@juspay/neurolink 9.88.3 → 9.88.5

package/dist/lib/processors/document/PptxProcessor.d.ts

@@ -42,6 +42,20 @@ export declare class PptxProcessor {
      * @throws Error if the buffer is not a valid ZIP/PPTX file
      */
     static extractText(content: Buffer): Promise<string | null>;
+    /**
+     * Extract the speaker notes for one slide.
+     *
+     * Speaker notes live in a separate `ppt/notesSlides/notesSlideN.xml` part
+     * whose number is NOT tied to the slide number — the link is declared in the
+     * slide's own relationships file (`ppt/slides/_rels/slideN.xml.rels`) via a
+     * relationship of type `.../notesSlide`. We resolve that target, then pull the
+     * `<a:t>` runs from the notes part. Returns null when the slide has no notes.
+     *
+     * @param zip - The opened PPTX archive
+     * @param slideNumber - 1-indexed slide number
+     * @returns The notes text (runs joined by a space), or null when absent
+     */
+    private static extractNotesForSlide;
     /**
      * Extract text strings from a slide XML document.
      * Finds all `<a:t>` elements and returns their text content.

package/dist/lib/processors/document/PptxProcessor.js

@@ -37,6 +37,14 @@ const TEXT_ELEMENT_REGEX = /<a:t[^>]*>([\s\S]*?)<\/a:t>/g;
  * Matches entries like "ppt/slides/slide1.xml", "ppt/slides/slide12.xml".
  */
 const SLIDE_ENTRY_REGEX = /^ppt\/slides\/slide(\d+)\.xml$/;
+/**
+ * A slide's per-slide relationships file, e.g.
+ * "ppt/slides/_rels/slide1.xml.rels", which links the slide to its speaker
+ * notes part (among other relationships).
+ */
+const SLIDE_RELS_NAME = (n) => `ppt/slides/_rels/slide${n}.xml.rels`;
+/** Matches a single `<Relationship .../>` tag inside a .rels document. */
+const RELATIONSHIP_TAG_REGEX = /<Relationship\b[^>]*\/?>/g;
 /**
  * Static utility class for extracting text from PPTX files.
  *
@@ -74,15 +82,65 @@ export class PptxProcessor {
         parts.push(`Presentation: ${slides.length} slide(s)\n`);
         for (const slide of slides) {
             const texts = PptxProcessor.extractTextFromXml(slide.xml);
-            if (texts.length > 0) {
+            const notes = PptxProcessor.extractNotesForSlide(zip, slide.slideNumber);
+            // Emit a slide section when it has either body text or speaker notes.
+            if (texts.length > 0 || notes) {
                 parts.push(`### Slide ${slide.slideNumber}`);
-                parts.push(texts.join("\n"));
+                if (texts.length > 0) {
+                    parts.push(texts.join("\n"));
+                }
+                if (notes) {
+                    parts.push(`**Speaker notes:** ${notes}`);
+                }
                 parts.push(""); // blank line between slides
             }
         }
         const result = parts.join("\n").trim();
         return result || null;
     }
+    /**
+     * Extract the speaker notes for one slide.
+     *
+     * Speaker notes live in a separate `ppt/notesSlides/notesSlideN.xml` part
+     * whose number is NOT tied to the slide number — the link is declared in the
+     * slide's own relationships file (`ppt/slides/_rels/slideN.xml.rels`) via a
+     * relationship of type `.../notesSlide`. We resolve that target, then pull the
+     * `<a:t>` runs from the notes part. Returns null when the slide has no notes.
+     *
+     * @param zip - The opened PPTX archive
+     * @param slideNumber - 1-indexed slide number
+     * @returns The notes text (runs joined by a space), or null when absent
+     */
+    static extractNotesForSlide(zip, slideNumber) {
+        const relsEntry = zip.getEntry(SLIDE_RELS_NAME(slideNumber));
+        if (!relsEntry) {
+            return null;
+        }
+        const relsXml = relsEntry.getData().toString("utf-8");
+        let notesTarget = null;
+        RELATIONSHIP_TAG_REGEX.lastIndex = 0;
+        for (let match = RELATIONSHIP_TAG_REGEX.exec(relsXml); match !== null; match = RELATIONSHIP_TAG_REGEX.exec(relsXml)) {
+            const tag = match[0];
+            if (/Type="[^"]*\/notesSlide"/.test(tag)) {
+                notesTarget = tag.match(/Target="([^"]+)"/)?.[1] ?? null;
+                break;
+            }
+        }
+        if (!notesTarget) {
+            return null;
+        }
+        // Targets are relative to the slide's folder (ppt/slides/), e.g.
+        // "../notesSlides/notesSlide1.xml" → "ppt/notesSlides/notesSlide1.xml".
+        const normalized = notesTarget
+            .replace(/^\.\.\//, "ppt/")
+            .replace(/^\/+/, "");
+        const notesEntry = zip.getEntry(normalized);
+        if (!notesEntry) {
+            return null;
+        }
+        const notes = PptxProcessor.extractTextFromXml(notesEntry.getData().toString("utf-8")).join(" ");
+        return notes.trim() || null;
+    }
     /**
      * Extract text strings from a slide XML document.
      * Finds all `<a:t>` elements and returns their text content.

