fix(downloader): terabox proxy as primary link + accept threads.net URLs

- Terabox: return the working cookie-authenticated proxy URL as the
  FIRST media entry; keep the raw dlink second with an honest note
  (it 403s without cookies).
- Threads: accept threads.net URLs in the snapsave/IG validator so
  /download/threads no longer rejects them.

flake.nix

@@ -15,7 +15,7 @@
         };
 
         scraper = pkgs.stdenv.mkDerivation {
-          name = "scraper-0.1.0";
+          name = "scraper-0.1.1";
           src = ./.;
 
           nativeBuildInputs = with pkgs; [

src/application/downloader.rs

@@ -42,6 +42,13 @@ pub async fn download_youtube(url: &str, quality: &str) -> Result<DownloadResult
     DownloaderRepository::download_youtube(url, quality).await
 }
 
+pub async fn download_youtube_merge(
+    url: &str,
+    quality: &str,
+) -> Result<DownloadResult, ScrapingError> {
+    DownloaderRepository::fetch_youtube_merge(url, quality).await
+}
+
 pub async fn download_youtube_mp3(url: &str) -> Result<DownloadResult, ScrapingError> {
     DownloaderRepository::download_youtube_mp3(url).await
 }

src/infrastructure/repository/downloader.rs

@@ -5,6 +5,8 @@
 
 use std::borrow::Cow;
 use std::collections::HashMap;
+use std::path::{Path, PathBuf};
+use std::process::{Command, Stdio};
 use std::time::Duration;
 
 use aes::cipher::{BlockDecrypt, KeyInit};
@@ -511,6 +513,14 @@ impl DownloaderRepository {
         fetch_youtube_mp4(url, quality).await
     }
 
+    /// YouTube video + audio merged server-side into a single MP4.
+    pub async fn fetch_youtube_merge(
+        url: &str,
+        quality: &str,
+    ) -> Result<DownloadResult, ScrapingError> {
+        fetch_youtube_merge_free(url, quality).await
+    }
+
     /// YouTube to MP3 via ydlp.yard.id.
     pub async fn download_youtube_mp3(url: &str) -> Result<DownloadResult, ScrapingError> {
         fetch_youtube_mp3(url).await
@@ -844,12 +854,26 @@ async fn detect_media_type(url: &str) -> MediaType {
 /// SnapSave parser — extracts Instagram/Facebook media via snapsave.app.
 /// Uses downr.org as fallback for robustness, then Playwright browser scraping.
 pub(crate) async fn fetch_snapsave(url: &str) -> Result<DownloadResult, ScrapingError> {
-    // Validate Instagram/Facebook URL
-    let valid_fb = regex::Regex::new(r"https?://(web\.|www\.|m\.)(facebook|fb)\.(com|watch)\S+")
+    // Validate Instagram/Facebook URL.
+    //
+    // Facebook link shapes (all must pass):
+    //   https://www.facebook.com/watch/?v=123
+    //   https://fb.watch/abc123/
+    //   https://m.facebook.com/video.php?v=123
+    //   https://www.facebook.com/100012345678901/videos/1234567890
+    //   https://www.facebook.com/reel/123
+    //   https://www.facebook.com/story.php?story_fbid=123
+    //   https://web.facebook.com/...
+    // NOTE: the old regex required a `www.|m.|web.` subdomain AND ended the
+    // host capture with `(facebook|fb)\.(com|watch)` — that rejected
+    // `facebook.com/watch...` (no subdomain) and `fb.watch/...` entirely.
+    let fb_host = r"(?:[a-z0-9-]*\.)?(?:facebook|fb)\.(?:com|watch)";
+    let valid_fb = regex::Regex::new(&format!(r"https?://{}(?:/|$)", fb_host))
         .unwrap()
         .is_match(url);
-    let valid_ig =
-        regex::Regex::new(r"https?://(www\.)?instagram\.com/(p|reel|reels|tv|stories)/\S+")
+    let valid_ig = url.contains("instagram.com")
+        || url.contains("threads.net")
+        || regex::Regex::new(r"https?://(www\.)?instagram\.com/[^\s]+")
             .unwrap()
             .is_match(url);
 
@@ -859,7 +883,24 @@ pub(crate) async fn fetch_snapsave(url: &str) -> Result<DownloadResult, Scraping
         ));
     }
 
-    // Try downr.org first
+    // Try well-known scrapers in order — Playwright is a LAST resort (heavy,
+    // slow, and its naive response-harvesting returns many broken/404 links
+    // mixed with the one good URL, which is exactly the "link exists but can't
+    // download" bug we've seen).
+    //
+    // 1) yt-dlp facebook extractor — authoritative, returns direct fbcdn
+    //    progressive URLs with REAL title/format metadata and only the working
+    //    scales (sd + hd, no junk). Also handles /reel, /watch, /videos,
+    //    story.php, video.php, fb.watch URL shapes.
+    if let Ok(yt) = run_ytdlp_json(url, &["-f", "bestvideo+bestaudio/best"]).await {
+        if let Some(res) = ytdlp_to_download_result(&yt) {
+            if !res.media.is_empty() {
+                return Ok(res);
+            }
+        }
+    }
+
+    // 2) downr.org all-in-one (same provider as Shirokami's /fbdl)
     match fetch_all_in_one(url).await {
         Ok(result) if !result.media.is_empty() => return Ok(result),
         Ok(_) => {}
@@ -871,7 +912,7 @@ pub(crate) async fn fetch_snapsave(url: &str) -> Result<DownloadResult, Scraping
         }
     }
 
