAAC;IACD,MAAM,MAAM,GAAiB,EAAE,CAAC;IAChC,KAAK,MAAM,IAAI,IAAI,IAAI,EAAE,CAAC;QACtB,MAAM,IAAI,GAAG,OAAO,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAE,IAAkC,EAAE,IAAI,CAAC;QACzF,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,CAAC,IAAI,EAAE,CAAC;YACpC,OAAO,EAAE,MAAM,EAAE,EAAE,EAAE,KAAK,EAAE,wBAAwB,EAAE,CAAC;QAC3D,CAAC;QACD,6BAA6B;QAC7B,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,sBAAsB,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QAC5E,IAAI,MAAM,CAAC,MAAM,GAAG,wBAAgB,EAAE,CAAC;YACnC,OAAO;gBACH,MAAM,EAAE,EAAE;gBACV,KAAK,EAAE,oBAAoB,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,gBAAgB,CAAC,wBAAgB,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,IAAI,MAAM;aAC/H,CAAC;QACN,CAAC;QACD,MAAM,QAAQ,GAAG,cAAc,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC;QAC5E,IAAI,CAAC,QAAQ,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC1C,OAAO,EAAE,MAAM,EAAE,EAAE,EAAE,KAAK,EAAE,sCAAsC,EAAE,CAAC;QACzE,CAAC;QACD,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC,CAAC;IAC5C,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,CAAC;AACtB,CAAC;AAED;;;;;GAKG;AACH,SAAgB,aAAa,CAAC,GAAY;IACtC,IAAI,GAAG,KAAK,SAAS,IAAI,GAAG,KAAK,IAAI,IAAI,GAAG,KAAK,EAAE,EAAE,CAAC;QAClD,OAAO,IAAI,CAAC;IAChB,CAAC;IACD,IAAI,OAAO,GAAG,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;QAChD,OAAO,EAAE,KAAK,EAAE,8DAA8D,EAAE,CAAC;IACrF,CAAC;IACD,MAAM,CAAC,GAAG,GAA8B,CAAC;IACzC,IAAI,MAAM,IAAI,CAAC,EAAE,CAAC;QACd,OAAO,EAAE,KAAK,EAAE,sEAAsE,EAAE,CAAC;IAC7F,CAAC;IACD,MAAM,GAAG,GAAG,CAAC,CAAU,EAAU,EAAE,CAAC,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAC5E,MAAM,GAAG,GAAG,CAAC,CAAU,EAAsB,EAAE,CAC3C,OAAO,CAAC,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAC1E,MAAM,GAAG,GAAa;QAClB,OAAO,EAAE,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC;QACvB,IAAI,EAAE,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC;QACrC,QAAQ,EAAE,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,SAAS;QACtC,WAAW,EAAE,GAAG,CAAC,CAAC,CAAC,WAAW,CAAC;QAC/B,IAAI,EAAE,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC;QACjB,KAAK,EAAE,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,SAAS;KACnC,CAAC;IACF,IAAI,CAAC,CAAC,GAAG,CAAC,OAAO,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,CAAC;QAC5C,OAAO,EAAE,KAAK,EAAE,gCAAgC,EAAE,CAAC;IACvD,CAAC;IACD,OAAO,GAAG,CAAC;AACf,CAAC;AAED,sDAAsD;AACtD,SAAgB,YAAY,CAAC,CAAsC;IAC/D,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,CAAC;AACpC,CAAC;AAED;;;;;;GAMG;AACH,SAAgB,YAAY,CAAI,KAAQ;IACpC,MAAM,IAAI,GAAG,CAAC,CAAU,EAAW,EAAE;QACjC,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;YACnB,OAAO,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACvB,CAAC;QACD,IAAI,CAAC,CAAC,IAAI,OAAO,CAAC,KAAK,QAAQ,EAAE,CAAC;YAC9B,OAAO,CAAC,CAAC;QACb,CAAC;QACD,MAAM,CAAC,GAAG,CAAwB,CAAC;QACnC,IAAI,CAAC,CAAC,IAAI,KAAK,WAAW,IAAI,OAAO,CAAC,CAAC,SAAS,EAAE,GAAG,KAAK,QAAQ,IAAI,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC;YACxG,OAAO,EAAE,GAAG,CAAC,EAAE,SAAS,EAAE,EAAE,GAAG,CAAC,CAAC,SAAS,EAAE,GAAG,EAAE,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;QACpF,CAAC;QACD,IAAI,CAAC,CAAC,IAAI,KAAK,OAAO,IAAI,CAAC,CAAC,MAAM,EAAE,IAAI,KAAK,QAAQ,IAAI,OAAO,CAAC,CAAC,MAAM,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YACzF,OAAO,EAAE,GAAG,CAAC,EAAE,MAAM,EAAE,EAAE,GAAG,CAAC,CAAC,MAAM,EAAE,IAAI,EAAE,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QAClG,CAAC;QACD,MAAM,GAAG,GAA4B,EAAE,CAAC;QACxC,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;YACrC,GAAG,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QACrB,CAAC;QACD,OAAO,GAAG,CAAC;IACf,CAAC,CAAC;IACF,OAAO,IAAI,CAAC,KAAK,CAAM,CAAC;AAC5B,CAAC;AAED,SAAS,SAAS,CAAC,IAAY,EAAE,QAAiB;IAC9C,MAAM,CAAC,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACtC,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,EAAE,CAAC,CAAC;IAChD,OAAO,IAAI,QAAQ,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,OAAO,KAAK,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC;AAClG,CAAC;AAKD;;;;;GAKG;AACH,SAAgB,iBAAiB,CAAC,IAAU,EAAE,KAAK,GAAG,CAAC;IACnD,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;QAChB,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,wBAAwB,CAAC,CAAC,CAAC,uBAAuB,CAAC;IAC1E,CAAC;IACD,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;QAChB,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,6BAA6B,CAAC,CAAC,CAAC,0BAA0B,CAAC;IAClF,CAAC;IACD,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,4BAA4B,CAAC,CAAC,CAAC,0BAA0B,CAAC;AACjF,CAAC;AAED;;;;GAIG;AACH,SAAgB,qBAAqB,CAAC,IAAU;IAC5C,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;QAChB,OAAO,gCAAgC,CAAC;IAC5C,CAAC;IACD,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;QAChB,OAAO,yCAAyC,CAAC;IACrD,CAAC;IACD,OAAO,2BAA2B,CAAC;AACvC,CAAC;AAED;;;;GAIG;AACH,SAAgB,YAAY,CAAC,IAAU;IACnC,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;QAChB,OAAO,uDAAuD,CAAC;IACnE,CAAC;IACD,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;QAChB,OAAO,yDAAyD,CAAC;IACrE,CAAC;IACD,OAAO,iDAAiD,CAAC;AAC7D,CAAC;AAED;;;;;;GAMG;AACH,SAAgB,cAAc,CAAC,MAAc,EAAE,KAAc;IACzD,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,IAAI,MAAM,EAAE,CAAC;QACT,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,MAAM,yBAAyB,CAAC,CAAC,CAAC,wBAAwB,CAAC,CAAC;IAC3F,CAAC;IACD,IAAI,KAAK,EAAE,CAAC;QACR,KAAK,CAAC,IAAI,CAAC,8FAA8F,CAAC,CAAC;IAC/G,CAAC;IACD,OAAO,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;AACtD,CAAC;AAED;;;;;;;GAOG;AACH,SAAgB,WAAW,CAAC,QAAgB,EAAE,MAAc,EAAE,KAAc;IACxE,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,MAAM,WAAW,CAAC,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAC3G,OAAO,CAAC,GAAG,KAAK,EAAE,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;AAC1D,CAAC","sourcesContent":["/**\n * Pictures and videos that come with a question (`ask` with `images` / `video`, e.g. from Telegram).\n *\n * Pictures go to the cloud model as they are (base64, the vendors take JPEG/PNG/GIF/WebP). Videos are not looked at\n * yet: the sender keeps them in the ioBroker file storage and only says where, so a video never travels through the\n * message bus as base64.\n */\n\n/** A picture for the model: base64 without a `data:` prefix and its type. */\nexport interface ImageInput {\n    data: string;\n    mimeType: string;\n}\n\n/** Where the sender stored a video (ioBroker file storage), with what it knows about it. */\nexport interface VideoRef {\n    /** File storage namespace, e.g. `telegram.0` */\n    adapter: string;\n    /** Path in it, e.g. `assistant/video_2026-11-04_10-00-00.mp4` */\n    path: string;\n    mimeType?: string;\n    durationSec?: number;\n    size?: number;\n    /** Set when the sender could not store it (e.g. too big to download) */\n    error?: string;\n}\n\n/** At most this many pictures per question (an album of Telegram has up to 10) */\nexport const MAX_IMAGES = 4;\n\n/** Largest picture as base64 text: Anthropic refuses images over 5 MB */\nexport const MAX_IMAGE_BASE64 = 5 * 1024 * 1024;\n\nconst IMAGE_TYPES = new Set(['image/jpeg', 'image/png', 'image/gif', 'image/webp']);\n\n/**\n * The type of a picture from its first bytes, `null` if it is none of the four the vendors take.\n *\n * @param buf the start of the file (16 bytes are enough)\n */\nexport function sniffImageType(buf: Buffer): string | null {\n    if (buf.length >= 3 && buf[0] === 0xff && buf[1] === 0xd8 && buf[2] === 0xff) {\n        return 'image/jpeg';\n    }\n    if (buf.length >= 8 && buf.subarray(0, 8).equals(Buffer.from([0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a]))) {\n        return 'image/png';\n    }\n    if (buf.length >= 6 && /^GIF8[79]a$/.test(buf.subarray(0, 6).toString('latin1'))) {\n        return 'image/gif';\n    }\n    if (buf.length >= 12 && buf.toString('latin1', 0, 4) === 'RIFF' && buf.toString('latin1', 8, 12) === 'WEBP') {\n        return 'image/webp';\n    }\n    return null;\n}\n\n/**\n * Check the `images` of an `ask`: base64 of JPEG/PNG/GIF/WebP, at most {@link MAX_IMAGES} and\n * {@link MAX_IMAGE_BASE64} each. The type is taken from the content, not from what the sender says.\n *\n * @param raw the `images` field: a list of `{ data, mimeType? }` (one object or a bare base64 string work too)\n * @returns the pictures, or the reason they are refused\n */\nexport function parseAskImages(raw: unknown): { images: ImageInput[]; error?: string } {\n    if (raw === undefined || raw === null || raw === '') {\n        return { images: [] };\n    }\n    const list = Array.isArray(raw) ? raw : [raw];\n    if (list.length > MAX_IMAGES) {\n        return { images: [], error: `too many pictures (${list.length}, at most ${MAX_IMAGES})` };\n    }\n    const images: ImageInput[] = [];\n    for (const item of list) {\n        const data = typeof item === 'string' ? item : (item as { data?: unknown } | null)?.data;\n        if (typeof data !