CHANGELOG.md

Changelog

All notable changes to this project are documented here. The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

Security

• A stored password or API key is now only ever sent to the host it was saved for. Testing or saving a connection with the secret field left blank fills it in from storage, and it then connected to whatever address the request carried — so anything that can reach the API could have Transpondarr hand your qBittorrent password, indexer key or ntfy token to an address of its choosing. Pointing at a different host now asks for that host's own credentials.

Added

• A .env.local file, read ahead of .env. A development convenience for running from a checkout: what is true of one working copy alone — a port, a local endpoint — goes there, leaving .env for what every checkout shares. A deployment should keep configuring through environment variables; neither file is committed, and neither reaches the Docker image.
• A title AniList publishes no episode count for can now be given one. Some older OVAs and specials have neither a count nor a broadcast schedule on record, which left them sitting in the library with no episodes, never searched for, and no way to fix it. The episodes tab now says the count is unknown and offers to create it, and the library list says the same instead of showing 0 / 0.

Changed

• The add dialog now says which episodes it will monitor, in both choices and before you pick one — All 12 episodes will be monitored, or 6 of 12 episodes will be monitored, from episode 7. When a choice would monitor nothing — a finished title, or one between cours — it says so, and says that an episode added later is still picked up.
• "Future only" is now "Only unaired", and the choice is no longer offered for a title that has not started airing, where both answers mean the same thing.
• Changing the admin password or authentication mode no longer holds up other requests. Saving either wrote to the database with every incoming request waiting behind it, so on a busy install one settings save could stall the whole app for a moment.

Fixed

• Two auth changes made at once can no longer disagree with what was saved. Changing the authentication mode from two places at the same moment could leave the running server enforcing one mode while the database held the other, so the mode changed on its own at the next restart with nothing having said so. Changing the admin credentials had the same gap.
• The Wanted page's Cutoff Unmet tab no longer slows down as a library gets healthier. It works out which held episodes fall below their profile's cutoff every time it is opened, and did so from scratch each time — so the more of a library that had reached its cutoff, the more work it did to show nothing. It now remembers what it read from each release name, and only the cutoff itself is worked out afresh, so editing a quality profile still moves the list straight away.
• Checking the release feed does much less work on a large library. Every release on the feed was being read once per tracked title rather than once, and episodes already in the library were being examined even when the quality profile has upgrades switched off. Neither happens now, so each check costs a fraction of what it did.
• The Calendar's Unmonitored filter can now actually reveal a title. An unmonitored title never received broadcast times, so the filter could only surface one that had been monitored at some point — and the No schedule data footer that exists to explain such absences left it out too. Air dates and metadata now keep up to date for unmonitored titles, behind every monitored one.
• The Calendar no longer says AniList published no air dates for a title it has not looked up yet. A title has no broadcast times until the airing sync reaches it — minutes for a freshly added one, longer after a bulk add — and it sat under No schedule data the whole time, beside titles that genuinely have none. Those now read Not checked yet until they have been.
• When a metadata provider fails, it now says why. Adding a title while AniList was down reported only Could not add title / failed to add series, while AniList's own explanation — The AniList API has been temporarily disabled due to severe stability issues — was dropped on the way through. Errors from anywhere in the API now carry the underlying cause, and a validation failure names the field it came from.
• The airing sync no longer reports the same failure once per series. During a provider outage its last error was a dozen identical copies of one message; it now reads once, with a count.
• A title added before it starts airing now monitors its episodes. Adding an upcoming series or film and choosing to monitor only what is still to come left every episode switched off, so it sat at 0 / 12, was never searched for, and gave no sign why. Nothing has aired yet means there is no back catalogue to skip, so all of it is monitored.
• Breaking: a settings save can no longer change a setting it never mentioned. Leaving a field out of a PUT /api/v1/settings/* body quietly reset it to the default — a flat library reverting to season folders, a custom qBittorrent category becoming transpondarr, a self-hosted ntfy server becoming ntfy.sh — and the same omission on POST /api/v1/auth/mode switched a local-network install back to requiring a login. Each of those fields is now required, so an incomplete body is refused with a 422 instead. Nothing changes in the web UI, which has always sent whole sections, but an API client that sends partial bodies must now send whole ones.
• Breaking: the same applies to quality profiles. Saving one over the API without mentioning upgrades switched them off, and creating one without mentioning resolutions or the v2 carve-out started it with no resolutions ranked and the carve-out off — not the defaults a profile created in the UI starts from. Those fields are required now too.
• A finished title whose episode count AniList never publishes no longer costs a metadata request every six hours forever. It is now re-checked weekly, which is often enough to notice a count that does arrive and rare enough to stop eating the request budget for one that never will.
• A missed page of the recent feed is now spotted even when a stale entry sits on it. When the feed moved on faster than Transpondarr could poll it, a single entry that had been on the page all along — an older release from a tracker catching up, or a dateless sticky post — was enough to hide that anything had been missed, and the episodes that slipped past waited for the scheduled search to come round to them, up to a day later. Those series go back to the front of the search queue as soon as the poll notices.

0.9.0 — 2026-08-15

Acquisition reliability: a download that goes nowhere stops holding its episode, and a release stops being blamed for things that were never about it. A torrent transferring nothing is given up on after six hours and the next-best release tried; a torrent that vanishes, loses its files, or arrives in a format Transpondarr cannot follow no longer blocks the release.

Fixed

• A download that never gets going no longer holds its episode forever. A torrent that finds nobody to download from, or a magnet stuck at Downloading metadata, used to sit in the queue indefinitely with no other release tried. One that has transferred nothing is now given up on after six hours and the next-best release taken. Configurable under Settings → Download client, where 0 waits forever.
• A download that has made any progress is never given up on, however long it then sits, and paused or queued ones are left alone. Nothing is removed from your download client — cancelling stays yours to do there.
• Every download now shows how long it has left. A new one reads giving up in 6h until its first bytes arrive: what would happen if nothing ever came, not a sign that anything is wrong.
• A download waiting its turn now reads Queued. It showed as Downloading · 0%, indistinguishable from one getting nowhere — which matters most if you cap how many downloads run at once.
• A release is no longer blocked because its torrent went missing. A download that disappears from the client, or whose files vanish from disk, still frees its episode — but the release is no longer remembered as a bad one, because neither says anything about it. A dropped mount used to block every release it touched, for a day, a week, or permanently on a third occurrence.
• A download whose data the client has lost now reads Data missing, and the next search moves on to another release. Grabbing such a release by hand fails outright rather than reporting success and quietly failing.
• One failed download is now remembered once, not twice. When a pack's episodes were added to the queue at different times, a torrent that vanished counted as two separate failures of the same release, pushing it up the blocklist's ladder twice as fast.
• A release in a torrent format Transpondarr does not support is refused when it is grabbed. A download published in BitTorrent v2 alone was accepted by qBittorrent but could not be followed, so it sat unaccounted for until the grace period expired — then was blocked for something never wrong with it.
• An import interrupted mid-transfer no longer leaves a file in the library. Transpondarr renames a temporary file into place on completion, so a crash never leaves a half-written episode under the real name — but the temporary file stayed. A new Library tidy job removes them once they are a day old.