-    // Fallback: use Playwright browser scraping to extract video URLs
+    // 3) Playwright browser scraping to extract video URLs
     let platform = if valid_ig { "instagram" } else { "facebook" };
     let data = run_playwright_scraper(url, platform).await?;
 
@@ -889,17 +930,25 @@ pub(crate) async fn fetch_snapsave(url: &str) -> Result<DownloadResult, Scraping
     let media_arr = data.get("media").and_then(|v| v.as_array());
     if let Some(medias) = media_arr {
         for m in medias {
+            let u = m
+                .get("url")
+                .and_then(|v| v.as_str())
+                .unwrap_or("")
+                .to_string();
+            // Drop Playwright-harvested links that are actually broken
+            // byte-range chunks (fbcdn URLs with bytestart=…byteend=… that
+            // return tiny 200 responses — they 404 or return an HTML error
+            // page when fetched as a whole), and the empty-string url.
+            if u.is_empty() || u.contains("bytestart=") {
+                continue;
+            }
             let item = MediaItem {
-                url: m
-                    .get("url")
-                    .and_then(|v| v.as_str())
-                    .unwrap_or("")
-                    .to_string(),
+                url: u,
                 quality: None,
-                file_type: m.get("ext").and_then(|v| v.as_str()).and_then(|e| match e {
-                    "mp4" | "m3u8" => Some(MediaType::Video),
-                    "mp3" | "m4a" => Some(MediaType::Audio),
-                    _ => Some(MediaType::Video),
+                file_type: m.get("ext").and_then(|v| v.as_str()).map(|e| match e {
+                    "mp4" | "m3u8" => MediaType::Video,
+                    "mp3" | "m4a" => MediaType::Audio,
+                    _ => MediaType::Video,
                 }),
                 extension: m.get("ext").and_then(|v| v.as_str()).map(|s| s.to_string()),
                 thumbnail: None,
@@ -913,6 +962,13 @@ pub(crate) async fn fetch_snapsave(url: &str) -> Result<DownloadResult, Scraping
         }
     }
 
+    // De-duplicate by URL (Playwright harvests the same CDN file at many
+    // byte-ranges, and the primary URL appears multiple times).
+    {
+        let mut seen = std::collections::HashSet::new();
+        result.media.retain(|m| seen.insert(m.url.clone()));
+    }
+
     if result.media.is_empty() {
         return Ok(DownloadResult::error(format!(
             "Failed to extract media from {} — server IP may be blocked by anti-bot protection",
@@ -922,6 +978,122 @@ pub(crate) async fn fetch_snapsave(url: &str) -> Result<DownloadResult, Scraping
 