== 'string' || !data) {\n            return { images: [], error: 'a picture without data' };\n        }\n        // a data URL is fine as well\n        const base64 = data.replace(/^data:[^;,]*;base64,/, '').replace(/\\s+/g, '');\n        if (base64.length > MAX_IMAGE_BASE64) {\n            return {\n                images: [],\n                error: `picture too big (${Math.round((base64.length * 3) / 4 / 1024)} KB, at most ${(MAX_IMAGE_BASE64 * 3) / 4 / 1024} KB)`,\n            };\n        }\n        const mimeType = sniffImageType(Buffer.from(base64.slice(0, 32), 'base64'));\n        if (!mimeType || !IMAGE_TYPES.has(mimeType)) {\n            return { images: [], error: 'not a JPEG, PNG, GIF or WebP picture' };\n        }\n        images.push({ data: base64, mimeType });\n    }\n    return { images };\n}\n\n/**\n * Check the `video` of an `ask`: only a reference into the ioBroker file storage, never the content.\n *\n * @param raw the `video` field\n * @returns the reference, `null` without one, or the reason it is refused (e.g. base64 content)\n */\nexport function parseAskVideo(raw: unknown): VideoRef | null | { error: string } {\n    if (raw === undefined || raw === null || raw === '') {\n        return null;\n    }\n    if (typeof raw !== 'object' || Array.isArray(raw)) {\n        return { error: 'video must be { adapter, path } of the ioBroker file storage' };\n    }\n    const v = raw as Record<string, unknown>;\n    if ('data' in v) {\n        return { error: 'a video is not sent as content - store it and send { adapter, path }' };\n    }\n    const str = (x: unknown): string => (typeof x === 'string' ? x.trim() : '');\n    const num = (x: unknown): number | undefined =>\n        typeof x === 'number' && Number.isFinite(x) && x >= 0 ? x : undefined;\n    const ref: VideoRef = {\n        adapter: str(v.adapter),\n        path: str(v.path).replace(/^\\/+/, ''),\n        mimeType: str(v.mimeType) || undefined,\n        durationSec: num(v.durationSec),\n        size: num(v.size),\n        error: str(v.error) || undefined,\n    };\n    if ((!ref.adapter || !ref.path) && !ref.error) {\n        return { error: 'video without adapter and path' };\n    }\n    return ref;\n}\n\n/** Whether {@link parseAskVideo} refused the video */\nexport function isVideoError(v: VideoRef | null | { error: string }): v is { error: string } {\n    return !!v && !('adapter' in v);\n}\n\n/**\n * Copy of a request / message list with the pictures replaced by a short note — for the trace and the debug log,\n * where hundreds of KB of base64 per round help nobody. Understands the OpenAI (`image_url` with a data URL) and the\n * Anthropic form (`image` with a base64 `source`).\n *\n * @param value what to copy\n */\nexport function redactImages<T>(value: T): T {\n    const walk = (v: unknown): unknown => {\n        if (Array.isArray(v)) {\n            return v.map(walk);\n        }\n        if (!v || typeof v !== 'object') {\n            return v;\n        }\n        const o = v as Record<string, any>;\n        if (o.type === 'image_url' && typeof o.image_url?.url === 'string' && o.image_url.url.startsWith('data:')) {\n            return { ...o, image_url: { ...o.image_url, url: imageNote(o.image_url.url) } };\n        }\n        if (o.type === 'image' && o.source?.type === 'base64' && typeof o.source.data === 'string') {\n            return { ...o, source: { ...o.source, data: imageNote(o.source.data, o.source.media_type) } };\n        }\n        const out: Record<string, unknown> = {};\n        for (const [k, x] of Object.entries(o)) {\n            out[k] = walk(x);\n        }\n        return out;\n    };\n    return walk(value) as T;\n}\n\nfunction imageNote(data: string, mimeType?: string): string {\n    const m = /^data:([^;,]+)/.exec(data);\n    const base64 = data.replace(/^data:[^,]*,/, '');\n    return `<${mimeType || m?.[1] || 'image'}, ${Math.round((base64.length * 3) / 4 / 1024)} KB>`;\n}\n\n/** Language of the texts below: de, ru, else en */\ntype Lang = string;\n\n/**\n * The question for pictures that came without one (a photo without caption).\n *\n * @param lang language of the answer\n * @param count number of pictures\n */\nexport function imageOnlyQuestion(lang: Lang, count = 1): string {\n    if (lang === 'ru') {\n        return count > 1 ? 'Что на этих картинках?' : 'Что на этой картинке?';\n    }\n    if (lang === 'de') {\n        return count > 1 ? 'Was ist auf diesen Bildern?' : 'Was ist auf diesem Bild?';\n    }\n    return count > 1 ? 'What is in these pictures?' : 'What is in this picture?';\n}\n\n/**\n * The answer to a video that came without a question.\n *\n * @param lang language of the answer\n */\nexport function videoNotSupportedText(lang: Lang): string {\n    if (lang === 'ru') {\n        return 'Видео я пока смотреть не умею.';\n    }\n    if (lang === 'de') {\n        return 'Videos kann ich mir noch nicht ansehen.';\n    }\n    return \"I can't watch videos yet.\";\n}\n\n/**\n * The answer to pictures when there is no cloud model to look at them.\n *\n * @param lang language of the answer\n */\nexport function noVisionText(lang: Lang): string {\n    if (lang === 'ru') {\n        return 'Картинки я могу посмотреть только с облачной моделью.';\n    }\n    if (lang === 'de') {\n        return 'Bilder kann ich mir nur mit einem Cloud-Modell ansehen.';\n    }\n    return 'I can only look at pictures with a cloud model.';\n}\n\n/**\n * Note for the model in the user turn: what came along besides the text. A model that does not get the pictures\n * (an older pi-agent) at least knows that there were some.\n *\n * @param images number of pictures attached\n * @param video whether a video came along\n */\nexport function attachmentNote(images: number, video: boolean): string {\n    const parts: string[] = [];\n    if (images) {\n        parts.push(images > 1 ? `${images} pictures are attached.` : 'A picture is attached.');\n    }\n    if (video) {\n        parts.push('The user also sent a video, which you cannot watch yet - say so if the question is about it.');\n    }\n    return parts.length ? `(${parts.join(' ')})` : '';\n}\n\n/**\n * How the question is kept in the conversation history: the pictures are not repeated in later turns, a marker says\n * there were some.\n *\n * @param question the question\n * @param images number of pictures\n * @param video whether a video came along\n */\nexport function historyText(question: string, images: number, video: boolean): string {\n    const marks = [images ? `[${images > 1 ? `${images} pictures` : 'picture'}]` : '', video ? '[video]' : ''];\n    return [...marks, question].filter(Boolean).join(' ');\n}\n"]}

package/build/lib/piAgent.js

@@ -77,6 +77,7 @@ class PiAgentHarness {
             conversation: conversationOf(req.conversation),
             instructions: req.instructions,
             ...(req.language ? { language: req.language } : {}),
+            ...(req.images?.length ? { images: req.images } : {}),
             requestId,
         }, { timeout: this.timeoutMs }));
         if (result?.error) {

package/build/lib/piAgent.js.map

@@ -1 +1 @@