Upgrade notes

• Downloads that have been stuck at 0% will start being given up on. The six-hour timeout is counted from the upgrade, not from however long a torrent has already sat, so one transferring nothing is failed six hours after you upgrade and its release blocked for a day. To hold one deliberately, pause it in your download client or set the wait to 0; either stops the clock.
• Releases already blocked under the old rule stay blocked. Entries recorded when a torrent went missing or lost its files carry over, including any at the permanent rung. A title's History tab lists them under Blocked releases with an Unblock on each, and Settings → Failure memory has a Forget all.

0.8.0 — 2026-08-14

Films: the release where Transpondarr tracks both anime formats as themselves. Add a film from AniList and it is searched on its title and release year, grabbed, and filed into a Movies library of its own; episodes keep going where they went before, and gain a flat filing option for the scanners that prefer it. The REST resource series is also now titles, which API clients need to know about.

Quality profiles are now chosen when you add a title rather than reassigned afterwards, and the controls around them hold up — the chip survives a slow or failed load, and names no longer collide only by case.

Added

• Anime films can be added, searched and grabbed. A film in AniList search or on the Discovery chart is no longer reserved and greyed out: adding one tracks it like any other title, and the same pipeline searches, grades, grabs and imports it. A film's releases are matched on its title and release year, which is what they are named for, so a manual search lists what is available and a monitored film joins the scheduled search queue and the recent-feed poll alongside your series, backing off the same way when nothing turns up. Format decides and episode count never does, so a one-episode OVA or special is a series and stays one. Two things to expect: automation waits for a release year rather than guessing at one — a film whose year AniList has not published yet is listed as ineligible until a metadata refresh fills it in, and a release naming a different year is listed with that mismatch as its reason — and a batch or season pack is held back for the same kind of reason, being as likely to be the parent series' pack as a multi-part release of the film. Searching and grabbing by hand work throughout, in both cases. Adds migration 00022.
• Films import into their own library directory. Plex and Jellyfin want a Movies library separate from Shows, so Settings → Library now takes a movies directory alongside the existing one, and a film places into it as Placeholder Film (2019)/Placeholder Film (2019).mkv, with the year left out of the name when the provider publishes none. Until that directory is set, a grabbed film waits in the Activity queue and says so rather than landing in the wrong library; setting one imports it on the next scan, and adding a film with no movies directory configured says so on the add form, so it is clear before the download finishes rather than after. A download that ships the feature alongside deleted scenes, an interview or a making-of imports the feature — the largest video in the payload. When two videos are the same size nothing is guessed: the import waits in the Activity queue for you to say which is the film.
• Films read as films across the web UI. A film's page carries a status card — wanted, downloading or in the library — with the release year in its header, and its add form asks only whether to monitor it. The library list reports what a film is doing: Downloading, Downloaded, not imported, Import blocked, In library, Wanted, or Not monitored when nothing is pursuing it. The calendar shows a film's premiere as a premiere, and places far more of them — the release date AniList publishes now fills in where a broadcast schedule never existed, so a dated film leaves the No schedule data footer and lands on its day, while one announced with only a year stays in the footer rather than being placed on a date nobody published. A film's Releases tab lists everything the search returned, and its Search button on the Wanted page links straight there. Wanted says Not released yet and states the date plainly; the Activity queue, the import notifications, the delete confirmation and the release-match reasons all name the film. Series screens keep their wording throughout, single-episode OVAs and specials included.
• Episodes can be filed without season folders. Settings → Library now takes a Series layout: keep the default season folders, or choose flat and have episodes land in Placeholder Saga/Placeholder Saga - S01E05.mkv for the scanner configurations that prefer it. Films are unaffected, and so are single-episode OVAs and specials, which stay series-shaped as before.

Changed

• A title's page has moved from #/series/:id to #/titles/:id, and the web UI calls it a title. The URL now matches what the rest of the product calls it. There is no redirect from the old address, so a saved link or a browser-history entry pointing at #/series/:id will not resolve — open the title from the Titles list instead. Every link inside the app already points at the new address. The sidebar and page title now read Titles to match, while genuinely episodic screens keep their wording.
• A title page's link parameters now do one job each. ?tab=releases opens the Releases tab and ?item=N filters it to one episode, where a single ?item=N used to do both. A saved link or bookmark carrying only ?item=N now opens the title on its usual landing tab rather than on Releases; a series' Search button on the Wanted page sends both, so it is unaffected.
• The REST resource is now titles, not series. An endpoint called series that returns films would be a wart the 1.0 stability promise freezes in place. Every /api/v1/series... route is now /api/v1/titles..., and the fields that name or reference one follow — in the webhook payload as well as the REST API, so the two agree. Discord and ntfy now announce Title added rather than "Series added". Nothing else changes — same behaviour, same data, same database, and your notification settings carry over untouched.
• Adding a title now opens a form for it, carrying both add-time decisions. The monitor choice was a single control above the whole result list, and the quality profile was not asked for at all — an added series took the default and had to be re-assigned from its own page afterwards. Clicking Add on a result now opens a form for that title with Monitor and Quality profile on it, defaulted for the common case, so a seasonal show is still one confirmation and a long-runner is narrowed before its first search. The global "Monitor on add" control and the Add · future only button annotation are gone, and Discovery opens the same form.

Fixed

• The quality-profile chip on a title's page no longer vanishes while it loads or when it fails. The picker rendered nothing until the profile list arrived, so the chips row shifted on every visit, and a failed fetch left no trace at all — just a missing control. It now holds its place while loading, shows a Profile unavailable chip you can click to try again, and links to Settings when there are no profiles to pick.
• Profile names no longer collide only by case. Saving a profile named Anime while anime exists is refused with the same conflict a duplicate name has always given, matching how group names within a profile are compared.