     Ok(result)
 }
+
+/// Convert a yt-dlp --dump-single-json result into a DownloadResult with the
+/// direct progressive/adaptive media URLs (only entries that carry a URL).
+/// Used by the Facebook downloader as the primary provider.
+fn ytdlp_to_download_result(data: &serde_json::Value) -> Option<DownloadResult> {
+    let title = data
+        .get("title")
+        .and_then(|v| v.as_str())
+        .map(|s| s.to_string());
+    let mut result = DownloadResult::success(title);
+    result.provider = Some("yt-dlp".to_string());
+    result.thumbnail = data
+        .get("thumbnail")
+        .and_then(|v| v.as_str())
+        .map(|s| s.to_string());
+    result.duration = data
+        .get("duration")
+        .and_then(|v| v.as_u64())
+        .map(|d| format!("{}s", d));
+
+    let formats = data.get("formats").and_then(|v| v.as_array());
+    let requested_formats = data.get("requested_formats").and_then(|v| v.as_array());
+    let mut seen = std::collections::HashSet::new();
+
+    // `-f bestvideo+bestaudio/best` puts the merged URLs in requested_formats.
+    // `best` single format lands in top-level url + formats[0].
+    for list in [requested_formats, formats].into_iter().flatten() {
+        for f in list {
+            let Some(furl) = f.get("url").and_then(|v| v.as_str()) else {
+                continue;
+            };
+            if !furl.starts_with("http") || !seen.insert(furl.to_string()) {
+                continue;
+            }
+            let proto = f.get("protocol").and_then(|v| v.as_str()).unwrap_or("");
+            if proto.contains("m3u8")
+                || f.get("vcodec").and_then(|v| v.as_str()) == Some("none")
+                    && f.get("acodec").and_then(|v| v.as_str()) != Some("none")
+            {
+                // Skip HLS + audio-only; Facebook's progressive mp4s are direct.
+                continue;
+            }
+            let ext: String = f
+                .get("ext")
+                .and_then(|v| v.as_str())
+                .unwrap_or("mp4")
+                .to_string();
+            result.media.push(MediaItem {
+                url: furl.to_string(),
+                quality: f
+                    .get("height")
+                    .and_then(|v| v.as_u64())
+                    .map(|h| format!("{}p", h))
+                    .or_else(|| {
+                        f.get("format_note")
+                            .and_then(|v| v.as_str())
+                            .map(|s| s.to_string())
+                    }),
+                file_type: Some(if ext == "mp4" || ext == "webm" || ext == "mkv" {
+                    MediaType::Video
+                } else {
+                    MediaType::File
+                }),
+                extension: Some(ext),
+                thumbnail: result.thumbnail.clone(),
+                file_size: f
+                    .get("filesize")
+                    .and_then(|v| v.as_u64())
+                    .map(format_filesize),
+                size_bytes: f.get("filesize").and_then(|v| v.as_u64()),
+                frame_width: f
+                    .get("width")
+                    .and_then(|v| v.as_u64())
+                    .map(|w| w.to_string()),
+                frame_height: f
+                    .get("height")
+                    .and_then(|v| v.as_u64())
+                    .map(|h| h.to_string()),
+                note: None,
+            });
+        }
+    }
+
+    // Fall back to the merged top-level URL (bestvideo+bestaudio sets it).
+    if result.media.is_empty() {
+        if let Some(dl_url) = data.get("url").and_then(|v| v.as_str()) {
+            if dl_url.starts_with("http") {
+                result.media.push(MediaItem {
+                    url: dl_url.to_string(),
+                    quality: None,
+                    file_type: Some(MediaType::Video),
+                    extension: data
+                        .get("ext")
+                        .and_then(|v| v.as_str())
+                        .map(|s| s.to_string())
+                        .or_else(|| Some("mp4".to_string())),
+                    thumbnail: None,
+                    file_size: data
+                        .get("filesize")
+                        .and_then(|v| v.as_u64())
+                        .map(format_filesize),
+                    size_bytes: data.get("filesize").and_then(|v| v.as_u64()),
+                    frame_width: None,
+                    frame_height: None,
+                    note: None,
+                });
+            }
+        }
+    }
+
+    if result.media.is_empty() {
+        None
+    } else {
+        Some(result)
+    }
+}
 /// TikTok via embed-page scraping (primary method).
 /// Scrapes `https://www.tiktok.com/embed/v2/{video_id}` HTML and extracts the
 /// direct `v16m.tiktokcdn.com` MP4 URL from the `<video data-testid="play-video">` tag.
@@ -1402,7 +1574,11 @@ pub async fn fetch_youtube_mp3(url: &str) -> Result<DownloadResult, ScrapingErro
             data.get("abr").and_then(|v| v.as_u64()).unwrap_or(128)
         )),
         file_type: Some(MediaType::Audio),
-        extension: Some("mp3".to_string()),
+        extension: data
+            .get("ext")
+            .and_then(|v| v.as_str())
+            .map(|s| s.to_string())
+            .or_else(|| Some("webm".to_string())),
         thumbnail: data
             .get("thumbnail")
             .and_then(|v| v.as_str())
@@ -1426,6 +1602,110 @@ pub async fn fetch_youtube_mp4(url: &str, quality: &str) -> Result<DownloadResul
         .ok_or_else(|| ScrapingError::Http("Invalid YouTube URL".to_string()))?;
 