-{"version":3,"file":"piAgent.js","sourceRoot":"","sources":["../../src/lib/piAgent.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;;;;;;;GAgBG;;;AAmCH,kDAaC;AAGD,wCAEC;AAnDD,8FAA8F;AACjF,QAAA,yBAAyB,GAAG,KAAK,CAAC;AAE/C,wFAAwF;AAC3E,QAAA,oBAAoB,GAAG,EAAE,CAAC;AAsBvC;;;;;;GAMG;AACH,SAAgB,mBAAmB,CAAC,QAAkB,EAAE,QAAkB;IACtE,MAAM,MAAM,GAAG,CAAC,GAAG,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC;IAC7C,MAAM,KAAK,GAAG,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IACvC,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,OAAO,KAAK,CAAC,MAAM,EAAE,CAAC;QAClB,MAAM,IAAI,GAAG,CAAC,GAAG,MAAM,EAAE,GAAG,KAAK,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAChD,IAAI,IAAI,CAAC,MAAM,IAAI,iCAAyB,EAAE,CAAC;YAC3C,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC;QAC7B,CAAC;QACD,KAAK,CAAC,GAAG,EAAE,CAAC;QACZ,OAAO,EAAE,CAAC;IACd,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,iCAAyB,CAAC,EAAE,OAAO,EAAE,CAAC;AACtF,CAAC;AAED,0FAA0F;AAC1F,SAAgB,cAAc,CAAC,MAAc;IACzC,OAAO,CAAC,MAAM,IAAI,SAAS,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;AACnE,CAAC;AAED,MAAa,cAAc;IAIF;IAED;IACC;IANb,GAAG,GAAG,CAAC,CAAC;IAEhB,YACqB,IAAiB;IAClC,wBAAwB;IACR,QAAgB,EACf,YAAoB,4BAAoB,GAAG,IAAI;QAH/C,SAAI,GAAJ,IAAI,CAAa;QAElB,aAAQ,GAAR,QAAQ,CAAQ;QACf,cAAS,GAAT,SAAS,CAAsC;IACjE,CAAC;IAEJ;;;OAGG;IACI,KAAK,CAAC,GAAG,CAAC,GAAmB;QAChC,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,oBAAoB,CAAC,kBAAkB,IAAI,CAAC,QAAQ,QAAQ,CAAC,CAAC;QAC5F,IAAI,CAAC,KAAK,EAAE,GAAG,EAAE,CAAC;YACd,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,CAAC,QAAQ,iBAAiB,CAAC,CAAC;QACvD,CAAC;QACD,uEAAuE;QACvE,MAAM,SAAS,GAAG,aAAa,IAAI,CAAC,GAAG,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,QAAQ,CAAC,EAAE,CAAC,EAAE,CAAC;QACtF,MAAM,MAAM,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,WAAW,CACvC,IAAI,CAAC,QAAQ,EACb,KAAK,EACL;YACI,IAAI,EAAE,GAAG,CAAC,IAAI;YACd,YAAY,EAAE,cAAc,CAAC,GAAG,CAAC,YAAY,CAAC;YAC9C,YAAY,EAAE,GAAG,CAAC,YAAY;YAC9B,GAAG,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,GAAG,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACnD,SAAS;SACZ,EACD,EAAE,OAAO,EAAE,IAAI,CAAC,SAAS,EAAE,CAC9B,CAA6D,CAAC;QAC/D,IAAI,MAAM,EAAE,KAAK,EAAE,CAAC;YAChB,MAAM,KAAK,GAAG,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YAC7F,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,CAAC,QAAQ,KAAK,KAAK,EAAE,CAAC,CAAC;QAClD,CAAC;QACD,MAAM,MAAM,GAAG,OAAO,MAAM,EAAE,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC9E,IAAI,CAAC,MAAM,EAAE,CAAC;YACV,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,CAAC,QAAQ,iBAAiB,CAAC,CAAC;QACvD,CAAC;QACD,OAAO,MAAM,CAAC;IAClB,CAAC;CACJ;AA3CD,wCA2CC","sourcesContent":["/**\n * The adapter `pi-agent` as the cloud tier (\"harness\"): instead of running its own tool loop, the assistant\n * hands the question to a pi-agent instance via `sendTo(<instance>, 'ask', …)`. pi-agent stores every model\n * and tool call as a checkpoint before it runs (an interrupted request continues after a restart, a write is\n * never repeated), enforces its own write permissions and confirmations, and can use the system-wide AI\n * settings of admin — none of which the built-in `LlmAgent` does.\n *\n * Each request source (satellite, `chat`, `telegram:<user>`) gets a conversation of its own in pi-agent\n * (`conversation`), so two rooms do not see each other's follow-ups. The source-independent part of the\n * prompt (style, follow-up marker, memory, device list) goes along as the `instructions` of that\n * conversation; what changes every turn (time, weather, presence) is prepended to the question.\n * pi-agent 0.3.0 and older ignore both fields and answer everything in their root conversation.\n *\n * With \"Confirmation by text in sendTo conversations\", pi-agent answers a write it may not do unasked with a\n * question ending in `[[LISTEN]]`, and the next `ask` of the conversation is the yes or no. `language` makes it\n * ask that question in the language the satellite listens in.\n */\n\n/** Longest instructions pi-agent accepts for a conversation; anything beyond is cut there. */\nexport const PI_AGENT_MAX_INSTRUCTIONS = 20000;\n\n/** Default for `piAgentTimeoutSec`: a tool loop with several reads can take a while. */\nexport const PI_AGENT_TIMEOUT_SEC = 60;\n\n/** The part of the adapter this module needs (kept small so tests can fake it). */\nexport interface PiAgentHost {\n    sendToAsync(instance: string, command: string, message: unknown, options?: { timeout?: number }): Promise<unknown>;\n    getForeignStateAsync(id: string): Promise<ioBroker.State | null | undefined>;\n}\n\nexport interface PiAgentRequest {\n    /** The text pi-agent answers: context lines + the question. */\n    text: string;\n    /** Conversation of the caller, e.g. the satellite name. */\n    conversation: string;\n    /** Instructions of that conversation. */\n    instructions: string;\n    /**\n     * Language the user speaks, e.g. `ru`. pi-agent asks its questions to confirm in it; older versions ask in the\n     * system language, and a satellite listening in Russian does not hear a German \"Ja\" (the STT returns nothing).\n     */\n    language?: string;\n}\n\n/**\n * Join instruction blocks, leaving out optional ones from the end until the result fits into pi-agent's\n * limit — pi-agent would cut the text in the middle, and a half device list is worse than none.\n *\n * @param required blocks that always go along (style, follow-up marker)\n * @param optional blocks in order of importance (memory before the device list)\n */\nexport function composeInstructions(required: string[], optional: string[]): { text: string; dropped: number } {\n    const blocks = [...required.filter(Boolean)];\n    const extra = optional.filter(Boolean);\n    let dropped = 0;\n    while (extra.length) {\n        const text = [...blocks, ...extra].join('\\n\\n');\n        if (text.length <= PI_AGENT_MAX_INSTRUCTIONS) {\n            return { text, dropped };\n        }\n        extra.pop();\n        dropped++;\n    }\n    return { text: blocks.join('\\n\\n').slice(0, PI_AGENT_MAX_INSTRUCTIONS), dropped };\n}\n\n/** Conversation name for a request source; pi-agent uses \"/\" as separator in its keys. */\nexport function conversationOf(source: string): string {\n    return (source || 'default').replace(/\\//g, '_').slice(0, 100);\n}\n\nexport class PiAgentHarness {\n    private seq = 0;\n\n    public constructor(\n        private readonly host: PiAgentHost,\n        /** e.g. `pi-agent.0` */\n        public readonly instance: string,\n        private readonly timeoutMs: number = PI_AGENT_TIMEOUT_SEC * 1000,\n    ) {}\n\n    /**\n     * Ask pi-agent. Fails fast when the instance is not running (otherwise the caller would wait for the whole\n     * timeout), and turns pi-agent's `{error}` into an exception so the caller can fall back to the local model.\n     */\n    public async ask(req: PiAgentRequest): Promise<string> {\n        const alive = await this.host.getForeignStateAsync(`system.adapter.${this.instance}.alive`);\n        if (!alive?.val) {\n            throw new Error(`${this.instance} is not running`);\n        }\n        // Unique per request: pi-agent answers a repeated requestId only once.\n        const requestId = `assistant-${Date.now().toString(36)}-${(++this.seq).toString(36)}`;\n        const result = (await this.host.sendToAsync(\n            this.instance,\n            'ask',\n            {\n                text: req.text,\n                conversation: conversationOf(req.conversation),\n                instructions: req.instructions,\n                ...(req.language ? { language: req.language } : {}),\n                requestId,\n            },\n            { timeout: this.timeoutMs },\n        )) as { answer?: unknown; error?: unknown } | null | undefined;\n        if (result?.error) {\n            const error = typeof result.error === 'string' ? result.error : JSON.stringify(result.error);\n            throw new Error(`${this.instance}: ${error}`);\n        }\n        const answer = typeof result?.answer === 'string' ? result.answer.trim() : '';\n        if (!answer) {\n            throw new Error(`${this.instance} gave no answer`);\n        }\n        return answer;\n    }\n}\n"]}
+{"version":3,"file":"piAgent.js","sourceRoot":"","sources":["../../src/lib/piAgent.ts"],"names":[],"mappings":";AAAA;;;;;;;;;;;;;;;;GAgBG;;;AA0CH,kDAaC;AAGD,wCAEC;AAxDD,8FAA8F;AACjF,QAAA,yBAAyB,GAAG,KAAK,CAAC;AAE/C,wFAAwF;AAC3E,QAAA,oBAAoB,GAAG,EAAE,CAAC;AA2BvC;;;;;;GAMG;AACH,SAAgB,mBAAmB,CAAC,QAAkB,EAAE,QAAkB;IACtE,MAAM,MAAM,GAAG,CAAC,GAAG,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC;IAC7C,MAAM,KAAK,GAAG,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IACvC,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,OAAO,KAAK,CAAC,MAAM,EAAE,CAAC;QAClB,MAAM,IAAI,GAAG,CAAC,GAAG,MAAM,EAAE,GAAG,KAAK,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAChD,IAAI,IAAI,CAAC,MAAM,IAAI,iCAAyB,EAAE,CAAC;YAC3C,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC;QAC7B,CAAC;QACD,KAAK,CAAC,GAAG,EAAE,CAAC;QACZ,OAAO,EAAE,CAAC;IACd,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,iCAAyB,CAAC,EAAE,OAAO,EAAE,CAAC;AACtF,CAAC;AAED,0FAA0F;AAC1F,SAAgB,cAAc,CAAC,MAAc;IACzC,OAAO,CAAC,MAAM,IAAI,SAAS,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;AACnE,CAAC;AAED,MAAa,cAAc;IAIF;IAED;IACC;IANb,GAAG,GAAG,CAAC,CAAC;IAEhB,YACqB,IAAiB;IAClC,wBAAwB;IACR,QAAgB,EACf,YAAoB,4BAAoB,GAAG,IAAI;QAH/C,SAAI,GAAJ,IAAI,CAAa;QAElB,aAAQ,GAAR,QAAQ,CAAQ;QACf,cAAS,GAAT,SAAS,CAAsC;IACjE,CAAC;IAEJ;;;OAGG;IACI,KAAK,CAAC,GAAG,CAAC,GAAmB;QAChC,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,oBAAoB,CAAC,kBAAkB,IAAI,CAAC,QAAQ,QAAQ,CAAC,CAAC;QAC5F,IAAI,CAAC,KAAK,EAAE,GAAG,EAAE,CAAC;YACd,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,CAAC,QAAQ,iBAAiB,CAAC,CAAC;QACvD,CAAC;QACD,uEAAuE;QACvE,MAAM,SAAS,GAAG,aAAa,IAAI,CAAC,GAAG,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,QAAQ,CAAC,EAAE,CAAC,EAAE,CAAC;QACtF,MAAM,MAAM,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,WAAW,CACvC,IAAI,CAAC,QAAQ,EACb,KAAK,EACL;YACI,IAAI,EAAE,GAAG,CAAC,IAAI;YACd,YAAY,EAAE,cAAc,CAAC,GAAG,CAAC,YAAY,CAAC;YAC9C,YAAY,EAAE,GAAG,CAAC,YAAY;YAC9B,GAAG,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,GAAG,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACnD,GAAG,CAAC,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrD,SAAS;SACZ,EACD,EAAE,OAAO,EAAE,IAAI,CAAC,SAAS,EAAE,CAC9B,CAA6D,CAAC;QAC/D,IAAI,MAAM,EAAE,KAAK,EAAE,CAAC;YAChB,MAAM,KAAK,GAAG,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YAC7F,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,CAAC,QAAQ,KAAK,KAAK,EAAE,CAAC,CAAC;QAClD,CAAC;QACD,MAAM,MAAM,GAAG,OAAO,MAAM,EAAE,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC9E,IAAI,CAAC,MAAM,EAAE,CAAC;YACV,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,CAAC,QAAQ,iBAAiB,CAAC,CAAC;QACvD,CAAC;QACD,OAAO,MAAM,CAAC;IAClB,CAAC;CACJ;AA5CD,wCA4CC","sourcesContent":["/**\n * The adapter `pi-agent` as the cloud tier (\"harness\"): instead of running its own tool loop, the assistant\n * hands the question to a pi-agent instance via `sendTo(<instance>, 'ask', …)`. pi-agent stores every model\n * and tool call as a checkpoint before it runs (an interrupted request continues after a restart, a write is\n * never repeated), enforces its own write permissions and confirmations, and can use the system-wide AI\n * settings of admin — none of which the built-in `LlmAgent` does.\n *\n * Each request source (satellite, `chat`, `telegram:<user>`) gets a conversation of its own in pi-agent\n * (`conversation`), so two rooms do not see each other's follow-ups. The source-independent part of the\n * prompt (style, follow-up marker, memory, device list) goes along as the `instructions` of that\n * conversation; what changes every turn (time, weather, presence) is prepended to the question.\n * pi-agent 0.3.0 and older ignore both fields and answer everything in their root conversation.\n *\n * With \"Confirmation by text in sendTo conversations\", pi-agent answers a write it may not do unasked with a\n * question ending in `[[LISTEN]]`, and the next `ask` of the conversation is the yes or no. `language` makes it\n * ask that question in the language the satellite listens in.\n */\n\nimport type { ImageInput } from './media';\n\n/** Longest instructions pi-agent accepts for a conversation; anything beyond is cut there. */\nexport const PI_AGENT_MAX_INSTRUCTIONS = 20000;\n\n/** Default for `piAgentTimeoutSec`: a tool loop with several reads can take a while. */\nexport const PI_AGENT_TIMEOUT_SEC = 60;\n\n/** The part of the adapter this module needs (kept small so tests can fake it). */\nexport interface PiAgentHost {\n    sendToAsync(instance: string, command: string, message: unknown, options?: { timeout?: number }): Promise<unknown>;\n    getForeignStateAsync(id: string): Promise<ioBroker.State | null | undefined>;\n}\n\nexport interface PiAgentRequest {\n    /** The text pi-agent answers: context lines + the question. */\n    text: string;\n    /** Conversation of the caller, e.g. the satellite name. */\n    conversation: string;\n    /** Instructions of that conversation. */\n    instructions: string;\n    /**\n     * Language the user speaks, e.g. `ru`. pi-agent asks its questions to confirm in it; older versions ask in the\n     * system language, and a satellite listening in Russian does not hear a German \"Ja\" (the STT returns nothing).\n     */\n    language?: string;\n    /**\n     * Pictures that came with the question (`{ data: base64, mimeType }`, the form of pi-ai's `ImageContent`).\n     * Versions of pi-agent before image support ignore them; the text then says that pictures came along.\n     */\n    images?: ImageInput[];\n}\n\n/**\n * Join instruction blocks, leaving out optional ones from the end until the result fits into pi-agent's\n * limit — pi-agent would cut the text in the middle, and a half device list is worse than none.\n *\n * @param required blocks that always go along (style, follow-up marker)\n * @param optional blocks in order of importance (memory before the device list)\n */\nexport function composeInstructions(required: string[], optional: string[]): { text: string; dropped: number } {\n    const blocks = [...required.filter(Boolean)];\n    const extra = optional.filter(Boolean);\n    let dropped = 0;\n    while (extra.length) {\n        const text = [...blocks, ...extra].join('\\n\\n');\n        if (text.length <= PI_AGENT_MAX_INSTRUCTIONS) {\n            return { text, dropped };\n        }\n        extra.pop();\n        dropped++;\n    }\n    return { text: blocks.join('\\n\\n').slice(0, PI_AGENT_MAX_INSTRUCTIONS), dropped };\n}\n\n/** Conversation name for a request source; pi-agent uses \"/\" as separator in its keys. */\nexport function conversationOf(source: string): string {\n    return (source || 'default').replace(/\\//g, '_').slice(0, 100);\n}\n\nexport class PiAgentHarness {\n    private seq = 0;\n\n    public constructor(\n        private readonly host: PiAgentHost,\n        /** e.g. `pi-agent.0` */\n        public readonly instance: string,\n        private readonly timeoutMs: number = PI_AGENT_TIMEOUT_SEC * 1000,\n    ) {}\n\n    /**\n     * Ask pi-agent. Fails fast when the instance is not running (otherwise the caller would wait for the whole\n     * timeout), and turns pi-agent's `{error}` into an exception so the caller can fall back to the local model.\n     */\n    public async ask(req: PiAgentRequest): Promise<string> {\n        const alive = await this.host.getForeignStateAsync(`system.adapter.${this.instance}.alive`);\n        if (!alive?.val) {\n            throw new Error(`${this.instance} is not running`);\n        }\n        // Unique per request: pi-agent answers a repeated requestId only once.\n        const requestId = `assistant-${Date.now().toString(36)}-${(++this.seq).toString(36)}`;\n        const result = (await this.host.sendToAsync(\n            this.instance,\n            'ask',\n            {\n                text: req.text,\n                conversation: conversationOf(req.conversation),\n                instructions: req.instructions,\n                ...(req.language ? { language: req.language } : {}),\n                ...(req.images?.length ? { images: req.images } : {}),\n                requestId,\n            },\n            { timeout: this.timeoutMs },\n        )) as { answer?: unknown; error?: unknown } | null | undefined;\n        if (result?.error) {\n            const error = typeof result.error === 'string' ? result.error : JSON.stringify(result.error);\n            throw new Error(`${this.instance}: ${error}`);\n        }\n        const answer = typeof result?.answer === 'string' ? result.answer.trim() : '';\n        if (!answer) {\n            throw new Error(`${this.instance} gave no answer`);\n        }\n        return answer;\n    }\n}\n"]}

package/build/lib/voice/engines.js

@@ -104,6 +104,9 @@ function buildSttEngine(provider, ctx) {
         case 'piper':
             throw new Error('Piper is a text-to-speech engine, not speech-to-text');
         default:
+            if (ctx.sttBaseUrl?.trim()) {
+                return new stt_1.OpenAiStt(ctx.sttServerKey || '', ctx.sttModel || undefined, ctx.sttBaseUrl);
+            }
             requireOpenAi(ctx.creds);
             return new stt_1.OpenAiStt(ctx.creds.openaiKey, ctx.sttModel || undefined);
     }

package/build/lib/voice/engines.js.map

@@ -1 +1 @@