Internal

• Built with Go 1.26.6. The 1.26.5 standard library carries six vulnerabilities the release binary reaches — in net/http, crypto/tls, net/url, encoding/xml and encoding/asn1 — all fixed upstream.
• Duplicate-name detection reads SQLite's result code instead of matching on the error message, so a driver that rewords its errors can no longer turn a conflict into a 500.
• The profiles list reads groups and usage counts in bulk. GET /api/v1/profiles fetched both one profile at a time, so every profile cost a full scan of the series table. Groups and counts now come back in one query each, leaving the endpoint at three queries whatever the profile count.

Upgrade notes

• Set a movies directory before a film can import. Settings → Library takes one alongside the existing library directory (or TRANSPONDARR_LIBRARY_MOVIES_DIR); with it unset, films are still added, searched and grabbed, but a completed film waits in the Activity queue instead of being placed. Point it at a root your media server has as a Movies library, under the same mount your downloads use so the hardlink works. Nothing about episodes changes if you never set it.

• Saved links into a title's page need re-making. #/series/:id has no redirect to #/titles/:id, and a bookmark carrying only ?item=N now opens the title on its landing tab rather than on Releases. Links inside the app are already updated; only your own bookmarks and browser history are affected.

• The new series layout setting defaults to season folders, so nothing moves on upgrade. If you switch it later, it applies to future imports only: episodes already in your library stay where they are, and an upgrade of one of them lands in the new shape and leaves the old file behind, which the log warns about. API clients that write library settings must now send series_layout with the rest of the body: omitting it is refused rather than quietly resetting the layout to season folders. The web UI needs nothing.

• API clients using X-Api-Key need one search-and-replace; the web UI needs nothing. The UI ships inside the binary and is already updated, so only machine clients notice. Every route under /api/v1/series moved to /api/v1/titles — including {id}, {id}/search, {id}/grab, {id}/grabs, {id}/blocklist, {id}/blocklist/{entryId}, {id}/pinned-group and {id}/profile — and these JSON fields were renamed:

  Before	After	Where
  series_id	title_id	activity queue and history, calendar, wanted groups, discovery entries, pinned-group and profile-assign responses
  series_title	title	activity queue and history, calendar, wanted groups
  series (list)	titles	GET /api/v1/titles
  series (name)	title	title search, grab history, title blocklist
  series (count)	titles	GET /api/v1/blocklist
  series_ids	title_ids	POST /api/v1/wanted/search
  series_queued	titles_queued	POST /api/v1/wanted/search, PATCH /api/v1/wanted/items
  series_count	title_count	GET /api/v1/profiles
  on_series_added	on_title_added	notification settings
  series_title	title	webhook payload
  event: "series_added"	event: "title_added"	webhook payload

  Generated clients also see renamed operation ids (list-series → list-titles, and so on) and schema names (SeriesDTO → TitleDTO, AddSeriesInputBody → AddTitleInputBody, and so on); regenerate against the new spec. Nothing is dropped alongside the renames — the series database table keeps its name and its columns, and saved notification toggles are read under their existing keys, so nothing needs reconfiguring.

  Films add fields rather than changing any: titles carry format and year, the calendar, wanted and activity rows carry format so a client can word a film as one, library settings take movies_dir, and adding a title accepts quality_profile_id. All are optional on the way in, so a client that ignores them behaves as it did.

  One rename fails quietly rather than loudly, so check it if you script your settings: a PUT /api/v1/settings/notifications that still sends on_series_added turns that notification off, because the toggles are plain booleans with no unset encoding and a save replaces all of them. Every other rename here reports itself — an old route 404s, and series_ids is a required field, so POST /api/v1/wanted/search returns 422 rather than resetting your whole library's search backoff.

0.7.0 — 2026-08-08

Monitoring, per episode: the release where you choose which episodes Transpondarr chases, not just which series. Unmonitor a recap, an episode you already have, or a thousand-episode back catalogue, and automation stops looking for it — and every progress count now reads against what you actually asked for rather than against every episode a series will ever have. Activity also gains a view of the downloads nothing is waiting on.

Added

• Per-episode monitoring. Monitoring was all or nothing per series: once an episode existed, the sweep chased it until a file landed — forever. Every episode now carries its own flag, so unmonitoring a recap, an episode you already have, or 1,038 episodes of a back catalogue takes them out of automation for good. Adding a series asks which episodes to monitor — all, or future only from the next broadcast onwards — and the answer keeps applying to episodes announced months later. The Episodes table gains row checkboxes and a bulk monitor/unmonitor toolbar; Wanted gains a per-row toggle and an unmonitored reason, with those rows behind the existing Unmonitored chip rather than removed, so the click is undoable. A manual search and a manual grab still work on an unmonitored episode, and a pack grabbed for its monitored episodes still imports every file it carries. Adds migration 00021. Closes #188.
• Downloads nothing is waiting on are now visible in Activity. A torrent could be left in the download client with nothing pointing at it — after a second grab replaced the first, or after deleting a series and choosing to keep its downloads — where it kept downloading, took up disk, and could not be seen or stopped from Transpondarr. Activity gains an Unmatched downloads section listing them with their size, age and state, each removable by hand with the option to delete its data too. Nothing is removed automatically, and only Transpondarr's own download category is ever listed — torrents outside it are yours and are never touched. Closes #131.

Changed

• A series' progress is now measured against what it is pursuing, not against every episode it will ever have. The denominator counts episodes that are monitored and already broadcast, following Sonarr, with the raw total kept beside it: a 12-episode show three episodes in reads 3 / 3 (12 total) instead of 3 / 12. A series with nothing aired yet says "Nothing aired yet (12 total)" and one with nothing monitored says "Nothing monitored", rather than 0 / 0; "Search all wanted" is disabled in both cases, because there is nothing to search for. The episode table's summary strip splits the same three groups — tracked, not yet aired, not monitored.
• The item status have is now in_library, and so is the field it derives from. have was the odd one out in a vocabulary of downloading / stuck / deferred / wanted, and the web UI has always rendered it as "In library" anyway, so this is a rename in the API, not in what you see. The target is deliberately not imported: adopting a pre-existing library would put an episode in the library without the importer touching it. The wanted_items.have column moves with it (migration 00019). No behaviour changes. Closes #84.
• A tracked title is now identified by which provider it came from and its id there, not by an AniList id column. series.anilist_id becomes the pair (provider, provider_id) — the same shape the metadata cache has always used. AniList remains the only provider and every existing row migrates to it, so nothing about behaviour changes; what changes is that the database no longer names its upstream in a column. Rebuilds the series table (migration 00020). Closes #74.