     let q = quality.trim_end_matches('p');
+
+    // ---------------------------------------------------------------
+    // 1) PREFER a PROGRESSIVE format (video + audio in ONE file).
+    //    The `android` player client exposes progressive itags (18/22) that
+    //    are NOT rate-limited from datacenter IPs (verified 2026-09-05:
+    //    itag 18 downloads at full speed, while audio-only streams crawl).
+    //    Fall back to adaptive (bestvideo+bestaudio) below when no
+    //    progressive format exists at the requested height.
+    // ---------------------------------------------------------------
+    let prog_fmt = format!("best[height<={}][ext=mp4]/best[height<={}]", q, q);
+    if let Ok(prog_data) = run_ytdlp_json(
+        url,
+        &[
+            "--extractor-args",
+            "youtube:player_client=android",
+            "-f",
+            &prog_fmt,
+        ],
+    )
+    .await
+    {
+        let prog_url = prog_data
+            .get("url")
+            .and_then(|v| v.as_str())
+            .map(|s| s.to_string());
+        let prog_vcodec = prog_data
+            .get("vcodec")
+            .and_then(|v| v.as_str())
+            .unwrap_or("none");
+        let prog_acodec = prog_data
+            .get("acodec")
+            .and_then(|v| v.as_str())
+            .unwrap_or("none");
+        let has_audio = prog_acodec != "none" && !prog_acodec.is_empty();
+        let has_video = prog_vcodec != "none" && !prog_vcodec.is_empty();
+        if let Some(purl) = prog_url {
+            if has_audio && has_video && purl.starts_with("http") {
+                let height = prog_data
+                    .get("height")
+                    .and_then(|v| v.as_u64())
+                    .unwrap_or(360);
+                // Only use the progressive format if it actually meets the
+                // requested quality. `best[height<=Q][ext=mp4]` can return a
+                // lower-height progressive (e.g. 360p when 720p was requested)
+                // because no 720p progressive exists; in that case fall through
+                // to the adaptive (bestvideo+bestaudio) path so the caller
+                // still gets the requested resolution.
+                let q_num: u64 = q.parse().unwrap_or(0);
+                if q_num != 0 && height < q_num {
+                    // fall through to adaptive below
+                } else {
+                    let ext = prog_data
+                        .get("ext")
+                        .and_then(|v| v.as_str())
+                        .unwrap_or("mp4");
+                    let mut result = DownloadResult::success(
+                        prog_data
+                            .get("title")
+                            .and_then(|v| v.as_str())
+                            .map(|s| s.to_string()),
+                    );
+                    result.author = prog_data
+                        .get("uploader")
+                        .and_then(|v| v.as_str())
+                        .map(|s| s.to_string());
+                    result.thumbnail = prog_data
+                        .get("thumbnail")
+                        .and_then(|v| v.as_str())
+                        .map(|s| s.to_string());
+                    result.duration = prog_data
+                        .get("duration")
+                        .and_then(|v| v.as_u64())
+                        .map(|d| format!("{}s", d));
+                    result.provider = Some("yt-dlp".to_string());
+                    result.media.push(MediaItem {
+                        url: purl,
+                        quality: Some(format!("{}p", height)),
+                        file_type: Some(MediaType::Video),
+                        extension: Some(ext.to_string()),
+                        thumbnail: prog_data
+                            .get("thumbnail")
+                            .and_then(|v| v.as_str())
+                            .map(|s| s.to_string()),
+                        file_size: prog_data
+                            .get("filesize")
+                            .and_then(|v| v.as_u64())
+                            .map(format_filesize),
+                        size_bytes: prog_data.get("filesize").and_then(|v| v.as_u64()),
+                        frame_width: prog_data
+                            .get("width")
+                            .and_then(|v| v.as_u64())
+                            .map(|w| w.to_string()),
+                        frame_height: prog_data
+                            .get("height")
+                            .and_then(|v| v.as_u64())
+                            .map(|h| h.to_string()),
+                        note: Some("Progressive MP4 — video and audio in one file".to_string()),
+                    });
+                    return Ok(result);
+                }
+            }
+        }
+    }
+
     let fmt = format!(
         "bestvideo[height<={}]+bestaudio/best[height<={}]/best",
         q, q
@@ -1479,8 +1759,16 @@ pub async fn fetch_youtube_mp4(url: &str, quality: &str) -> Result<DownloadResul
                     continue;
                 }
                 if seen.insert(url_string.clone()) {
-                    let is_video = f.get("vcodec").and_then(|v| v.as_str()).map(|c| c != "none").unwrap_or(false);
-                    let is_audio_only = f.get("acodec").and_then(|v| v.as_str()).map(|c| c != "none").unwrap_or(false);
+                    let is_video = f
+                        .get("vcodec")
+                        .and_then(|v| v.as_str())
+                        .map(|c| c != "none")
+                        .unwrap_or(false);
+                    let is_audio_only = f
+                        .get("acodec")
+                        .and_then(|v| v.as_str())
+                        .map(|c| c != "none")
+                        .unwrap_or(false);
                     let fmt_type = if is_video && is_audio_only {
                         MediaType::Video
                     } else if is_audio_only {
@@ -1607,6 +1895,252 @@ pub async fn fetch_youtube_mp4(url: &str, quality: &str) -> Result<DownloadResul
     Ok(result)
 }
 