-{"version":3,"file":"engines.js","sourceRoot":"","sources":["../../../src/lib/voice/engines.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA+FA,0CAYC;AAuBD,0CAeC;AAuDD,gCAmBC;AAGD,sCAKC;AAGD,sCAEC;AAxOD;;;;GAIG;AACH,+BAAkD;AAClD,+BAAoE;AACpE,mCAA8D;AAC9D,+BAAuE;AACvE,iCAAiD;AACjD,mCAAoD;AACpD,6CAOsB;AACtB,gDAAkC;AAElC,yCAAuC;AACvC,yCAAwE;AACxE,iCAAqC;AAgDrC,SAAS,YAAY,CAAC,KAAuB;IACzC,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC;QACxC,MAAM,IAAI,KAAK,CAAC,kDAAkD,CAAC,CAAC;IACxE,CAAC;AACL,CAAC;AAED,SAAS,UAAU,CAAC,KAAuB;IACvC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,eAAe,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,EAAE,CAAC;QAC5E,MAAM,IAAI,KAAK,CAAC,iEAAiE,CAAC,CAAC;IACvF,CAAC;AACL,CAAC;AAED,SAAS,aAAa,CAAC,KAAuB;IAC1C,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC;QACnB,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACtD,CAAC;AACL,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAuB;IAC9C,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,CAAC;QACvB,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAC;IAC1D,CAAC;AACL,CAAC;AAED,SAAgB,eAAe,CAAC,QAAwB,EAAE,GAAkB;IACxE,MAAM,OAAO,GAA2B,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,cAAc,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAAC;IAClG,MAAM,QAAQ,GAAG,GAAG,CAAC,WAAW,IAAI,GAAG,CAAC,WAAW,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,CAAC;IACxF,IAAI,CAAC,QAAQ,EAAE,CAAC;QACZ,OAAO,OAAO,CAAC,MAAM,CAAC;IAC1B,CAAC;IACD,IAAI,CAAC;QACD,OAAO,IAAI,sBAAW,CAAC,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,cAAc,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAAC,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;IAC1G,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACT,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,4BAA4B,QAAQ,kBAAmB,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;QAC3F,OAAO,OAAO,CAAC,MAAM,CAAC;IAC1B,CAAC;AACL,CAAC;AAED,SAAS,cAAc,CAAC,QAAwB,EAAE,GAAkB;IAChE,QAAQ,QAAQ,EAAE,CAAC;QACf,KAAK,OAAO;YACR,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACxB,OAAO,IAAI,gBAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;QACnE,KAAK,KAAK;YACN,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACtB,OAAO,IAAI,YAAM,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QACrC,KAAK,MAAM;YACP,OAAO,IAAI,cAAO,CAAC,GAAG,CAAC,OAAO,EAAE,GAAG,CAAC,SAAS,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;QAC5D,KAAK,YAAY;YACb,iBAAiB,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YAC7B,OAAO,IAAI,0BAAa,CAAC,GAAG,CAAC,KAAK,CAAC,aAAa,EAAE,GAAG,CAAC,kBAAkB,CAAC,CAAC;QAC9E,KAAK,OAAO;YACR,MAAM,IAAI,KAAK,CAAC,sDAAsD,CAAC,CAAC;QAC5E;YACI,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACzB,OAAO,IAAI,eAAS,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,EAAE,GAAG,CAAC,QAAQ,IAAI,SAAS,CAAC,CAAC;IAC7E,CAAC;AACL,CAAC;AAED,SAAgB,eAAe,CAAC,QAAwB,EAAE,GAAkB;IACxE,sGAAsG;IACtG,iGAAiG;IACjG,MAAM,OAAO,GAA2B,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,SAAS,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAAC;IAC7F,MAAM,QAAQ,GAAG,GAAG,CAAC,WAAW,IAAI,GAAG,CAAC,WAAW,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,CAAC;IACxF,IAAI,CAAC,QAAQ,EAAE,CAAC;QACZ,OAAO,OAAO,CAAC,MAAM,CAAC;IAC1B,CAAC;IACD,IAAI,CAAC;QACD,OAAO,IAAI,sBAAW,CAAC,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,SAAS,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAAC,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;IACrG,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACT,sGAAsG;QACtG,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,4BAA4B,QAAQ,kBAAmB,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;QAC3F,OAAO,OAAO,CAAC,MAAM,CAAC;IAC1B,CAAC;AACL,CAAC;AAED;;;GAGG;AACH,SAAS,SAAS,CAAC,QAAwB,EAAE,GAAkB;IAC3D,MAAM,KAAK,GACP,QAAQ,KAAK,OAAO;QAChB,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK;QAClB,CAAC,CAAC,QAAQ,KAAK,KAAK;YAClB,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG;YAChB,CAAC,CAAC,QAAQ,KAAK,OAAO;gBACpB,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK;gBAClB,CAAC,CAAC,QAAQ,KAAK,YAAY;oBACzB,CAAC,CAAC,GAAG,GAAG,CAAC,kBAAkB,IAAI,yCAA4B,IAAI,GAAG,CAAC,MAAM,CAAC,UAAU,EAAE;oBACtF,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC;IAClC,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,UAAU,EAAE,GAAG,QAAQ,IAAI,CAAC,KAAK,IAAI,SAAS,CAAC,CAAC,OAAO,CAAC,UAAU,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;IAC/G,OAAO,IAAI,oBAAS,CAAC,cAAc,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,GAAG,EAAE,QAAQ,EAAE,GAAG,CAAC,WAAW,EAAE,CAAC,CAAC;AAC1G,CAAC;AAED,SAAS,cAAc,CAAC,QAAwB,EAAE,GAAkB;IAChE,QAAQ,QAAQ,EAAE,CAAC;QACf,KAAK,OAAO;YACR,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACxB,OAAO,IAAI,gBAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,KAAK,CAAC,WAAW,EAAE,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACrF,KAAK,KAAK;YACN,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACtB,OAAO,IAAI,YAAM,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QACrD,KAAK,OAAO;YACR,OAAO,IAAI,gBAAQ,CAAC,GAAG,CAAC,OAAO,EAAE,GAAG,CAAC,MAAM,CAAC,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;QAChE,KAAK,MAAM;YACP,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;QAC3E,KAAK,YAAY;YACb,iBAAiB,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YAC7B,OAAO,IAAI,0BAAa,CACpB,GAAG,CAAC,KAAK,CAAC,aAAa,EACvB,GAAG,CAAC,kBAAkB,EACtB,GAAG,CAAC,MAAM,CAAC,UAAU,EACrB,GAAG,CAAC,WAAW,CAClB,CAAC;QACN;YACI,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACzB,OAAO,IAAI,eAAS,CAChB,GAAG,CAAC,KAAK,CAAC,SAAS,EACnB,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,IAAI,OAAO,CAAgB,EAC7C,GAAG,CAAC,QAAQ,IAAI,SAAS,CAC5B,CAAC;IACV,CAAC;AACL,CAAC;AAED,uDAAuD;AACvD,MAAM,aAAa,GAAG,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC;AAE5E,sFAAsF;AAC/E,KAAK,UAAU,UAAU,CAAC,QAAwB,EAAE,GAAkB,EAAE,QAAgB;IAC3F,QAAQ,QAAQ,EAAE,CAAC;QACf,KAAK,OAAO;YACR,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACxB,OAAO,IAAA,uBAAe,EAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,KAAK,CAAC,WAAW,EAAE,IAAA,kBAAW,EAAC,QAAQ,CAAC,CAAC,CAAC;QAC7F,KAAK,KAAK;YACN,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACtB,OAAO,IAAA,qBAAe,EAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,IAAA,kBAAW,EAAC,QAAQ,CAAC,CAAC,CAAC;QACjE,KAAK,OAAO;YACR,OAAO,IAAA,uBAAe,GAAE,CAAC;QAC7B,KAAK,YAAY,CAAC,CAAC,CAAC;YAChB,6DAA6D;YAC7D,MAAM,KAAK,GAAG,GAAG,CAAC,kBAAkB,IAAI,yCAA4B,CAAC;YACrE,MAAM,MAAM,GAAG,MAAM,IAAA,uCAA0B,EAAC,QAAQ,CAAC,CAAC;YAC1D,OAAO,IAAA,8BAAiB,EAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,KAAK,CAAC,EAAE,MAAM,IAAI,EAAE,EAAE,QAAQ,CAAC,CAAC;QACvF,CAAC;QACD;YACI,OAAO,aAAa,CAAC;IAC7B,CAAC;AACL,CAAC;AAED,kHAAkH;AAC3G,KAAK,UAAU,aAAa,CAAC,QAAwB,EAAE,QAAgB;IAC1E,IAAI,QAAQ,KAAK,YAAY,EAAE,CAAC;QAC5B,OAAO,CAAC,MAAM,IAAA,uCAA0B,EAAC,eAAe,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAC9E,CAAC;IACD,OAAO,QAAQ,KAAK,MAAM,CAAC,CAAC,CAAC,IAAA,qBAAc,EAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AAC/D,CAAC;AAED,8FAA8F;AACvF,KAAK,UAAU,aAAa,CAAC,QAAwB;IACxD,OAAO,QAAQ,KAAK,YAAY,CAAC,CAAC,CAAC,CAAC,MAAM,IAAA,uCAA0B,EAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AACxG,CAAC","sourcesContent":["/**\n * Factory that builds the configured STT/TTS engine. Cloud engines (OpenAI/Azure/AWS) use resolved\n * credentials; local engines (Vosk/Piper) download their models on demand into the data dir.\n * Adding a provider = one new engine class + one `case` here.\n */\nimport { OpenAiStt, type SttEngine } from './stt';\nimport { OpenAiTts, type OpenAiVoice, type TtsEngine } from './tts';\nimport { AzureStt, AzureTts, listAzureVoices } from './azure';\nimport { AwsStt, AwsTts, listPollyVoices, type AwsCreds } from './aws';\nimport { VoskStt, listVoskModels } from './vosk';\nimport { PiperTts, listPiperVoices } from './piper';\nimport {\n    OpenRouterStt,\n    OpenRouterTts,\n    listOpenRouterSpeechModels,\n    voicesForLanguage,\n    DEFAULT_OPENROUTER_TTS_MODEL,\n    type AudioDecoder,\n} from './openrouter';\nimport * as path from 'node:path';\n\nimport { CachedTts } from './ttsCache';\nimport { FallbackStt, FallbackTts, type NamedEngine } from './fallback';\nimport { isoToLocale } from './lang';\nimport type { VoiceLogger } from './download';\n\nexport type SpeechProvider = 'openai' | 'openrouter' | 'azure' | 'aws' | 'vosk' | 'piper';\n\n/** All speech credentials, resolved from config (or the central credential store). */\nexport interface VoiceCredentials {\n    openaiKey: string;\n    openrouterKey: string;\n    azureKey: string;\n    azureRegion: string;\n    aws: AwsCreds;\n}\n\n/** Provider-specific voice ids (each provider names voices differently). */\nexport interface VoiceNames {\n    openai: string; // alloy/echo/…\n    azure: string; // e.g. de-DE-KatjaNeural\n    aws: string; // e.g. Vicki\n    piper: string; // e.g. de_DE-thorsten-medium\n    openrouter: string; // depends on the model, e.g. Kore, george\n}\n\n/** Everything the factory needs to build any engine. */\nexport interface EngineContext {\n    creds: VoiceCredentials;\n    voices: VoiceNames;\n    /** Optional second engine, used when the primary one fails ('' = none). */\n    sttFallback?: SpeechProvider | '';\n    ttsFallback?: SpeechProvider | '';\n    /** Writable instance data dir (for the local engines' downloads). */\n    dataDir: string;\n    log: VoiceLogger;\n    /** Local model overrides ('' → default per language). */\n    voskModel: string;\n    /** OpenAI STT model ('' → 'whisper-1'; e.g. 