Fixed

• An episode you already have can now be searched from the Episodes tab. The row's Search button appeared only on wanted and deferred episodes, so an episode already in your library could not be searched from the series page at all — even though Cutoff Unmet linked to the same view and the server had always accepted the request. It is now on every row, in-flight grabs included. Closes #195.

Upgrade notes

• Every series card's progress changes on upgrade, including ones with nothing unmonitored. Nothing is recalculated or written — the denominator is simply counted differently, and the old number stays visible as the total in parentheses. The API keeps total unchanged and adds tracked and monitored_items beside it, so a client reading the old field still gets the old number. in_library does change: it now carries the same monitored-and-aired filter, so a held unaired or unmonitored episode cannot push a series past its own denominator.

• API clients using X-Api-Key need one search-and-replace; the web UI needs nothing. Three fields change on the JSON responses:

  Endpoint	Before	After
  GET /api/v1/series	have: 12 (count)	in_library: 12
  GET /api/v1/series/{id}, POST /api/v1/series	have: true (per item)	in_library: true
  GET /api/v1/series/{id}, GET /api/v1/calendar, GET /api/v1/wanted/cutoff-unmet	status: "have"	status: "in_library"

  The other status values are untouched. The migration renames the column in place, so no data moves and a downgrade to 0.6.x needs goose down for migration 00019 as usual.

• Breaking API change: adding a series takes a provider pair. POST /api/v1/series now expects {"provider": "anilist", "provider_id": 123, "monitored": true}. anilist_id is gone, with no compatibility alias, and provider is required rather than defaulted — a default would hide which id space the caller meant, which is the ambiguity this change exists to remove. A request in the old shape is rejected with 422.

  The series responses change to match: GET /api/v1/series/{id}, the add response, GET /api/v1/metadata/search and GET /api/v1/browse/season all emit provider and provider_id in place of anilist_id. The web UI ships inside the server binary and is updated in lockstep, so this only affects scripts and third-party clients calling the API directly.

• The migration rebuilds the series table. It is applied automatically on first start and preserves every row, including everything hanging off a series — wanted items, grabs, grab history and blocklist entries. As with any schema-changing release, back up transpondarr.db before upgrading; downgrading to 0.6.x is supported and converts the pair back.

0.6.0 — 2026-08-06

Completeness: the release where you can see what the library is still missing, and finish it. v0.5.0 told you what automation did; this one answers the harder question — what it hasn't done, and why. A new Wanted page lists every episode still missing across the library with the reason for each one, down to the release the last search found and turned down. Season packs now import episode by episode, so a back catalogue arrives in one grab instead of a dozen. And an episode you already hold can be re-grabbed when a better release appears — opt-in per quality profile, bounded by a cutoff.

Added

• Wanted is a real page: what the library is still missing, and why. Two tabs. Missing lists every episode still worth acquiring, grouped by series with the newest gap first, and says why each one is still missing — automation is off, the series is unmonitored, blocklisted or waiting its turn in the search queue, the last grab failed, or it simply has not aired. Unaired episodes and unmonitored series sit behind filters, since the Calendar owns the forward-looking view. Cutoff Unmet lists episodes you hold that score below their quality profile's cutoff, with the release you have and what the profile still wants from it ("group TopSubs (+100) · resolution 1080p (+100)"). An episode's Search opens the Releases tab focused on that episode; selecting series and hitting Search selected or Search all puts them back at the front of the search queue and starts a run — the confirmation says queued rather than done, and under notify-only that the run will grab nothing. Series headers stay in view while you scroll, groups collapse once you are done with them, and both lists page as you go rather than loading a whole library at once.

  Missing also says when a search found releases and turned them down. An episode blocked by your profile's floor, a blocked group or an exclude used to look exactly like one nothing had searched for yet — and those are the ones where you have something to change. Each episode carries what the last pass decided about it: the release it declined and why, the pinned group it is waiting for, the one the download client refused, or that a search genuinely found nothing. Under notify-only, a rehearsal's decisions show up here too rather than only in your notifier. It is the last pass's answer, so the badge is dated — "Releases declined · 2h ago" — and a grab since then replaces it. Adds migration 00018 (pass_outcomes).

• An episode you already have can be re-grabbed when a better release turns up. Off by default and enabled per quality profile: Upgrade until cutoff in the profile editor, with the cutoff picked from the same landmarks your score weights compose ("top group, best resolution") rather than typed as a bare number. It is a cutoff, not a chase — while what you hold scores below it, any strictly better release replaces it; once it reaches the cutoff, the episode is left alone forever. A second switch keeps taking the same group's v2 or repack of the very file you hold even above the cutoff, since that is a fix for a broken release rather than a better one. Upgrades ride the recent feed, which costs one request for the whole library; the scheduled search never spends a search on a complete series, though a series it searched anyway takes an upgrade it happens to find. A manual search now offers releases for episodes you already have too, and grabs them without asking about the cutoff — profiles inform manual actions and gate only automation. The library file is replaced in place, with anything the superseded release left under the same episode name (a different container, a sidecar) cleared with it; a failed or unresolvable upgrade leaves the episode exactly as it was. Nothing removes the superseded torrent from your download client. Adds migration 00017 (upgrades_enabled, cutoff_score, upgrade_v2_above_cutoff on quality profiles, and the release each held episode is holding, backfilled from its import).

• Season packs and batches now import episode by episode, and automation prefers them. A multi-episode payload used to settle every grab row it covered as "downloaded (batch)" — terminal, with the bytes on disk and the filing left to you — and automation was deliberately kept away from packs for exactly that reason. The importer now walks a completed payload once and maps its files onto the episodes the release covered, placing each on its own, so the refusal is gone: a back-catalog series that has both singles and a pack takes the pack, one grab instead of one per episode. Ranking gained a coverage tier to make that happen deliberately rather than by seeder count, below a pinned group, and a weekly single-episode release still ranks purely on your profile score. A pack whose numbering runs past the entry — a 01-48 pack against a 12-episode season — is now refused as a possible absolute/season mismatch instead of silently claiming episodes 1-12.

• A payload file for an episode the release never claimed is imported too, when that episode exists, is still wanted, and has no download of its own — the release titled 03 that ships 03 and 04.

• A stuck import can be fixed by hand from the Activity queue. "Needs import fix" rows get a Fix import action listing every file in the payload with what its name parsed to, and a per-file episode picker preseeded with the importer's own suggestion. Use it when a filename is unreadable and nothing could be matched to it. Automation never reopens a settled import on its own.