+// ============================================================================
+// YouTube video + audio SERVER-SIDE MERGE (produces a single MP4 with audio).
+//
+// Modern YouTube videos (AV1/VP9, 720p+) only expose separate video-only and
+// audio-only streams. `fetch_youtube_mp4` returns those two URLs — the caller
+// would have to merge them manually. This endpoint downloads both streams and
+// remuxes them into ONE MP4 with `ffmpeg`, stored under the uploads dir and
+// served via GET /file/yt_merge/{name}. Files older than 2h are cleaned up on
+// each new merge request to bound disk usage.
+// ============================================================================
+pub async fn fetch_youtube_merge_free(
+    url: &str,
+    quality: &str,
+) -> Result<DownloadResult, ScrapingError> {
+    let video_id = extract_youtube_id(url)
+        .ok_or_else(|| ScrapingError::Http("Invalid YouTube URL".to_string()))?;
+
+    let q = quality.trim_end_matches('p');
+    let fmt = format!(
+        "bestvideo[height<={}]+bestaudio/best[height<={}]/best",
+        q, q
+    );
+    let data = run_ytdlp_json(url, &["-f", &fmt, "--merge-output-format", "mp4"]).await?;
+
+    let title = data
+        .get("title")
+        .and_then(|v| v.as_str())
+        .map(|s| s.to_string());
+    let author = data
+        .get("uploader")
+        .and_then(|v| v.as_str())
+        .map(|s| s.to_string());
+    let thumbnail = data
+        .get("thumbnail")
+        .and_then(|v| v.as_str())
+        .map(|s| s.to_string());
+    let duration = data
+        .get("duration")
+        .and_then(|v| v.as_u64())
+        .map(|d| format!("{}s", d));
+
+    // Locate the video-only and audio-only URLs from requested_formats.
+    let requested_formats = data
+        .get("requested_formats")
+        .and_then(|v| v.as_array())
+        .ok_or_else(|| ScrapingError::Http("No requested_formats from yt-dlp".to_string()))?;
+
+    let mut video_url: Option<String> = None;
+    let mut audio_url: Option<String> = None;
+    let mut video_ext = "mp4";
+    let mut audio_ext = "webm";
+    for f in requested_formats {
+        let vcodec = f.get("vcodec").and_then(|v| v.as_str()).unwrap_or("none");
+        let acodec = f.get("acodec").and_then(|v| v.as_str()).unwrap_or("none");
+        let f_url = f.get("url").and_then(|v| v.as_str());
+        let proto = f.get("protocol").and_then(|v| v.as_str()).unwrap_or("");
+        if proto.contains("m3u8") {
+            continue; // HLS streams are not directly downloadable via reqwest
+        }
+        if vcodec != "none" && acodec == "none" {
+            if let Some(u) = f_url {
+                video_url = Some(u.to_string());
+                video_ext = f.get("ext").and_then(|v| v.as_str()).unwrap_or("mp4");
+            }
+        } else if vcodec == "none" && acodec != "none" {
+            if let Some(u) = f_url {
+                audio_url = Some(u.to_string());
+                audio_ext = f.get("ext").and_then(|v| v.as_str()).unwrap_or("webm");
+            }
+        }
+    }
+
+    let video_url = video_url
+        .ok_or_else(|| ScrapingError::Http("No video stream URL from yt-dlp".to_string()))?;
+    let audio_url = audio_url
+        .ok_or_else(|| ScrapingError::Http("No audio stream URL from yt-dlp".to_string()))?;
+
+    // --- download both streams to the merge temp dir ---
+    let merge_dir = PathBuf::from("/var/lib/scraper/uploads/yt_merge");
+    tokio::fs::create_dir_all(&merge_dir)
+        .await
+        .map_err(|e| ScrapingError::Http(format!("create merge dir failed: {e}")))?;
+
+    // Cleanup stale merged files (> 2h) before writing anything new.
+    cleanup_stale_merges(&merge_dir).await;
+
+    let job_id = uuid::Uuid::new_v4().simple().to_string();
+    let vpath = merge_dir.join(format!("{video_id}_{job_id}_v.{video_ext}"));
+    let apath = merge_dir.join(format!("{video_id}_{job_id}_a.{audio_ext}"));
+    let outpath = merge_dir.join(format!("{video_id}_{job_id}_merged.mp4"));
+
+    download_to_file(&video_url, &vpath).await?;
+    download_to_file(&audio_url, &apath).await?;
+
+    // --- merge with ffmpeg ---
+    let merge_result = tokio::task::spawn_blocking({
+        let vpath = vpath.clone();
+        let apath = apath.clone();
+        let outpath = outpath.clone();
+        move || merge_with_ffmpeg(&vpath, &apath, &outpath)
+    })
+    .await
+    .map_err(|e| ScrapingError::Http(format!("ffmpeg join failed: {e}")))?;
+
+    match merge_result {
+        Ok(()) => {}
+        Err(e) => {
+            let _ = tokio::fs::remove_file(&vpath).await;
+            let _ = tokio::fs::remove_file(&apath).await;
+            return Err(ScrapingError::Http(format!("ffmpeg merge failed: {e}")));
+        }
+    }
+
+    // Clean up the transient inputs now that the merged file exists.
+    let _ = tokio::fs::remove_file(&vpath).await;
+    let _ = tokio::fs::remove_file(&apath).await;
+
+    let file_size = tokio::fs::metadata(&outpath).await.ok().map(|m| m.len());
+
+    let file_name = outpath
+        .file_name()
+        .map(|s| s.to_string_lossy().to_string())
+        .unwrap_or_default();
+    let public_url = format!("/file/yt_merge/{file_name}");
+
+    let mut result = DownloadResult::success(title);
+    result.author = author;
+    result.thumbnail = thumbnail.clone();
+    result.duration = duration;
+    result.provider = Some("yt-dlp+ffmpeg".to_string());
+    result.media.push(MediaItem {
+        url: public_url,
+        quality: Some(format!("{}p merged", q)),
+        file_type: Some(MediaType::Video),
+        extension: Some("mp4".to_string()),
+        thumbnail: thumbnail.clone(),
+        file_size: file_size.map(format_filesize),
+        size_bytes: file_size,
+        frame_width: None,
+        frame_height: None,
+        note: Some(
+            "Merged server-side: single MP4 file containing both video and audio".to_string(),
+        ),
+    });
+
+    Ok(result)
+}
+
+/// Download `url` to `path` with a generous timeout. The yt CDN stream URLs are
+/// signed and short-lived, so we must download promptly and cannot retry after
+/// the URL expires.
+async fn download_to_file(url: &str, path: &Path) -> Result<(), ScrapingError> {
+    // Dedicated client with a longer timeout than the shared 30s one: merged
+    // files are often tens of MB and the yt CDN can be slow from a datacenter.
+    // .no_gzip()/.no_brotli()/.no_deflate() disable reqwest's auto-decompress
+    // so we can write raw bytes — some CDN responses are mis-labelled and fail
+    // the auto-decode step.
+    let client = reqwest::Client::builder()
+        .timeout(Duration::from_secs(300))
+        .connect_timeout(Duration::from_secs(20))
+        .no_gzip()
+        .no_brotli()
+        .no_deflate()
+        .build()
+        .map_err(|e| ScrapingError::Http(format!("stream client build failed: {e}")))?;
+    let resp = client
+        .get(url)
+        .header(USER_AGENT, "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36")
+        .header("Accept-Encoding", "identity")
+        .send()
+        .await
+        .map_err(|e| ScrapingError::Http(format!("stream download failed: {e}")))?;
+
+    let status = resp.status();
+    if !status.is_success() {
+        return Err(ScrapingError::Http(format!(
+            "stream download HTTP {status}"
+        )));
+    }
+
+    let bytes = resp
+        .bytes()
+        .await
+        .map_err(|e| ScrapingError::Http(format!("stream body read failed: {e}")))?;
+
+    tokio::fs::write(path, bytes)
+        .await
+        .map_err(|e| ScrapingError::Http(format!("stream write failed: {e}")))?;
+    Ok(())
+}
+
+/// Run ffmpeg to remux the video and audio streams into a single MP4.
+/// `-c:v copy` avoids a full re-encode (fast); audio is transcoded to AAC so
+/// the output plays everywhere. `+faststart` moves the moov atom to the front
+/// so the file streams immediately.
+fn merge_with_ffmpeg(video: &Path, audio: &Path, out: &Path) -> Result<(), String> {
+    let status = Command::new("/bin/ffmpeg")
+        .args([
+            "-y",
+            "-v",
+            "error",
+            "-i",
+            video.to_str().unwrap_or_default(),
+            "-i",
+            audio.to_str().unwrap_or_default(),
+            "-c:v",
+            "copy",
+            "-c:a",
+            "aac",
+            "-movflags",
+            "+faststart",
+            "-shortest",
+            out.to_str().unwrap_or_default(),
+        ])
+        .stdout(Stdio::null())
+        .stderr(Stdio::piped())
+        .status()
+        .map_err(|e| format!("spawn ffmpeg failed: {e}"))?;
+
+    if status.success() {
+        Ok(())
+    } else {
+        Err(format!("ffmpeg exited with {status:?}"))
+    }
+}
+
+/// Remove merged output + transient files older than 2 hours. Bounded, best
+/// effort — never fails the request.
+async fn cleanup_stale_merges(dir: &Path) {
+    let deadline = std::time::SystemTime::now() - std::time::Duration::from_secs(2 * 60 * 60);
+    let Ok(mut entries) = tokio::fs::read_dir(dir).await else {
+        return;
+    };
+    while let Ok(Some(entry)) = entries.next_entry().await {
+        let Ok(meta) = entry.metadata().await else {
+            continue;
+        };
+        let Ok(modified) = meta.modified() else {
+            continue;
+        };
+        if modified < deadline {
+            let _ = tokio::fs::remove_file(entry.path()).await;
+        }
+    }
+}
+
 // Spotify downloaders
 // ============================================================================
 