'gpt-4o-transcribe'). */\n    sttModel: string;\n    /** OpenAI TTS model ('' → 'tts-1'; e.g. 'gpt-4o-mini-tts'). */\n    ttsModel: string;\n    /** OpenRouter models ('' → the defaults of openrouter.ts) */\n    openrouterSttModel: string;\n    openrouterTtsModel: string;\n    /** Decoder for compressed speech (OpenRouter models that only return mp3); absent → those models fail */\n    decodeAudio?: AudioDecoder;\n    /** Longest text that is spoken, in characters; longer ones are cut at a sentence end. 0 = no limit, absent = 400. */\n    ttsMaxChars?: number;\n}\n\nfunction requireAzure(creds: VoiceCredentials): void {\n    if (!creds.azureKey || !creds.azureRegion) {\n        throw new Error('Azure speech needs a subscription key and region');\n    }\n}\n\nfunction requireAws(creds: VoiceCredentials): void {\n    if (!creds.aws.accessKeyId || !creds.aws.secretAccessKey || !creds.aws.region) {\n        throw new Error('AWS speech needs an access key id, secret access key and region');\n    }\n}\n\nfunction requireOpenAi(creds: VoiceCredentials): void {\n    if (!creds.openaiKey) {\n        throw new Error('OpenAI speech needs an API key');\n    }\n}\n\nfunction requireOpenRouter(creds: VoiceCredentials): void {\n    if (!creds.openrouterKey) {\n        throw new Error('OpenRouter speech needs an API key');\n    }\n}\n\nexport function createSttEngine(provider: SpeechProvider, ctx: EngineContext): SttEngine {\n    const primary: NamedEngine<SttEngine> = { name: provider, engine: buildSttEngine(provider, ctx) };\n    const fallback = ctx.sttFallback && ctx.sttFallback !== provider ? ctx.sttFallback : '';\n    if (!fallback) {\n        return primary.engine;\n    }\n    try {\n        return new FallbackStt([primary, { name: fallback, engine: buildSttEngine(fallback, ctx) }], ctx.log);\n    } catch (e) {\n        ctx.log.warn(`Speech-to-text fallback '${fallback}' unavailable: ${(e as Error).message}`);\n        return primary.engine;\n    }\n}\n\nfunction buildSttEngine(provider: SpeechProvider, ctx: EngineContext): SttEngine {\n    switch (provider) {\n        case 'azure':\n            requireAzure(ctx.creds);\n            return new AzureStt(ctx.creds.azureKey, ctx.creds.azureRegion);\n        case 'aws':\n            requireAws(ctx.creds);\n            return new AwsStt(ctx.creds.aws);\n        case 'vosk':\n            return new VoskStt(ctx.dataDir, ctx.voskModel, ctx.log);\n        case 'openrouter':\n            requireOpenRouter(ctx.creds);\n            return new OpenRouterStt(ctx.creds.openrouterKey, ctx.openrouterSttModel);\n        case 'piper':\n            throw new Error('Piper is a text-to-speech engine, not speech-to-text');\n        default:\n            requireOpenAi(ctx.creds);\n            return new OpenAiStt(ctx.creds.openaiKey, ctx.sttModel || undefined);\n    }\n}\n\nexport function createTtsEngine(provider: SpeechProvider, ctx: EngineContext): TtsEngine {\n    // Each engine gets the shared behaviour (disk cache, length limit, SSML) and its own cache directory,\n    // so a fallback engine's voice can never be served later while the primary one is healthy again.\n    const primary: NamedEngine<TtsEngine> = { name: provider, engine: cachedTts(provider, ctx) };\n    const fallback = ctx.ttsFallback && ctx.ttsFallback !== provider ? ctx.ttsFallback : '';\n    if (!fallback) {\n        return primary.engine;\n    }\n    try {\n        return new FallbackTts([primary, { name: fallback, engine: cachedTts(fallback, ctx) }], ctx.log);\n    } catch (e) {\n        // A fallback that cannot even be built (missing key, wrong provider) must not take the main one down.\n        ctx.log.warn(`Text-to-speech fallback '${fallback}' unavailable: ${(e as Error).message}`);\n        return primary.engine;\n    }\n}\n\n/**\n * One provider binding plus the shared cache/length/SSML layer. One cache directory per provider+voice —\n * a voice change must not serve audio in the old voice.\n */\nfunction cachedTts(provider: SpeechProvider, ctx: EngineContext): TtsEngine {\n    const voice =\n        provider === 'azure'\n            ? ctx.voices.azure\n            : provider === 'aws'\n              ? ctx.voices.aws\n              : provider === 'piper'\n                ? ctx.voices.piper\n                : provider === 'openrouter'\n                  ? `${ctx.openrouterTtsModel || DEFAULT_OPENROUTER_TTS_MODEL}-${ctx.voices.openrouter}`\n                  : ctx.voices.openai;\n    const dir = path.join(ctx.dataDir, 'ttscache', `${provider}-${(voice || 'default').replace(/[^\\w.-]/g, '_')}`);\n    return new CachedTts(buildTtsEngine(provider, ctx), { dir, log: ctx.log, maxChars: ctx.ttsMaxChars });\n}\n\nfunction buildTtsEngine(provider: SpeechProvider, ctx: EngineContext): TtsEngine {\n    switch (provider) {\n        case 'azure':\n            requireAzure(ctx.creds);\n            return new AzureTts(ctx.creds.azureKey, ctx.creds.azureRegion, ctx.voices.azure);\n        case 'aws':\n            requireAws(ctx.creds);\n            return new AwsTts(ctx.creds.aws, ctx.voices.aws);\n        case 'piper':\n            return new PiperTts(ctx.dataDir, ctx.voices.piper, ctx.log);\n        case 'vosk':\n            throw new Error('Vosk is a speech-to-text engine, not text-to-speech');\n        case 'openrouter':\n            requireOpenRouter(ctx.creds);\n            return new OpenRouterTts(\n                ctx.creds.openrouterKey,\n                ctx.openrouterTtsModel,\n                ctx.voices.openrouter,\n                ctx.decodeAudio,\n            );\n        default:\n            requireOpenAi(ctx.creds);\n            return new OpenAiTts(\n                ctx.creds.openaiKey,\n                (ctx.voices.openai || 'alloy') as OpenAiVoice,\n                ctx.ttsModel || undefined,\n            );\n    }\n}\n\n/** OpenAI TTS voices are a fixed set (no list API). */\nconst OPENAI_VOICES = ['alloy', 'echo', 'fable', 'onyx', 'nova', 'shimmer'];\n\n/** List the available TTS voices for a provider (for the settings voice dropdown). */\nexport async function listVoices(provider: SpeechProvider, ctx: EngineContext, language: string): Promise<string[]> {\n    switch (provider) {\n        case 'azure':\n            requireAzure(ctx.creds);\n            return listAzureVoices(ctx.creds.azureKey, ctx.creds.azureRegion, isoToLocale(language));\n        case 'aws':\n            requireAws(ctx.creds);\n            return listPollyVoices(ctx.creds.aws, isoToLocale(language));\n        case 'piper':\n            return listPiperVoices();\n        case 'openrouter': {\n            // public list, no key needed: the voices of the chosen model\n            const model = ctx.openrouterTtsModel || DEFAULT_OPENROUTER_TTS_MODEL;\n            const models = await listOpenRouterSpeechModels('speech');\n            return voicesForLanguage(models.find(m => m.id === model)?.voices || [], language);\n        }\n        default:\n            return OPENAI_VOICES;\n    }\n}\n\n/** Suggested STT model names for the settings dropdown: the local Vosk catalogue, or the models of OpenRouter. */\nexport async function listSttModels(provider: SpeechProvider, language: string): Promise<string[]> {\n    if (provider === 'openrouter') {\n        return (await listOpenRouterSpeechModels('transcription')).map(m => m.id);\n    }\n    return provider === 'vosk' ? listVoskModels(language) : [];\n}\n\n/** TTS model names for the settings dropdown — only OpenRouter has a choice worth listing. */\nexport async function listTtsModels(provider: SpeechProvider): Promise<string[]> {\n    return provider === 'openrouter' ? (await listOpenRouterSpeechModels('speech')).map(m => m.id) : [];\n}\n"]}