package/dist/lib/utils/fileDetector.js

@@ -1441,20 +1441,34 @@ class MagicBytesStrategy {
         if (this.isPDF(input)) {
             return this.result("pdf", "application/pdf", 95);
         }
-        // MP4/MOV: "ftyp" at offset 4
+        // ISO-BMFF ("ftyp" at offset 4): MP4 video, QuickTime MOV, or M4A/M4B/M4P
+        // audio all share this box — disambiguate by the major brand at offset 8-11,
+        // otherwise an .m4a audio file is misrouted to the video pipeline.
         if (input.length >= 8 &&
             input[4] === 0x66 &&
             input[5] === 0x74 &&
             input[6] === 0x79 &&
             input[7] === 0x70) {
+            const brand = input.length >= 12 ? input.toString("latin1", 8, 12) : "";
+            if (/^(M4A|M4B|M4P|F4A|F4B)/.test(brand)) {
+                return this.result("audio", "audio/mp4", 95);
+            }
+            if (brand.startsWith("qt")) {
+                return this.result("video", "video/quicktime", 95);
+            }
             return this.result("video", "video/mp4", 95);
         }
-        // MKV/WebM: EBML header
+        // EBML container (MKV/WebM) — both share the 0x1A45DFA3 header; the DocType
+        // string in the header disambiguates WebM from generic Matroska.
         if (input.length >= 4 &&
             input[0] === 0x1a &&
             input[1] === 0x45 &&
             input[2] === 0xdf &&
             input[3] === 0xa3) {
+            const head = input.toString("latin1", 0, Math.min(input.length, 64));
+            if (head.includes("webm")) {
+                return this.result("video", "video/webm", 92);
+            }
             return this.result("video", "video/x-matroska", 90);
         }
         // AVI: "RIFF" + "AVI "
@@ -1488,7 +1502,13 @@ class MagicBytesStrategy {
             input[2] === 0x33) {
             return this.result("audio", "audio/mpeg", 95);
         }
-        // MP3: sync word
+        // AAC (ADTS): 12-bit syncword 0xFFF with the 2 layer bits == 00. This must be
+        // checked before the MP3 sync word below, because an ADTS header also satisfies
+        // the looser 11-bit MPEG sync mask and would otherwise be mislabeled audio/mpeg.
+        if (input.length >= 2 && input[0] === 0xff && (input[1] & 0xf6) === 0xf0) {
+            return this.result("audio", "audio/aac", 85);
+        }
+        // MP3: sync word (MPEG audio — layer bits are non-zero, unlike ADTS AAC above)
         if (input.length >= 2 && input[0] === 0xff && (input[1] & 0xe0) === 0xe0) {
             return this.result("audio", "audio/mpeg", 80);
         }