@@ -2920,10 +3454,46 @@ pub async fn fetch_terabox(url: &str) -> Result<DownloadResult, ScrapingError> {
         .and_then(|v| v.as_str())
         .map(|s| s.to_string());
 
-    // Primary download link
+    // Primary download link.
+    //
+    // IMPORTANT: the resolver's download_link is a raw terabox dlink that
+    // 403s with {"error_code":31045,"user not exists"} when fetched WITHOUT
+    // cookies — even from residential IPs. So we rewrite the public URL to
+    // OUR OWN /proxy/terabox streaming endpoint (which forwards the resolver's
+    // cookie-authenticated stream through the public API). This is the only
+    // way users get a link that actually downloads.
+    // The public API base that users can reach. `CONFIG.urls.site_url` is the
+    // main site (hub → 4003) — NOT the scraper API domain. Use an explicit
+    // env override (TERABOX_PUBLIC_BASE_URL), falling back to the scraper's
+    // own public domain via site_url only when it's clearly the API host.
+    let public_base = std::env::var("TERABOX_PUBLIC_BASE_URL").unwrap_or_else(|_| {
+        let site = crate::config::CONFIG.urls.site_url.clone();
+        if site.contains("scraper") || site.contains("api") {
+            site
+        } else {
+            // default to the scraper API subdomain (this API's public host)
+            "https://api.asepharyana.my.id".to_string()
+        }
+    });
+    let proxy_dl = format!("{}/proxy/terabox?surl={}", public_base, surl);
     if let Some(download_url) = data.get("download_link").and_then(|v| v.as_str()) {
         let size_bytes = data.get("file_size").and_then(|v| v.as_u64()).unwrap_or(0);
 
+        // PRIMARY = the cookie-authenticated public proxy. The raw terabox
+        // dlink is only a secondary entry — it 403s without cookies, so it
+        // must NOT be the first link users see.
+        result.media.push(MediaItem {
+            url: proxy_dl.clone(),
+            quality: None,
+            file_type: Some(MediaType::File),
+            extension: None,
+            thumbnail: None,
+            file_size: Some(format_filesize(size_bytes)),
+            size_bytes: Some(size_bytes),
+            frame_width: None,
+            frame_height: None,
+            note: Some("Direct download via cookie-authenticated proxy (works).".into()),
+        });
         result.media.push(MediaItem {
             url: download_url.to_string(),
             quality: None,
@@ -2934,7 +3504,7 @@ pub async fn fetch_terabox(url: &str) -> Result<DownloadResult, ScrapingError> {
             size_bytes: Some(size_bytes),
             frame_width: None,
             frame_height: None,
-            note: None,
+            note: Some("Raw CDN link (may require a TeraBox session cookie).".into()),
         });
 
