v0.11.3

Changed

• Version is now baked into the source (internal/version, default
  v0.11.3, bumped on each release). Docker builds report the correct
  version with zero configuration — the VERSION build-arg /
  ARIAFLOW_VERSION env is now an optional override instead of a
  requirement.
• /start version line can also append the short VCS commit when available
  (local/git builds).

v0.11.2

Added

• Bot version in the /start welcome message — shows the build version
  (injected via -ldflags "-X ariaflow/internal/version.Version=…"),
  host uptime, storage free/total, and aria2 active/stopped counts. Docker
  builds accept VERSION build-arg (compose: ARIAFLOW_VERSION).

v0.11.1

Added

• Recovery for interrupted Telegram-forwarded downloads. If the bot dies
  after the local Bot API server finished fetching a forwarded file but
  before it was staged/organized, startup recovery now finishes it: it
  matches leftover server temp files to the task by exact expected size
  (persisted at registration time) and a stable age of ≥2 minutes, then
  stages + organizes them. Partial or size-unknown downloads are never
  touched.

Fixed

• Blank line between GID and the system footer in yt-dlp live messages —
  the CPU/DISK/RAM + DL/UL footer now sits one empty line below the task
  block, exactly like the aria2 task list.

v0.11.0

Changed

• Unified live download message structure. Telegram-forwarded files and
  yt-dlp downloads now render the exact same progress frame as normal direct
  URLs in the aria2 task list:
  Filename → progress bar + percent → Downloaded → Speed → ETA →
  GID, followed by the CPU/DISK/RAM + DL/UL system line.
  • Telegram-forwarded files (local engine) use the frame in the "Downloading"
    view, including the system footer when no aria2 tasks are shown.
  • yt-dlp live messages now use the same frame (previously a free-form
    "⬇️ Downloading / bar / URL" block), rendered with Markdown like the
    aria2 list, with the source URL as the Filename and the task GID shown.
  • Processing stage shows the same envelope with a Status line.
• System-info footer is cached for 5 s (cpu.Percent samples for 1 s;
  recomputing every tick would stall progress updates).

v0.10.0

Fixed

• Magnet .torrent metadata files are now cleaned up reliably. aria2
  writes <infohash>.torrent into the download folder
  (--bt-save-metadata); cleanup previously depended on the in-memory task
  record surviving, so after a restart the file lingered forever even with
  keepTorrent=false. Cleanup now falls back to aria2's own infoHash, and
  startup recovery sweeps orphaned <infohash>.torrent files for completed
  downloads.
• Task store is now actually persisted (tasks.json) — persist() was
  a no-op, so ownership, links and download history vanished on every
  restart. Tasks now survive restarts (atomic temp+rename writes).
• monitor can no longer panic the bot — GetDiskPercent dereferenced
  a nil disk.Usage result (e.g. USB/NAS mount hiccup) and
  GetCpuPercent indexed an empty slice; both now return 0 safely.
• TMAllowDownloads map race fixed — the torrent file-selection map was
  written from the Telegram handler and read from the testTMStop
  goroutine without a lock; Go maps are not concurrent-safe and could fatal
  with "concurrent map read and map write". Now mutex-guarded.
• Websocket RPC write errors are surfaced — a failed WriteJSON now
  triggers the reconnect loop instead of silently waiting out timeouts.
• Removed dead tool/controller code — legacy CopyFiles/RemoveFiles
  (panic-on-error, weak path guard) were referenced by nothing.
• runOrganizeDual deletes its "Download completed" messages when
  organizing is disabled instead of leaving them in the chat.
• Vet-clean RPC client — Call value receivers containing sync.Once
  flagged copylocks; switched to pointer receivers.

v0.9.0

Fixed

• Bot no longer crashes on transient aria2 RPC timeouts — the 5-second
  live progress poll used to panic on any RPC error (e.g. aria2 busy
  finalising a large file on a slow disk), killing the whole bot. RPC/data
  errors are now logged and the view degrades gracefully instead.
• Crash-recovery now covers all completed downloads, not just archives —
  if the bot dies after a download completes but before the organize pipeline
  moves it, the startup recovery re-runs the media pipeline for every
  completed aria2 task whose files are still in the download folder (torrent
  folders included), with a 7-day recency guard so old intentional leftovers
  are never touched.
• BitTorrent completion now triggers the organize pipeline — previously
  only onDownloadComplete did, so a torrent that finished and stopped
  seeding (e.g. seed-time=0) would silently stay in the download folder.
• Deduplicated organize runs — onDownloadComplete/onBtDownloadComplete
  and startup recovery can no longer process the same download twice.
• Release-group folder naming follows the media pipeline — movie release
  directories with site/group prefixes (www.Site.org - Title ...) now land
  in Movies/Title (Year)/Title (Year).ext derived from the video file,
  instead of polluting the folder name with the prefix.
• Log rendering of fatal errors fixed — %w is not supported by zap;
  panics now log a readable message instead of %!w(....).
• Torrent upload failures are non-fatal — a failed fetch of an uploaded
  .torrent file logs and skips instead of killing the process.

v0.8.0

Added

• Jellyfin/Plex library refresh — after organizing a download into a
  media-library category (movies/series/anime/music), the bot can now
  optionally trigger an on-demand rescan on Jellyfin and/or Plex so the new
  content appears immediately instead of waiting for the periodic scan. Both
  targets run concurrently with a short timeout; the outcome (success or
  failure) is shown on the final "All done!" summary. Configure via
  organize.refreshJellyfin/jellyfinUrl/jellyfinApiKey and
  organize.refreshPlex/plexUrl/plexToken (env: REFRESH_JELLYFIN,
  JELLYFIN_*, REFRESH_PLEX, PLEX_*).

v0.7.0

Added

• Docker Compose deployment — now the recommended path. A
  docker-compose.yml plus docker/ stack runs the bot, an aria2c
  daemon, and an optional local Bot API server as containers sharing one
  media volume; no manual binary or dependency installation.
• Project renamed to AriaFlow: the Go module, binary, and release
  artifacts are now ariaflow (ariaflow-linux-amd64 / -arm64), and the
  GitHub repository moved to spaulll/AriaFlow.
• Docker image bundles the full runtime toolset, including unrar (non-free)
  alongside yt-dlp, ffmpeg/ffprobe, 7z and unzip.

Changed

• README restructured to lead with the Docker quick start; the prebuilt-binary
  path is documented as the alternative.
• Containerized build no longer silently drops unrar (non-free repo enabled).

v0.6.1

Fixed

• Percent-encoded filenames (e.g. %20 for spaces) are URL-decoded before
  media routing: previously the trailing 20 of an encoded space could glue
  to the x264/x265 codec tag, faking an episode marker and mis-sorting a
  movie into Series/.../Season 20. Codec tags never count as episode
  markers now, and cleaned movie names carry no percent-escapes.

v0.6.0

Added

• Disk guard: links whose size is known via HEAD are refused up front
  with ⚠️ Not enough disk space when they provably don't fit the download
  filesystem. Unknown sizes (magnets, torrents) always allowed.
• Orphan sweeper: startup + hourly scan of the download folder deletes
  stale *.aria2 debris and their partials (no referencing task, older than
  30 min). Directories, lone data files, and anything live are never touched.
• Duplicate guard: warns before re-downloading — same link already active,
  or normalized show+season / episode / movie already in the library — with
  override buttons (never blocks: quality upgrades are legit). Opaque links
  (e.g. googleusercontent) are re-checked once the real filename arrives;
  on a hit the download pauses and asks to keep or stop+delete.