package/dist/processors/document/PptxProcessor.d.ts

@@ -42,6 +42,20 @@ export declare class PptxProcessor {
      * @throws Error if the buffer is not a valid ZIP/PPTX file
      */
     static extractText(content: Buffer): Promise<string | null>;
+    /**
+     * Extract the speaker notes for one slide.
+     *
+     * Speaker notes live in a separate `ppt/notesSlides/notesSlideN.xml` part
+     * whose number is NOT tied to the slide number — the link is declared in the
+     * slide's own relationships file (`ppt/slides/_rels/slideN.xml.rels`) via a
+     * relationship of type `.../notesSlide`. We resolve that target, then pull the
+     * `<a:t>` runs from the notes part. Returns null when the slide has no notes.
+     *
+     * @param zip - The opened PPTX archive
+     * @param slideNumber - 1-indexed slide number
+     * @returns The notes text (runs joined by a space), or null when absent
+     */
+    private static extractNotesForSlide;
     /**
      * Extract text strings from a slide XML document.
      * Finds all `<a:t>` elements and returns their text content.

package/dist/processors/document/PptxProcessor.js

@@ -37,6 +37,14 @@ const TEXT_ELEMENT_REGEX = /<a:t[^>]*>([\s\S]*?)<\/a:t>/g;
  * Matches entries like "ppt/slides/slide1.xml", "ppt/slides/slide12.xml".
  */
 const SLIDE_ENTRY_REGEX = /^ppt\/slides\/slide(\d+)\.xml$/;
+/**
+ * A slide's per-slide relationships file, e.g.
+ * "ppt/slides/_rels/slide1.xml.rels", which links the slide to its speaker
+ * notes part (among other relationships).
+ */
+const SLIDE_RELS_NAME = (n) => `ppt/slides/_rels/slide${n}.xml.rels`;
+/** Matches a single `<Relationship .../>` tag inside a .rels document. */
+const RELATIONSHIP_TAG_REGEX = /<Relationship\b[^>]*\/?>/g;
 /**
  * Static utility class for extracting text from PPTX files.
  *
@@ -74,15 +82,65 @@ export class PptxProcessor {
         parts.push(`Presentation: ${slides.length} slide(s)\n`);
         for (const slide of slides) {
             const texts = PptxProcessor.extractTextFromXml(slide.xml);
-            if (texts.length > 0) {
+            const notes = PptxProcessor.extractNotesForSlide(zip, slide.slideNumber);
+            // Emit a slide section when it has either body text or speaker notes.
+            if (texts.length > 0 || notes) {
                 parts.push(`### Slide ${slide.slideNumber}`);
-                parts.push(texts.join("\n"));
+                if (texts.length > 0) {
+                    parts.push(texts.join("\n"));
+                }
+                if (notes) {
+                    parts.push(`**Speaker notes:** ${notes}`);
+                }
                 parts.push(""); // blank line between slides
             }
         }
         const result = parts.join("\n").trim();
         return result || null;
     }
+    /**
+     * Extract the speaker notes for one slide.
+     *
+     * Speaker notes live in a separate `ppt/notesSlides/notesSlideN.xml` part
+     * whose number is NOT tied to the slide number — the link is declared in the
+     * slide's own relationships file (`ppt/slides/_rels/slideN.xml.rels`) via a
+     * relationship of type `.../notesSlide`. We resolve that target, then pull the
+     * `<a:t>` runs from the notes part. Returns null when the slide has no notes.
+     *
+     * @param zip - The opened PPTX archive
+     * @param slideNumber - 1-indexed slide number
+     * @returns The notes text (runs joined by a space), or null when absent
+     */
+    static extractNotesForSlide(zip, slideNumber) {
+        const relsEntry = zip.getEntry(SLIDE_RELS_NAME(slideNumber));
+        if (!relsEntry) {
+            return null;
+        }
+        const relsXml = relsEntry.getData().toString("utf-8");
+        let notesTarget = null;
+        RELATIONSHIP_TAG_REGEX.lastIndex = 0;
+        for (let match = RELATIONSHIP_TAG_REGEX.exec(relsXml); match !== null; match = RELATIONSHIP_TAG_REGEX.exec(relsXml)) {
+            const tag = match[0];
+            if (/Type="[^"]*\/notesSlide"/.test(tag)) {
+                notesTarget = tag.match(/Target="([^"]+)"/)?.[1] ?? null;
+                break;
+            }
+        }
+        if (!notesTarget) {
+            return null;
+        }
+        // Targets are relative to the slide's folder (ppt/slides/), e.g.
+        // "../notesSlides/notesSlide1.xml" → "ppt/notesSlides/notesSlide1.xml".
+        const normalized = notesTarget
+            .replace(/^\.\.\//, "ppt/")
+            .replace(/^\/+/, "");
+        const notesEntry = zip.getEntry(normalized);
+        if (!notesEntry) {
+            return null;
+        }
+        const notes = PptxProcessor.extractTextFromXml(notesEntry.getData().toString("utf-8")).join(" ");
+        return notes.trim() || null;
+    }
     /**
      * Extract text strings from a slide XML document.
      * Finds all `<a:t>` elements and returns their text content.