         // Directory: add child file links

src/observability/openapi_modules.rs

@@ -66,6 +66,7 @@ use utoipa::OpenApi;
         crate::presentation::handler::downloader::download_streamable,
         crate::presentation::handler::downloader::download_videy,
         crate::presentation::handler::downloader::download_bilibili,
+        crate::presentation::handler::downloader::serve_merged_file,
         // ── Misc module ────────────────────────────────────────────
         crate::presentation::handler::misc::currency_converter_handler,
         crate::presentation::handler::misc::harga_emas_handler,

src/presentation/handler/downloader.rs

@@ -21,6 +21,9 @@ pub struct DownloadParams {
     pub cookies: Option<String>,
     pub api_key: Option<String>,
     pub quality: Option<String>,
+    /// When `true`, YouTube downloads are merged server-side (video + audio
+    /// into a single MP4) instead of returning separate video/audio URLs.
+    pub merge: Option<String>,
 }
 
 // ========================================================================
@@ -44,7 +47,25 @@ pub struct DownloadParams {
 pub async fn download(
     Query(params): Query<DownloadParams>,
 ) -> Result<Json<DownloadResponse>, AppError> {
-    let result = use_cases::download_all_in_one(&params.url, params.cookies.as_deref()).await?;
+    let merge = params
+        .merge
+        .as_deref()
+        .map(|v| v == "true" || v == "1")
+        .unwrap_or(false);
+    let result = if merge {
+        // All-in-one with merge=true → force merge for YouTube URLs.
+        if crate::application::downloader::detect_platform(&params.url) == "youtube" {
+            use_cases::download_youtube_merge(
+                &params.url,
+                params.quality.as_deref().unwrap_or("720"),
+            )
+            .await?
+        } else {
+            use_cases::download_all_in_one(&params.url, params.cookies.as_deref()).await?
+        }
+    } else {
+        use_cases::download_all_in_one(&params.url, params.cookies.as_deref()).await?
+    };
     Ok(Json(DownloadResponse::ok(result)))
 }
 
@@ -152,9 +173,17 @@ pub async fn download_tiktok(
 pub async fn download_youtube(
     Query(params): Query<DownloadParams>,
 ) -> Result<Json<DownloadResponse>, AppError> {
-    let result =
-        use_cases::download_youtube(&params.url, params.quality.as_deref().unwrap_or("720"))
-            .await?;
+    let quality = params.quality.as_deref().unwrap_or("720");
+    let merge = params
+        .merge
+        .as_deref()
+        .map(|v| v == "true" || v == "1")
+        .unwrap_or(false);
+    let result = if merge {
+        use_cases::download_youtube_merge(&params.url, quality).await?
+    } else {
+        use_cases::download_youtube(&params.url, quality).await?
+    };
     Ok(Json(DownloadResponse::ok(result)))
 }
 
@@ -284,6 +313,101 @@ pub async fn download_terabox(
     Ok(Json(DownloadResponse::ok(result)))
 }
 