+{"version":3,"file":"engines.js","sourceRoot":"","sources":["../../../src/lib/voice/engines.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAsGA,0CAYC;AA0BD,0CAeC;AAuDD,gCAmBC;AAGD,sCAKC;AAGD,sCAEC;AAlPD;;;;GAIG;AACH,+BAAkD;AAClD,+BAAoE;AACpE,mCAA8D;AAC9D,+BAAuE;AACvE,iCAAiD;AACjD,mCAAoD;AACpD,6CAOsB;AACtB,gDAAkC;AAElC,yCAAuC;AACvC,yCAAwE;AACxE,iCAAqC;AAuDrC,SAAS,YAAY,CAAC,KAAuB;IACzC,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC;QACxC,MAAM,IAAI,KAAK,CAAC,kDAAkD,CAAC,CAAC;IACxE,CAAC;AACL,CAAC;AAED,SAAS,UAAU,CAAC,KAAuB;IACvC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,eAAe,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,EAAE,CAAC;QAC5E,MAAM,IAAI,KAAK,CAAC,iEAAiE,CAAC,CAAC;IACvF,CAAC;AACL,CAAC;AAED,SAAS,aAAa,CAAC,KAAuB;IAC1C,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC;QACnB,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACtD,CAAC;AACL,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAuB;IAC9C,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,CAAC;QACvB,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAC;IAC1D,CAAC;AACL,CAAC;AAED,SAAgB,eAAe,CAAC,QAAwB,EAAE,GAAkB;IACxE,MAAM,OAAO,GAA2B,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,cAAc,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAAC;IAClG,MAAM,QAAQ,GAAG,GAAG,CAAC,WAAW,IAAI,GAAG,CAAC,WAAW,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,CAAC;IACxF,IAAI,CAAC,QAAQ,EAAE,CAAC;QACZ,OAAO,OAAO,CAAC,MAAM,CAAC;IAC1B,CAAC;IACD,IAAI,CAAC;QACD,OAAO,IAAI,sBAAW,CAAC,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,cAAc,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAAC,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;IAC1G,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACT,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,4BAA4B,QAAQ,kBAAmB,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;QAC3F,OAAO,OAAO,CAAC,MAAM,CAAC;IAC1B,CAAC;AACL,CAAC;AAED,SAAS,cAAc,CAAC,QAAwB,EAAE,GAAkB;IAChE,QAAQ,QAAQ,EAAE,CAAC;QACf,KAAK,OAAO;YACR,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACxB,OAAO,IAAI,gBAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;QACnE,KAAK,KAAK;YACN,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACtB,OAAO,IAAI,YAAM,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QACrC,KAAK,MAAM;YACP,OAAO,IAAI,cAAO,CAAC,GAAG,CAAC,OAAO,EAAE,GAAG,CAAC,SAAS,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;QAC5D,KAAK,YAAY;YACb,iBAAiB,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YAC7B,OAAO,IAAI,0BAAa,CAAC,GAAG,CAAC,KAAK,CAAC,aAAa,EAAE,GAAG,CAAC,kBAAkB,CAAC,CAAC;QAC9E,KAAK,OAAO;YACR,MAAM,IAAI,KAAK,CAAC,sDAAsD,CAAC,CAAC;QAC5E;YACI,IAAI,GAAG,CAAC,UAAU,EAAE,IAAI,EAAE,EAAE,CAAC;gBACzB,OAAO,IAAI,eAAS,CAAC,GAAG,CAAC,YAAY,IAAI,EAAE,EAAE,GAAG,CAAC,QAAQ,IAAI,SAAS,EAAE,GAAG,CAAC,UAAU,CAAC,CAAC;YAC5F,CAAC;YACD,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACzB,OAAO,IAAI,eAAS,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,EAAE,GAAG,CAAC,QAAQ,IAAI,SAAS,CAAC,CAAC;IAC7E,CAAC;AACL,CAAC;AAED,SAAgB,eAAe,CAAC,QAAwB,EAAE,GAAkB;IACxE,sGAAsG;IACtG,iGAAiG;IACjG,MAAM,OAAO,GAA2B,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,SAAS,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAAC;IAC7F,MAAM,QAAQ,GAAG,GAAG,CAAC,WAAW,IAAI,GAAG,CAAC,WAAW,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,CAAC;IACxF,IAAI,CAAC,QAAQ,EAAE,CAAC;QACZ,OAAO,OAAO,CAAC,MAAM,CAAC;IAC1B,CAAC;IACD,IAAI,CAAC;QACD,OAAO,IAAI,sBAAW,CAAC,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,SAAS,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAAC,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;IACrG,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACT,sGAAsG;QACtG,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,4BAA4B,QAAQ,kBAAmB,CAAW,CAAC,OAAO,EAAE,CAAC,CAAC;QAC3F,OAAO,OAAO,CAAC,MAAM,CAAC;IAC1B,CAAC;AACL,CAAC;AAED;;;GAGG;AACH,SAAS,SAAS,CAAC,QAAwB,EAAE,GAAkB;IAC3D,MAAM,KAAK,GACP,QAAQ,KAAK,OAAO;QAChB,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK;QAClB,CAAC,CAAC,QAAQ,KAAK,KAAK;YAClB,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG;YAChB,CAAC,CAAC,QAAQ,KAAK,OAAO;gBACpB,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK;gBAClB,CAAC,CAAC,QAAQ,KAAK,YAAY;oBACzB,CAAC,CAAC,GAAG,GAAG,CAAC,kBAAkB,IAAI,yCAA4B,IAAI,GAAG,CAAC,MAAM,CAAC,UAAU,EAAE;oBACtF,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC;IAClC,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,UAAU,EAAE,GAAG,QAAQ,IAAI,CAAC,KAAK,IAAI,SAAS,CAAC,CAAC,OAAO,CAAC,UAAU,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;IAC/G,OAAO,IAAI,oBAAS,CAAC,cAAc,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,GAAG,EAAE,QAAQ,EAAE,GAAG,CAAC,WAAW,EAAE,CAAC,CAAC;AAC1G,CAAC;AAED,SAAS,cAAc,CAAC,QAAwB,EAAE,GAAkB;IAChE,QAAQ,QAAQ,EAAE,CAAC;QACf,KAAK,OAAO;YACR,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACxB,OAAO,IAAI,gBAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,KAAK,CAAC,WAAW,EAAE,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACrF,KAAK,KAAK;YACN,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACtB,OAAO,IAAI,YAAM,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QACrD,KAAK,OAAO;YACR,OAAO,IAAI,gBAAQ,CAAC,GAAG,CAAC,OAAO,EAAE,GAAG,CAAC,MAAM,CAAC,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;QAChE,KAAK,MAAM;YACP,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;QAC3E,KAAK,YAAY;YACb,iBAAiB,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YAC7B,OAAO,IAAI,0BAAa,CACpB,GAAG,CAAC,KAAK,CAAC,aAAa,EACvB,GAAG,CAAC,kBAAkB,EACtB,GAAG,CAAC,MAAM,CAAC,UAAU,EACrB,GAAG,CAAC,WAAW,CAClB,CAAC;QACN;YACI,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACzB,OAAO,IAAI,eAAS,CAChB,GAAG,CAAC,KAAK,CAAC,SAAS,EACnB,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,IAAI,OAAO,CAAgB,EAC7C,GAAG,CAAC,QAAQ,IAAI,SAAS,CAC5B,CAAC;IACV,CAAC;AACL,CAAC;AAED,uDAAuD;AACvD,MAAM,aAAa,GAAG,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC;AAE5E,sFAAsF;AAC/E,KAAK,UAAU,UAAU,CAAC,QAAwB,EAAE,GAAkB,EAAE,QAAgB;IAC3F,QAAQ,QAAQ,EAAE,CAAC;QACf,KAAK,OAAO;YACR,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACxB,OAAO,IAAA,uBAAe,EAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,KAAK,CAAC,WAAW,EAAE,IAAA,kBAAW,EAAC,QAAQ,CAAC,CAAC,CAAC;QAC7F,KAAK,KAAK;YACN,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;YACtB,OAAO,IAAA,qBAAe,EAAC,GAAG,CAAC,KAAK,CAAC,GAAG,EAAE,IAAA,kBAAW,EAAC,QAAQ,CAAC,CAAC,CAAC;QACjE,KAAK,OAAO;YACR,OAAO,IAAA,uBAAe,GAAE,CAAC;QAC7B,KAAK,YAAY,CAAC,CAAC,CAAC;YAChB,6DAA6D;YAC7D,MAAM,KAAK,GAAG,GAAG,CAAC,kBAAkB,IAAI,yCAA4B,CAAC;YACrE,MAAM,MAAM,GAAG,MAAM,IAAA,uCAA0B,EAAC,QAAQ,CAAC,CAAC;YAC1D,OAAO,IAAA,8BAAiB,EAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,KAAK,CAAC,EAAE,MAAM,IAAI,EAAE,EAAE,QAAQ,CAAC,CAAC;QACvF,CAAC;QACD;YACI,OAAO,aAAa,CAAC;IAC7B,CAAC;AACL,CAAC;AAED,kHAAkH;AAC3G,KAAK,UAAU,aAAa,CAAC,QAAwB,EAAE,QAAgB;IAC1E,IAAI,QAAQ,KAAK,YAAY,EAAE,CAAC;QAC5B,OAAO,CAAC,MAAM,IAAA,uCAA0B,EAAC,eAAe,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAC9E,CAAC;IACD,OAAO,QAAQ,KAAK,MAAM,CAAC,CAAC,CAAC,IAAA,qBAAc,EAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AAC/D,CAAC;AAED,8FAA8F;AACvF,KAAK,UAAU,aAAa,CAAC,QAAwB;IACxD,OAAO,QAAQ,KAAK,YAAY,CAAC,CAAC,CAAC,CAAC,MAAM,IAAA,uCAA0B,EAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AACxG,CAAC","sourcesContent":["/**\n * Factory that builds the configured STT/TTS engine. Cloud engines (OpenAI/Azure/AWS) use resolved\n * credentials; local engines (Vosk/Piper) download their models on demand into the data dir.\n * Adding a provider = one new engine class + one `case` here.\n */\nimport { OpenAiStt, type SttEngine } from './stt';\nimport { OpenAiTts, type OpenAiVoice, type TtsEngine } from './tts';\nimport { AzureStt, AzureTts, listAzureVoices } from './azure';\nimport { AwsStt, AwsTts, listPollyVoices, type AwsCreds } from './aws';\nimport { VoskStt, listVoskModels } from './vosk';\nimport { PiperTts, listPiperVoices } from './piper';\nimport {\n    OpenRouterStt,\n    OpenRouterTts,\n    listOpenRouterSpeechModels,\n    voicesForLanguage,\n    DEFAULT_OPENROUTER_TTS_MODEL,\n    type AudioDecoder,\n} from './openrouter';\nimport * as path from 'node:path';\n\nimport { CachedTts } from './ttsCache';\nimport { FallbackStt, FallbackTts, type NamedEngine } from './fallback';\nimport { isoToLocale } from './lang';\nimport type { VoiceLogger } from './download';\n\nexport type SpeechProvider = 'openai' | 'openrouter' | 'azure' | 'aws' | 'vosk' | 'piper';\n\n/** All speech credentials, resolved from config (or the central credential store). */\nexport interface VoiceCredentials {\n    openaiKey: string;\n    openrouterKey: string;\n    azureKey: string;\n    azureRegion: string;\n    aws: AwsCreds;\n}\n\n/** Provider-specific voice ids (each provider names voices differently). */\nexport interface VoiceNames {\n    openai: string; // alloy/echo/…\n    azure: string; // e.g. de-DE-KatjaNeural\n    aws: string; // e.g. Vicki\n    piper: string; // e.g. de_DE-thorsten-medium\n    openrouter: string; // depends on the model, e.g. Kore, george\n}\n\n/** Everything the factory needs to build any engine. */\nexport interface EngineContext {\n    creds: VoiceCredentials;\n    voices: VoiceNames;\n    /** Optional second engine, used when the primary one fails ('' = none). */\n    sttFallback?: SpeechProvider | '';\n    ttsFallback?: SpeechProvider | '';\n    /** Writable instance data dir (for the local engines' downloads). */\n    dataDir: string;\n    log: VoiceLogger;\n    /** Local model overrides ('' → default per language). */\n    voskModel: string;\n    /** OpenAI STT model ('' → 'whisper-1'; e.g. 