package/dist/utils/fileDetector.js

@@ -1441,20 +1441,34 @@ class MagicBytesStrategy {
         if (this.isPDF(input)) {
             return this.result("pdf", "application/pdf", 95);
         }
-        // MP4/MOV: "ftyp" at offset 4
+        // ISO-BMFF ("ftyp" at offset 4): MP4 video, QuickTime MOV, or M4A/M4B/M4P
+        // audio all share this box — disambiguate by the major brand at offset 8-11,
+        // otherwise an .m4a audio file is misrouted to the video pipeline.
         if (input.length >= 8 &&
             input[4] === 0x66 &&
             input[5] === 0x74 &&
             input[6] === 0x79 &&
             input[7] === 0x70) {
+            const brand = input.length >= 12 ? input.toString("latin1", 8, 12) : "";
+            if (/^(M4A|M4B|M4P|F4A|F4B)/.test(brand)) {
+                return this.result("audio", "audio/mp4", 95);
+            }
+            if (brand.startsWith("qt")) {
+                return this.result("video", "video/quicktime", 95);
+            }
             return this.result("video", "video/mp4", 95);
         }
-        // MKV/WebM: EBML header
+        // EBML container (MKV/WebM) — both share the 0x1A45DFA3 header; the DocType
+        // string in the header disambiguates WebM from generic Matroska.
         if (input.length >= 4 &&
             input[0] === 0x1a &&
             input[1] === 0x45 &&
             input[2] === 0xdf &&
             input[3] === 0xa3) {
+            const head = input.toString("latin1", 0, Math.min(input.length, 64));
+            if (head.includes("webm")) {
+                return this.result("video", "video/webm", 92);
+            }
             return this.result("video", "video/x-matroska", 90);
         }
         // AVI: "RIFF" + "AVI "
@@ -1488,7 +1502,13 @@ class MagicBytesStrategy {
             input[2] === 0x33) {
             return this.result("audio", "audio/mpeg", 95);
         }
-        // MP3: sync word
+        // AAC (ADTS): 12-bit syncword 0xFFF with the 2 layer bits == 00. This must be
+        // checked before the MP3 sync word below, because an ADTS header also satisfies
+        // the looser 11-bit MPEG sync mask and would otherwise be mislabeled audio/mpeg.
+        if (input.length >= 2 && input[0] === 0xff && (input[1] & 0xf6) === 0xf0) {
+            return this.result("audio", "audio/aac", 85);
+        }
+        // MP3: sync word (MPEG audio — layer bits are non-zero, unlike ADTS AAC above)
         if (input.length >= 2 && input[0] === 0xff && (input[1] & 0xe0) === 0xe0) {
             return this.result("audio", "audio/mpeg", 80);
         }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@juspay/neurolink",
-  "version": "9.88.3",
+  "version": "9.88.5",
   "packageManager": "pnpm@10.15.1",
   "description": "TypeScript AI SDK with 24+ LLM providers behind one consistent API. MCP-native (58+ servers), voice TTS/STT/realtime, RAG, agents, memory, context compaction. OpenAI · Anthropic · Gemini · Bedrock · Azure · Ollama · DeepSeek · NVIDIA NIM and more.",
   "author": {