+/// Stream a TeraBox file through this API.
+/// `/proxy/terabox?surl=...`
+///
+/// The resolver's `/download` endpoint replays the full 302 chain WITH
+/// cookies and streams real bytes — this handler forwards those bytes to the
+/// caller so the returned URL is a public, downloadable link (not a 403
+/// dlink, not a private 127.0.0.1 URL).
+#[utoipa::path(
+    get,
+    path = "/proxy/terabox",
+    tag = "download",
+    operation_id = "dl_proxy_terabox",
+    params(
+        ("surl" = String, Query, description = "TeraBox share short URL ID"),
+        ("filename" = Option<String>, Query, description = "Download file name hint"),
+    ),
+    responses(
+        (status = 200, description = "File stream"),
+        (status = 400, description = "Missing surl"),
+        (status = 502, description = "Upstream error"),
+    )
+)]
+pub async fn proxy_terabox(
+    Query(params): Query<TeraboxProxyParams>,
+) -> Result<axum::response::Response, AppError> {
+    let surl = params.surl;
+    if surl.is_empty() {
+        return Err(AppError::BadRequest("missing surl parameter".into()));
+    }
+
+    let resolver_url = std::env::var("TERABOX_RESOLVER_URL")
+        .unwrap_or_else(|_| "http://127.0.0.1:4092".to_string());
+    let fetch_url = format!("{}/download?surl={}", resolver_url, surl);
+
+    // Ask the resolver for the redirect chain (it follows 302s internally with
+    // cookies) — but we only want the final CDN stream URL. The resolver
+    // already streams; simplest reliable path is to just proxy the resolver's
+    // response body as-is.
+    let client = crate::infrastructure::utils::http_client::http_client();
+
+    let resp = client
+        .client()
+        .get(&fetch_url)
+        .header("User-Agent", "Mozilla/5.0")
+        .timeout(std::time::Duration::from_secs(120))
+        .send()
+        .await
+        .map_err(|e| AppError::ScraperError(format!("terabox proxy fetch failed: {e}")))?;
+
+    if !resp.status().is_success() {
+        let status = resp.status();
+        let body = resp.text().await.unwrap_or_default();
+        return Err(AppError::ScraperError(format!(
+            "terabox proxy returned HTTP {status}: {body}"
+        )));
+    }
+
+    let content_type = resp
+        .headers()
+        .get(reqwest::header::CONTENT_TYPE)
+        .and_then(|v| v.to_str().ok())
+        .unwrap_or("application/octet-stream")
+        .to_string();
+
+    let content_disposition = resp
+        .headers()
+        .get(reqwest::header::CONTENT_DISPOSITION)
+        .and_then(|v| v.to_str().ok())
+        .map(|s| s.to_string())
+        .unwrap_or_else(|| {
+            let fname = params
+                .filename
+                .filter(|s| !s.is_empty())
+                .unwrap_or_else(|| "terabox-download.bin".to_string());
+            format!("attachment; filename=\"{fname}\"")
+        });
+
+    let stream = resp.bytes_stream();
+
+    Ok(axum::response::Response::builder()
+        .header("Content-Type", content_type)
+        .header("Content-Disposition", content_disposition)
+        .header("X-Accel-Buffering", "no")
+        .header("Cache-Control", "no-store")
+        .body(axum::body::Body::from_stream(stream))
+        .map_err(|e| AppError::Internal(format!("build stream response: {e}")))?)
+}
+
+/// Params for the TeraBox streaming proxy.
+#[derive(Debug, Deserialize, IntoParams)]
+pub struct TeraboxProxyParams {
+    pub surl: String,
+    pub filename: Option<String>,
+}
+
 /// Download from Google Drive.
 /// `/download/gdrive?url=...`
 #[utoipa::path(
@@ -556,3 +680,62 @@ pub async fn download_bilibili(
     let result = use_cases::download_bilibili(&params.url).await?;
     Ok(Json(DownloadResponse::ok(result)))
 }
+
+// ========================================================================
+// Merged-file static serving
+// ========================================================================
+
+/// Serve a merged YouTube MP4 produced by `/download/youtube?merge=true`.
+/// Path is sanitised: only a plain filename (no `/`, `..`) inside the
+/// `yt_merge` uploads dir is allowed.
+///
+/// `/file/yt_merge/{filename}`
+#[utoipa::path(
+    get,
+    path = "/file/yt_merge/{filename}",
+    tag = "download",
+    operation_id = "dl_serve_merged_file",
+    params(
+        ("filename" = String, Path, description = "Merged MP4 file name")
+    ),
+    responses(
+        (status = 200, description = "Merged MP4 file", content_type = "video/mp4"),
+        (status = 404, description = "File not found or invalid name")
+    )
+)]
+pub async fn serve_merged_file(
+    axum::extract::Path(filename): axum::extract::Path<String>,
+) -> Result<axum::response::Response, AppError> {
+    // Reject any path traversal or nested segments.
+    if filename.contains('/')
+        || filename.contains('\\')
+        || filename.contains("..")
+        || filename.is_empty()
+    {
+        return Err(AppError::NotFound(format!("invalid file name: {filename}")));
+    }
+
+    let dir = std::path::PathBuf::from("/var/lib/scraper/uploads/yt_merge");
+    let path = dir.join(&filename);
+    let bytes = tokio::fs::read(&path)
+        .await
+        .map_err(|_| AppError::NotFound(format!("file not found: {filename}")))?;
+
+    let content_type = if filename.ends_with(".mp4") {
+        "video/mp4"
+    } else if filename.ends_with(".webm") {
+        "video/webm"
+    } else {
+        "application/octet-stream"
+    };
+
+    Ok(axum::response::Response::builder()
+        .header("Content-Type", content_type)
+        .header("Content-Length", bytes.len().to_string())
+        .header(
+            "Content-Disposition",
+            format!("attachment; filename=\"{filename}\""),
+        )
+        .body(axum::body::Body::from(bytes))
+        .map_err(|e| AppError::Internal(format!("build response: {e}")))?)
+}

src/presentation/router.rs

@@ -184,6 +184,10 @@ pub fn build_router(app_state: Arc<AppState>) -> anyhow::Result<Router> {
             "/download/youtube/mp3",
             axum::routing::get(crate::presentation::handler::downloader::download_youtube_mp3),
         )
+        .route(
+            "/file/yt_merge/{filename}",
+            axum::routing::get(crate::presentation::handler::downloader::serve_merged_file),
+        )
         .route(
             "/download/spotify",
             axum::routing::get(crate::presentation::handler::downloader::download_spotify),
@@ -204,6 +208,10 @@ pub fn build_router(app_state: Arc<AppState>) -> anyhow::Result<Router> {
             "/download/terabox",
             axum::routing::get(crate::presentation::handler::downloader::download_terabox),
         )
+        .route(
+            "/proxy/terabox",
+            axum::routing::get(crate::presentation::handler::downloader::proxy_terabox),
+        )
         .route(
             "/download/gdrive",
             axum::routing::get(crate::presentation::handler::downloader::download_gdrive),