• An episode's Search button now searches for that episode. It used to hand you the series-wide result list and leave you to scan it for the row whose Match badge said E7. The Releases tab now opens pre-filtered to the candidates that cover the episode you asked about — a single or the batch that contains it — with a dismissible chip naming the focus and a count of what is hidden. Nothing covering it says so, and offers the full list in one click. The search itself is unchanged and still series-wide: Search all wanted and clicking the Releases tab directly both stay that way, and a grab is still decided by the server.

• The indexer can be pointed at specific Newznab categories. A new Categories field on the indexer settings (env TRANSPONDARR_TORZNAB_CATEGORIES) takes a comma-separated list of IDs — anime is usually 5070 — and sends it as cat= on every search and on the recent feed. It matters most for the feed: one page is about 100 entries covering the whole endpoint, so on a general-purpose indexer music, books and software eat the window an anime release needs to be seen in. Empty is the default and keeps today's behaviour exactly — every request unfiltered — so nothing changes until you fill it in. A cat you baked into a Prowlarr or Jackett feed URL is left alone while the field is empty.

Changed

• The Calendar's "Unmonitored" control is now a filter chip rather than a switch. A switch says you are changing a stored preference — which is what Monitored on a series is — and using the same control for a view filter made both mean less. It is now an "Unmonitored" pill under an Include label, lighting up when on and matching the new Wanted page. Same behaviour, same keyboard and screen-reader semantics.

• An episode you have with a download in flight now reads as "downloading" rather than as simply had — the state an upgrade in progress puts it in.

• One import notification per release, not one per episode. A pack landing six episodes now sends a single "Import succeeded" naming the range (Episodes 1-3, 5). The generic webhook's payload gains an items array carrying the raw numbers — always present, empty for a single-episode import, with item_number then 0.

• "Downloaded (batch)" is now "Needs import fix", and says which episode had no file and how many files were left unmatched. It means one file could not be picked out of a payload, not that a whole batch was refused. When nothing in the payload matched an episode, that episode now goes back to wanted and the release is remembered, so the next search picks a different one instead of parking the item forever.

• A download that holds only an archive now says so, and can be finished by hand. Transpondarr does not unpack archives — a scene-style RAR set used to settle with a flat "the payload holds no video file", and Fix import then opened on an empty list, which was a dead end. The deferral now names the archive and how many volumes it spans (a 12-volume set counts as one thing to extract, not twelve), and the dialog lists it with what to do: extract it into the download folder and retry, which imports the extracted episode. An archive is listed beside the payload's files and can never be assigned to an episode. An episode an archive holds stays waiting for you on every path — retrying before you have extracted anything tells you so and leaves the download where it was, and a download holding one loose episode plus an archived one keeps the second waiting instead of giving up on it. Password-protected and corrupt archives are indistinguishable from healthy ones without an unpacker, so all three defer the same way.

Fixed

• A single-file .rar download is no longer filed into the library as the episode. A payload that is one plain file was taken as the episode whatever its extension, so a torrent containing a lone archive was hardlinked into place under an episode's name — a file no player can open, with the episode marked as had. Such a payload now defers as the archive it is.

• A downloaded episode whose filename happens to contain an extras word no longer parks itself. The payload resolver drops files marked preview, promo, nc, ncop and similar so a bundled creditless opening is never mistaken for the episode. When those words appear in the episode's own title, a folder holding exactly one video filtered to nothing and the grab settled as deferred — settled for good, with the file sitting right there. A payload whose sole video carries such a word is now imported. A payload where several videos all carry one still defers, since there is nothing to distinguish them, and a file marked sample is still never imported however alone it is: it is a truncated copy of the episode, not a title that happens to read that way.

• A releasing title whose episode count AniList never publishes now adds with its episodes. Such a title previously created a series with no wanted items and sat at 0 / 0 until a background pass caught up, because the add asked only for the episode count — precisely the field that is null. AniList exposes the broadcast schedule as a field on the title rather than a separate query, so one page of it now rides along in the request the add already makes: zero extra API calls, and the item set arrives immediately. Separately, the schedule skips an episode when two share a broadcast slot — it reads 1, 3, 4 — so both the add and the background sync now fill the gaps rather than transcribing, which is the only thing that creates an episode nothing else ever claimed should exist. A filled-in episode has no air date (a normal state here) and resets the series' search cadence, so it is looked for on the next pass instead of waiting out an accumulated backoff.

• The feed poll no longer spends AniList requests on title variants. Every poll resolved each wanted series' name variants through the metadata provider, and currently-airing titles — exactly the ones carrying wanted items — expire from the cache every 6 hours, so a library of airing shows could blow AniList's budget on a single 15-minute tick. The poll now reads variants from the local metadata cache only, serving even a stale snapshot (names don't go bad the way episode counts do); a series with no snapshot still matches on its stored title, and the bounded sweep covers it with full variants.

• An episode that fell through a gap in the recent feed is now searched within the next sweep, not up to a day later. When a busy indexer publishes more than one page of releases between two polls, the poll recognises nothing on the page it fetches and knows its coverage was broken — but it only warned, leaving whatever aired in the meantime to the scheduled search, which with a feed configured no longer aims at broadcast times and can be a full day away. Such a poll now puts the affected series back at the front of the search queue: monitored series still missing an episode that aired inside the gap (plus an hour of slack before it, since a rip is published after it airs). The reset is deliberately capped at five series per gap event — the same number one scheduled pass searches — and skips series already due, so a routine gap on a high-volume indexer never queues more searching than the schedule can spend.

Upgrade notes

• A long-running series already in your library will gain its full back catalogue. This affects monitored series whose AniList entry publishes no episode count — in practice the very long ones (One Piece, Detective Conan, Pokémon, Crayon Shin-chan). AniList keeps only a recent window of broadcast records for these (One Piece's first page starts at episode 1123), so until now everything below that window was created by nothing at all and the series quietly tracked only its recent tail. On the next metadata refresh or airing sync each one materializes its whole run — for One Piece, about 1,170 wanted items in place of roughly 25.

  Nothing is grabbed by that write, but the new items are wanted and searchable immediately, so expect the series to read as far from complete and to stay in the search queue while it drains. It costs no additional AniList requests, and no additional indexer load per episode either: a sweep spends one search per series, whatever its item count. If you would rather not track a long-runner's back catalogue, unmonitor the series before upgrading — an unmonitored series is skipped by the metadata refresh and the airing sync, which are the two passes that would create the items, and by the search sweep that would go looking for them.

0.5.0 — 2026-08-02

Visibility: the release where unattended acquisition stops being silent. v0.4.0 made Transpondarr act on its own; this one tells you what it did — pushed to Discord, a webhook, or ntfy, and collected in an Activity feed — and lets you rehearse the whole thing first. Turn automation to notify-only and watch a week of real decisions across your library without a single byte reaching the download client.