'gpt-4o-transcribe'). */\n    sttModel: string;\n    /**\n     * OpenAI-compatible transcription server instead of OpenAI ('' = OpenAI), e.g. a local whisper container.\n     * It gets `sttServerKey`, never the OpenAI key — that one is only for OpenAI.\n     */\n    sttBaseUrl?: string;\n    /** Key for `sttBaseUrl` ('' = the server needs none). */\n    sttServerKey?: string;\n    /** OpenAI TTS model ('' → 'tts-1'; e.g. 'gpt-4o-mini-tts'). */\n    ttsModel: string;\n    /** OpenRouter models ('' → the defaults of openrouter.ts) */\n    openrouterSttModel: string;\n    openrouterTtsModel: string;\n    /** Decoder for compressed speech (OpenRouter models that only return mp3); absent → those models fail */\n    decodeAudio?: AudioDecoder;\n    /** Longest text that is spoken, in characters; longer ones are cut at a sentence end. 0 = no limit, absent = 400. */\n    ttsMaxChars?: number;\n}\n\nfunction requireAzure(creds: VoiceCredentials): void {\n    if (!creds.azureKey || !creds.azureRegion) {\n        throw new Error('Azure speech needs a subscription key and region');\n    }\n}\n\nfunction requireAws(creds: VoiceCredentials): void {\n    if (!creds.aws.accessKeyId || !creds.aws.secretAccessKey || !creds.aws.region) {\n        throw new Error('AWS speech needs an access key id, secret access key and region');\n    }\n}\n\nfunction requireOpenAi(creds: VoiceCredentials): void {\n    if (!creds.openaiKey) {\n        throw new Error('OpenAI speech needs an API key');\n    }\n}\n\nfunction requireOpenRouter(creds: VoiceCredentials): void {\n    if (!creds.openrouterKey) {\n        throw new Error('OpenRouter speech needs an API key');\n    }\n}\n\nexport function createSttEngine(provider: SpeechProvider, ctx: EngineContext): SttEngine {\n    const primary: NamedEngine<SttEngine> = { name: provider, engine: buildSttEngine(provider, ctx) };\n    const fallback = ctx.sttFallback && ctx.sttFallback !== provider ? ctx.sttFallback : '';\n    if (!fallback) {\n        return primary.engine;\n    }\n    try {\n        return new FallbackStt([primary, { name: fallback, engine: buildSttEngine(fallback, ctx) }], ctx.log);\n    } catch (e) {\n        ctx.log.warn(`Speech-to-text fallback '${fallback}' unavailable: ${(e as Error).message}`);\n        return primary.engine;\n    }\n}\n\nfunction buildSttEngine(provider: SpeechProvider, ctx: EngineContext): SttEngine {\n    switch (provider) {\n        case 'azure':\n            requireAzure(ctx.creds);\n            return new AzureStt(ctx.creds.azureKey, ctx.creds.azureRegion);\n        case 'aws':\n            requireAws(ctx.creds);\n            return new AwsStt(ctx.creds.aws);\n        case 'vosk':\n            return new VoskStt(ctx.dataDir, ctx.voskModel, ctx.log);\n        case 'openrouter':\n            requireOpenRouter(ctx.creds);\n            return new OpenRouterStt(ctx.creds.openrouterKey, ctx.openrouterSttModel);\n        case 'piper':\n            throw new Error('Piper is a text-to-speech engine, not speech-to-text');\n        default:\n            if (ctx.sttBaseUrl?.trim()) {\n                return new OpenAiStt(ctx.sttServerKey || '', ctx.sttModel || undefined, ctx.sttBaseUrl);\n            }\n            requireOpenAi(ctx.creds);\n            return new OpenAiStt(ctx.creds.openaiKey, ctx.sttModel || undefined);\n    }\n}\n\nexport function createTtsEngine(provider: SpeechProvider, ctx: EngineContext): TtsEngine {\n    // Each engine gets the shared behaviour (disk cache, length limit, SSML) and its own cache directory,\n    // so a fallback engine's voice can never be served later while the primary one is healthy again.\n    const primary: NamedEngine<TtsEngine> = { name: provider, engine: cachedTts(provider, ctx) };\n    const fallback = ctx.ttsFallback && ctx.ttsFallback !== provider ? ctx.ttsFallback : '';\n    if (!fallback) {\n        return primary.engine;\n    }\n    try {\n        return new FallbackTts([primary, { name: fallback, engine: cachedTts(fallback, ctx) }], ctx.log);\n    } catch (e) {\n        // A fallback that cannot even be built (missing key, wrong provider) must not take the main one down.\n        ctx.log.warn(`Text-to-speech fallback '${fallback}' unavailable: ${(e as Error).message}`);\n        return primary.engine;\n    }\n}\n\n/**\n * One provider binding plus the shared cache/length/SSML layer. One cache directory per provider+voice —\n * a voice change must not serve audio in the old voice.\n */\nfunction cachedTts(provider: SpeechProvider, ctx: EngineContext): TtsEngine {\n    const voice =\n        provider === 'azure'\n            ? ctx.voices.azure\n            : provider === 'aws'\n              ? ctx.voices.aws\n              : provider === 'piper'\n                ? ctx.voices.piper\n                : provider === 'openrouter'\n                  ? `${ctx.openrouterTtsModel || DEFAULT_OPENROUTER_TTS_MODEL}-${ctx.voices.openrouter}`\n                  : ctx.voices.openai;\n    const dir = path.join(ctx.dataDir, 'ttscache', `${provider}-${(voice || 'default').replace(/[^\\w.-]/g, '_')}`);\n    return new CachedTts(buildTtsEngine(provider, ctx), { dir, log: ctx.log, maxChars: ctx.ttsMaxChars });\n}\n\nfunction buildTtsEngine(provider: SpeechProvider, ctx: EngineContext): TtsEngine {\n    switch (provider) {\n        case 'azure':\n            requireAzure(ctx.creds);\n            return new AzureTts(ctx.creds.azureKey, ctx.creds.azureRegion, ctx.voices.azure);\n        case 'aws':\n            requireAws(ctx.creds);\n            return new AwsTts(ctx.creds.aws, ctx.voices.aws);\n        case 'piper':\n            return new PiperTts(ctx.dataDir, ctx.voices.piper, ctx.log);\n        case 'vosk':\n            throw new Error('Vosk is a speech-to-text engine, not text-to-speech');\n        case 'openrouter':\n            requireOpenRouter(ctx.creds);\n            return new OpenRouterTts(\n                ctx.creds.openrouterKey,\n                ctx.openrouterTtsModel,\n                ctx.voices.openrouter,\n                ctx.decodeAudio,\n            );\n        default:\n            requireOpenAi(ctx.creds);\n            return new OpenAiTts(\n                ctx.creds.openaiKey,\n                (ctx.voices.openai || 'alloy') as OpenAiVoice,\n                ctx.ttsModel || undefined,\n            );\n    }\n}\n\n/** OpenAI TTS voices are a fixed set (no list API). */\nconst OPENAI_VOICES = ['alloy', 'echo', 'fable', 'onyx', 'nova', 'shimmer'];\n\n/** List the available TTS voices for a provider (for the settings voice dropdown). */\nexport async function listVoices(provider: SpeechProvider, ctx: EngineContext, language: string): Promise<string[]> {\n    switch (provider) {\n        case 'azure':\n            requireAzure(ctx.creds);\n            return listAzureVoices(ctx.creds.azureKey, ctx.creds.azureRegion, isoToLocale(language));\n        case 'aws':\n            requireAws(ctx.creds);\n            return listPollyVoices(ctx.creds.aws, isoToLocale(language));\n        case 'piper':\n            return listPiperVoices();\n        case 'openrouter': {\n            // public list, no key needed: the voices of the chosen model\n            const model = ctx.openrouterTtsModel || DEFAULT_OPENROUTER_TTS_MODEL;\n            const models = await listOpenRouterSpeechModels('speech');\n            return voicesForLanguage(models.find(m => m.id === model)?.voices || [], language);\n        }\n        default:\n            return OPENAI_VOICES;\n    }\n}\n\n/** Suggested STT model names for the settings dropdown: the local Vosk catalogue, or the models of OpenRouter. */\nexport async function listSttModels(provider: SpeechProvider, language: string): Promise<string[]> {\n    if (provider === 'openrouter') {\n        return (await listOpenRouterSpeechModels('transcription')).map(m => m.id);\n    }\n    return provider === 'vosk' ? listVoskModels(language) : [];\n}\n\n/** TTS model names for the settings dropdown — only OpenRouter has a choice worth listing. */\nexport async function listTtsModels(provider: SpeechProvider): Promise<string[]> {\n    return provider === 'openrouter' ? (await listOpenRouterSpeechModels('speech')).map(m => m.id) : [];\n}\n"]}