Added

• Notifications, with Discord, generic-webhook, and ntfy adapters. Once acquisition runs unattended, silence is the default failure mode: a stuck import, a failed grab, an episode landing — all of it previously visible only by opening the UI. A new Notifier seam carries one structured event that each adapter flattens natively: Discord to a colored embed with per-field detail, the generic webhook to a documented JSON contract you can script against, ntfy to title/priority/tags. ntfy is first-class rather than reached through the generic webhook because its priority mapping is the feature — a stuck import buzzes at high priority, an episode landing does not. Six event kinds (grabbed, imported, import stuck, grab failed, series added, and rehearsal), per-adapter per-event toggles, and a test button per adapter so a config can be verified without waiting for a real event. Delivery is fire-and-forget: a failing notifier logs and never blocks or fails the pipeline that triggered it. A manual grab stays deliberately silent — you were there.
• Notification-only mode: rehearse the sweep without grabbing. The global automation toggle grew a third state — off / notify-only / on. In notify-only the sweep and feed poll run for real (search, decide, cadence, pinned-group holds) and report what they would have taken, without recording a grab or touching the download client. The negative half reports too, which is the more useful half: no eligible candidate, held for a pinned group, refused by the season-pack guard — each with its reason. A rehearsal is a firehose by design, so ntfy maps it to low priority. Switching to on afterwards resets every series' search cadence, because a rehearsal settles nothing and would otherwise leave the library backed off for releases the feed has already consumed.
• Activity page: the global download and import feed. The placeholder is now the answer to "what did automation do while I wasn't looking?" — a queue of every in-flight grab across the library, joined with live client-reported state, and a paginated history feed of grab and import events, newest first. The queue surfaces paused, stalled, and checking per row, which the derived status vocabulary could not express: a paused torrent previously read as "Downloading" forever. With the download client unconfigured or unreachable it degrades to grab state rather than erroring.
• Per-job "Run now" from the Background jobs card. The runner's trigger has been implemented and tested since v0.3.0 with no caller; it now has a button. A manual run bypasses the automation kill switch on the same precedent that governs manual search and grab — an explicit action is intent, not something to gate — so the card confirms in words when a hand-triggered sweep will grab for real. Every eligibility rule still applies, and a run in notify-only rehearses like any other, so the mode's guarantee holds no matter who triggered it. The trigger only queues, so the button never pretends to wait for a result.
• Delete a series. Anything added previously stayed forever, occupying the library list and keeping the sweep searching. DELETE /api/v1/series/{id} removes the series with its episodes, grab history, and blocklist memory in one transaction, behind a confirmation that names what goes and what stays. Library files are never touched — a decision, not an omission: deleting media is a bigger call than deleting a tracking row. An optional flag also clears the series' torrents from the download client; without it they are left seeding.

Changed

• Breaking: PUT /api/v1/settings/automation now takes mode (off | notify_only | on) in place of the enabled boolean, and GET /api/v1/settings reports it the same way — an old client body is rejected. Stored values and TRANSPONDARR_AUTOMATION_ENABLED still accept the old booleans, so no configuration needs migrating.
• A re-grab no longer erases the attempt before it. The per-series History tab now reads an append-only event log instead of the single current grab row, so a completed lifecycle shows both "Grabbed" and "Imported" and a replaced release keeps its record. "Import blocked" leaves history in the process — it is live state, owned by the Episodes tab and the Activity queue, not an event that happened at a point in time.

Fixed

• The Discovery page built 88 hidden year options on every render. Radix keeps a closed Select's items mounted so the trigger can resolve its own text, so the year picker's full 1940-onward range was constructed on every render — about half the page's mount cost, for a dropdown that is rarely opened. The items now render only while the menu is open.

Internal

• An append-only grab_events table records the lifecycle the grabs table structurally cannot: one row per wanted item, overwritten by the next attempt. Events are written at the two convergence points — inside the grab transaction, and as the importer settles — with the importer's writes best-effort, because history must never wedge the pipeline. The migration backfills from surviving grab rows so an upgrade does not start empty.
• The API's first pagination, keyset-cursor over (created_at, id) with an opaque cursor, set as the precedent for the endpoints that follow.
• The notification dispatcher fans out under context.WithoutCancel with its own timeout, so neither a request-scoped context nor a shutdown cancels an in-flight send, and a hung endpoint cannot leak a goroutine.

0.4.0 — 2026-08-01

Monitoring and scheduled search: the release where the monitored flag starts doing real work. Add a series, touch nothing, and episodes arrive — searched on a broadcast-aware cadence, graded against your quality profile, grabbed, imported, and remembered when they fail. Automation ships off by default; enable it under Settings → Automation.

Added

• Scheduled search sweep. A wanted-search job periodically searches monitored series with outstanding wanted items once their episodes have aired (an unknown air date counts as searchable — absence is normal, not an error), runs the same match-and-score path as manual search, and grabs only a fully eligible release: matched, clear of excludes, over the profile floor. No eligible candidate means grab nothing — "nothing yet" is a correct outcome. Empty passes back off exponentially from an hour toward a day, clamped so a weekly show is still searched at broadcast however many empty passes came before; a new episode airing resets the clock, and integration edits in Settings apply on the next pass without a restart.
• Recent-feed polling. The sweep costs one indexer query per series, so its interval must be long; the Torznab recent feed inverts that — one request returns the newest releases across the endpoint, matched against everything wanted at once. The feed is the hot path, polled on a short interval, and the sweep becomes the safety net. A new release for a wanted, aired episode is grabbed on the next poll, not the next sweep. An indexer without a feed degrades to sweep-only as a supported configuration, and a quiet poll costs one request.
• A pinned group can now mean "wait for," not just "win when present." The sweep holds another group's release for a delay window measured from the episode's broadcast, so the pinned group's slower release gets its chance before automation settles. Global default plus a per-series override; a manual grab is never held.
• Automation controls. A global auto-search toggle and the default pin delay, persisted and editable live in Settings → Automation — flipping the switch takes effect at the next tick, off or on, without a restart. Off until you enable it, deliberately: an install that has never configured an indexer must not start grabbing on its own. A monitored series page says so when automation is globally off, since monitored now means "will be grabbed automatically," not just "appears on the calendar."
• Failure memory: a release that fails is not chosen again. The sweep's determinism made its own bug — a failed download reverted its item to wanted, the next pass re-derived the same ranking, and the same release won forever. Failed releases are now blocklisted per series with an escalating expiry (24 hours, then 7 days, then permanent): repeat failure of the same release is the only signal separating a dead release from a bad day, so first failures stay recoverable while a proven-dead release converges to permanent. Identity is info hash and normalized title, because Torznab feeds often omit the hash. Automation degrades to the next-best eligible release; blocked releases appear in the series History tab with the reason, an unblock, and bulk clear actions — and a manual grab is never refused and never leaves a block behind.
• An environmental-fault breaker guards the failure memory itself. A full disk or a client reaping torrents fails a different release every pass — the escalation ladder never triggers, and the candidate pool drains into the blocklist one entry at a time. Failures across distinct items library-wide now trip a breaker that suppresses recording (never grabbing) until the window passes; Settings says so in words, and clearing the library blocklist also closes the breaker so a fixed fault doesn't wait out its window. Refused adds are remembered too, when the client can attribute the failure to the release itself — an unreachable client blocklists nothing.
• Background jobs status in Settings. Each runner job's last run, duration, and last error (styled as one), refreshed live — "did automation actually run, and did it fail?" no longer requires reading server logs.
• Season packs are matched honestly, and automation declines them. Packs previously surfaced as unmatched with a reason; they now match their items and are declined by automation with an explicit ineligible reason instead — the importer can only defer a multi-episode payload, and unattended grabbing must not volunteer for a state it cannot finish. A manual grab of a pack still succeeds, landing in the deferred flow as before. Per-file batch import is tracked for a later release.

Fixed

• Search terms with unicode punctuation found nothing. AniList titles carry typography that never appears in release names (×, ☆, ・), so a HUNTER×HUNTER-class title searched verbatim died before matching ever ran — and unattended search would have inherited the silence. Queries are now sanitized (× → x, separators to spaces), zero results fall back to the English then native title variants, and matching transliterates instead of deleting, so the romaji variant matches real releases on its own.
• Dual-titled scene releases misparsed the alternate title as the release group. The trailing (Romaji Title, Multi-Audio, Multi-Subs) parenthetical became the group, so group ranking, blocking, and pinning silently never applied to a whole common release family. An implausible group is now distrusted and the real scene group recovered from the codec-dash pattern. The same family's no-episode-title variant also lost its source and codec to a misfiled tag run; both now survive.
• Saving any Settings section no longer blanks the Authentication card. The section writers returned a settings body without the auth block and the UI cached it whole; all writers now share one complete response.

Internal

• Search, decide, and grab moved into one core package (internal/core/ acquire), shared by the manual routes, the sweep, and the feed poll — one matcher, one eligibility path, three entry points. Recording a grab is now a single transaction per release, so a multi-episode grab can no longer be half-recorded by a mid-write failure.
• Per-series search cadence lives in the store (last searched, backoff, next due) with an epoch guard so a reset landing mid-sweep — a series growing, or being re-monitored — beats the sweep's stale write.
• The recent-feed capability is a type assertion, not a wider interface, per the repo's optional-capability rule: an indexer that lacks it is a supported configuration, and decorators forward it only when the inner source really has one.

0.3.0 — 2026-07-29

Quality profiles and airing awareness: releases are now scored against a profile instead of ranked by seeders, and Transpondarr now knows when episodes air — feeding a calendar, a seasonal discovery page, and a scheduled refresh that keeps a releasing series growing on its own.

Added

• Quality profiles drive release selection. Releases are scored against a profile — ranked release groups (dominant by construction), resolution order, and reward-only preferences over source, subtitle type, codec, and repack/version — instead of ranked by raw seeders. Hard excludes and an optional minimum-score floor let automation answer "nothing yet" rather than grab something unwanted.
• Profile management UI and per-series assignment — create, edit, reorder, and delete profiles under Settings → Quality profiles (deleting a profile in use prompts to migrate its series first), and assign a profile per series.
• The Releases tab shows each release's score and its breakdown — a per-axis tooltip explains the ranking, and a release outside the profile is shown with an amber score and the profile's reason, never hidden.
• Profiles inform manual grabs; they gate only automation. A manual grab always succeeds in one request — no confirm step; when the release falls outside the profile, the response carries ineligible_reason and the UI reports it after the fact.
• Per-series pinned release group — "for this show, this group's release is definitive." The pin is an absolute sort tier above profile scoring, so no stack of bonuses can outrank it — but it never bypasses eligibility: a pinned release that trips a hard exclude or the score floor stays refused.
• Air dates from AniList — wanted items now carry when they air, synced in the background off the job runner. Absence is normal, not an error: AniList's schedule coverage thins out before ~2015 and can skip episodes even on modern titles.
• Calendar view — upcoming episodes for monitored series (GET /api/v1/calendar), each carrying its acquisition status; monitored series with no schedule are listed as unscheduled instead of silently missing.
• Discovery page — a browse-and-add seasonal chart with format, airing-status, and genre filters, backed by a per-season cache; entries already in the library are marked as tracked.
• A releasing series now grows on its own. A scheduled metadata refresh adds newly-announced episodes as wanted items, and the airing sync creates the items its schedule names — covering long-runners whose total episode count AniList never publishes.
• GET /api/v1/system/jobs reports background job status — each job's interval, last run, how long it took, its last error, and when it runs next. "Did the refresh run, and did it fail?" was previously answerable only by reading server logs.

Fixed

• Dimension-form resolutions now count. A release named with 1920x1080 kept the literal dimension string as its resolution, so it scored zero on the resolution axis and slipped past a 1080p hard exclude; the parser now folds dimension forms to the height form profiles are written in.
• Blocking a group in the profile editor no longer yanks the row away from the cursor. Blocked groups still serialize last on save, but the list no longer re-sorts mid-edit.

Security

• Upgraded React Router to v8 to clear GHSA-qwww-vcr4-c8h2, a CSRF bypass in its unstable RSC APIs. The UI's declarative routing was never exploitable, but the only patched release line is 8.x.

Internal

• Background work runs on a named job runner (internal/core/jobs) instead of a bare go per feature. Jobs are registered by name with an interval, a panicking job is contained to the run that caused it (its own loop and every other job survive), and shutdown drains in-flight runs before the store closes. Deliberately not a cron library and not a persistent queue: intervals only, nothing durable across restarts.
• The importer now runs on the job runner instead of its own goroutine, so it reports a real last run, duration, and error like every other job, and the entrypoint is down to one background mechanism and one drain.
• Session cleanup no longer races the database close on shutdown. It was started as an unawaited goroutine, so a sweep could be mid-DELETE when the store closed; it now drains with everything else inside the shutdown budget. The shutdown warning also names which worker overran instead of just reporting that something did.
• Interval loops are now tested with testing/synctest, so scheduling behaviour is asserted exactly against a virtual clock rather than polled against wall-clock deadlines.
• make test now runs under the race detector. The job runner's status fields are written by each job's goroutine and read by the HTTP handler, so a missing lock would be invisible without it. The existing suite was already race-clean; the whole run costs about a second more.
• A fuzz target guards parser.Parse against panics. Release titles come from external indexers through an unmaintained parsing dependency, so "never panics on arbitrary input" is now checked in rather than assumed.
• The frontend suite is split into node and happy-dom vitest projects, so pure-logic suites skip the ~350ms-per-file DOM build.
• Local verification is scoped to what changed, with CI named as the full-suite enforcement point — and the CI gates that stance relies on hardened.

0.2.1 — 2026-07-23

Frontend foundations: the web UI is now held to the same test/lint/format bar as the Go backend, plus one History-tab fix.

Fixed

• The History tab no longer presents a failed grab as "Downloading". A grab that settled as failed (the download errored in the client, or vanished past the missing-from-client grace period) rendered with the in-progress icon and verb; it now gets a distinct destructive-toned Failed row.

Internal

• The frontend now has a test runner (Vitest), wired into make test — the web UI is no longer exempt from the repo's TDD process.
• Formatting and linting are enforced end to end: Prettier (checked in make web-lint), oxlint's react/exhaustive-deps and suspicious rule category, and a self-installing pre-commit hook (gofmt + prettier) so a quick manual edit can't reach CI unformatted.
• CI now fails when the committed api-types.ts drifts from the OpenAPI spec, keeping the typed frontend client honest.

0.2.0 — 2026-07-23

A reliability release: no download can appear stuck as "downloading" forever anymore, and imports now survive crashes, power loss, and restarts.

Fixed

• Every permanently-stuck "downloading" state is gone. The three ways a grab could wedge forever are each fixed or surfaced honestly:
  • A torrent removed from the download client out-of-band is now reconciled: after a 5-minute grace period the grab fails and the episode reverts to wanted (this closes the known limitation from 0.1.0).
  • A deferred batch/season-pack grab now shows a distinct deferred status instead of "downloading" forever. It stays honest — the bytes are on disk and seeding — and manually grabbing a single-episode release replaces the deferred grab cleanly.
  • An import that keeps failing (a qBittorrent path-mapping gap, library permissions, disk full) now surfaces as stuck, with the actual error shown on the grab, instead of retrying silently with the reason visible only in the logs.
• Folder-wrapped downloads now import. A single-episode torrent that delivers a directory payload is resolved to its one episode file at completion time; import_deferred is reserved for payloads that genuinely can't be disambiguated (real batches).
• Imports are crash-safe. Durability and shutdown ordering across the whole import path:
  • Copy-mode imports fsync the file before renaming it into the library — previously a power loss could land the rename before the data, leaving a truncated episode that was treated as already-imported forever.
  • Hardlink-mode imports fsync the linked inode (and directory entry) before the grab settles, closing the same window.
  • Truncated files left in the library by past crashes are detected and reclaimed for re-import instead of being invisible.
  • Graceful shutdown now waits for the importer: the store can no longer close mid-import, in-flight multi-GB copies abort promptly within the shutdown budget, and a completed Place always gets its have/status writes.
• API failures in the web UI render as error states with retry, not as misleading empty states ("No titles found" after a rate-limited AniList search). Abandoned searches are now aborted instead of left running, and an untrimmed search cache key no longer causes duplicate AniList queries.
• Expired sessions are swept on a daily ticker, not only at startup, so long-lived instances no longer accumulate expired session rows.

Security

• Rate-limited the change-password endpoint (POST /api/v1/auth/password). It verifies the current password but was not throttled, so repeated wrong guesses were unmetered. This matters most under TRANSPONDARR_AUTH_REQUIRED=local, where any loopback/private-network client is admitted without a credential: the endpoint was an unauthenticated password-guessing oracle for anyone on the LAN. In the default enabled mode a valid session was already required, so there it is re-authentication hygiene. Unmetered argon2id verification was also a cheap CPU/memory exhaustion lever. Login and change-password now share a single per-client bucket (5 attempts per 15 minutes) rather than getting one each, since both verify the same admin password.

0.1.0 — 2026-07-22

The initial release: the full anime acquisition loop, end to end.

Added

• Acquisition pipeline — add a series from AniList, search a Torznab/Prowlarr indexer, grab via qBittorrent, and hardlink the finished file into a Plex/Jellyfin-ready library (seeding-safe, correct Season 01 naming).
• Web UI (embedded React SPA) — series list, add-series (live AniList search), per-episode wanted/downloading/have status, release matching + grab, and history.
• Runtime settings UI — edit qBittorrent / indexer / library integrations live, with changes applied without a restart.
• Forms authentication — first-run admin setup, session login for the web UI, and a separate machine API key (X-Api-Key) for dashboards and scripts.
• AniList metadata behind a status-aware read-through SQLite cache and a rate-limit-respecting client (min-interval limiter + 429/Retry-After backoff).
• Content-agnostic core keyed on WantedItem, with pluggable Indexer / DownloadClient / LibraryTarget interfaces.
• Typed HTTP API under /api/v1 with an OpenAPI 3.1 spec (transpondarrd openapi prints it) and a public, unauthenticated health endpoint (GET /api/v1/health).
• Docker runtime — multi-arch distroless image that fixes /config ownership automatically: it starts as root, chowns the config volume, and drops to PUID/PGID (default 1000:1000) before serving, so there's no pre-creating the bind-mount dir with sudo chown (pass --user to skip the root phase; an unwritable data dir fails fast with an actionable error). Ships with a built-in container HEALTHCHECK (the binary probes its own health endpoint — no shell or curl needed) and an example docker-compose.yml.
• Packaging — single static binary (CGO_ENABLED=0), GoReleaser + GitHub Actions CI/release, with SPDX SBOMs and build-provenance attestations for the release archives and the Docker image.

Known limitations

• Indexing is via Torznab/Prowlarr only currently.
• Season-pack / batch releases are recognised but not yet importable (they are surfaced unmatched with a clear reason rather than grabbed).
• No remote path-mapping: qBittorrent and Transpondarr must resolve the same paths (satisfied by mounting the shared volume identically).
• A torrent removed from the client out-of-band is not yet reconciled (a torrent that errors in the client is marked failed and the item becomes grabbable again).