Lingarr Next 1.2.1 [Closest Match]

Source matching uses the closest language code. Connections can skip
unmonitored titles when scanning.

• Source subtitle matching uses the closest language code (en-US can use en.srt)
  and ignores letter case. Forced and regular tracks can be translated separately
  when Ignore captions is off; a captioned target no longer blocks a matching
  regular source.
• Content translation (Bazarr and other API clients) keeps running if the HTTP
  client disconnects. An active duplicate returns HTTP 409. A completed request
  with the same source lines is reused.
• Machine-translation language catalogues load from the app directory instead of
  a hardcoded /app path.
• Dashboard changelog highlighting and the show-title directive write text nodes
  instead of HTML.
• Connections can skip unmonitored Radarr movies and Sonarr series or episodes
  when scanning. The switches start off, and turning them on leaves already
  imported titles in Lingarr. Migration M0046 seeds the three switches off.
• Batch structured-output schemas name the target language on the translated
  line.
• Expanding a season loads subtitles from that season's stored folder.
• Movie, show, and translation toolbars wrap from the large breakpoint so tablet
  widths stay on one row less often.
• A Radarr or Sonarr quality replace clears language coverage, the media hash,
  and the folder stamp so daily discovery downloads for the new file.
• A completed translation for an old filename does not count as coverage for
  the current file.
• Daily light discovery keeps the Radarr/Sonarr import cursor when the per-run
  translation cap leaves imported titles unprocessed.
• Download queues translation when English is already on disk (extracted or
  placed earlier) and providers miss the target language.
• Daily discovery no longer treats an empty target sidecar as a finished
  language pair. ProcessMedia drops unusable leftovers and records only
  usable coverage. Embedded extract ignores empty sidecars, deletes a
  0-byte extract, and waits up to 10 minutes so a large video can finish.

Lingarr Next 1.2.0 [Download First]

Lingarr can download a subtitle that already exists on a provider before it
translates.

Reliability update — 2026-10-20

• Subtitle downloads skip unavailable media folders and stop cleanly after a
  write or permission failure; they no longer recreate missing library folders.
  Bazarr retries wait for the folder to return within the existing retry window.
• Subtitle collection reports an unavailable folder with a storage/mapping hint.
  Bazarr HTTP failures are reported as search failures instead of empty results.
• HTML subtitle providers retain configured cookies and User-Agent through
  result and download pages, with cookies restricted to the provider's origin.
  HTML-encoded link query separators are decoded before following a link.
• JSON subtitle providers preserve endpoint query parameters (including Titlovi
  credentials) and resolve relative download links against the search endpoint.
• Provider HTTP failures and timeouts are warnings with provider/status details;
  diagnostic messages omit request URLs and credential-bearing exceptions.
• No new migrations or settings are required. Existing provider order and version
  metadata remain unchanged. Validation: 570 server tests and 19 API smoke checks
  passed; the Docker image was rebuilt and deployed.

Download First features

• New movies and episodes extract embedded source and target text tracks
  from the video before downloaders or translation.
• Settings → Connections lists each Plex library with a refresh control and
  a schedule switch. One cron runs enabled libraries in turn, waits 15
  minutes between libraries, then selects the configured subtitle when that
  sidecar is on the item.
• Movies and TV episodes have Download, Translate, and Retranslate controls.
  Download searches configured subtitle providers for a missing source and
  for target languages. If only a source file is placed, Lingarr queues
  translation for the rest. Translate and Retranslate queue from a source
  sidecar already on disk and stay inactive until that file exists.
• Settings → Translation → Providers downloads subtitles from Bazarr's
  provider list before translation. A matching target file is placed and the
  media server is notified. Lingarr translates only the languages still
  missing. The tab switch is on by default. Providers Bazarr marks as
  removed, broken, captcha-gated, AniDB-backed, private-tracker, Whisper,
  embedded-extract, or hash-only workaround stay out of the list.
• Settings → Translation → Providers can download a missing English subtitle
  for every movie and episode. The run does not translate. Each miss is a
  warning in System logs, and the page shows placed, no-match, and failed
  counts.
• A subtitle provider that sends an unknown character set such as iso-8859-15
  is treated as a miss, and Lingarr keeps walking the rest of the chain.
• An empty sidecar, a web page saved as a subtitle, or a file in the wrong
  language does not count as that language. Download searches providers for
  it. Daily translation ignores a web page saved as a source subtitle and
  searches providers for a real file. Translation fails when the source file
  has no subtitle lines.
• A subtitle in Arabic or another non-Latin script does not count as English.
  German, Spanish, Italian, or Portuguese dialogue labeled as English is
  skipped the same way French already is. UTF-16 subtitle files and HTML
  pages are decoded before those checks. Lingarr then tries the next file
  from that provider.
• Subtitle search uses the year from the file name. HTML providers follow a
  result page to a real subtitle file, including relative subs/ links, and
  keep spaces and brackets encoded on that follow. OpenSubtitles retries
  without a file hash when the hash search is empty.
• Subtitle search ignores punctuation in a title, so a colon, ampersand, or
  accent still matches provider pages that use dots or spaces.
• A short title such as Box or 9 must stand as its own words next to the
  year, so Black Box or District 9 do not steal that search. A short movie
  title still matches when the file puts The, A, or An in front of it.
  Episode subtitle search uses the series name and the SxxExx tag.
• A download link with no language tag is tried as the requested language.
  A file tagged as another language is skipped.
• A zip that contains several subtitle files uses the one whose name matches
  the requested language.
• JSON providers try later download URLs when the first file is a web page
  or the wrong language.
• An English sidecar whose release name drifted, such as WEBRip next to a
  WEBDL video, still counts as present for that title.
• A downloaded sidecar is named from the video file, not from a leftover
  .shifted subtitle sitting in the folder.
• Plex library refresh uses a 10-minute request timeout. A timed-out
  metadata or refresh call is a warning, and the job keeps going.
• Translate and Include sit in the center of those columns on movie, show,
  season, and episode tables.

Lingarr Next 1.1.6 [Infinitely Extendable]

A plugin can extend Lingarr without a fork. Drop a DLL in PLUGINS_PATH and restart. Every plugin starts off.

• Settings → Plugins shows each plugin as its own card. The switch is in the upper right. An off card is dim. Cards keep their own height.
• A plugin can add a tab and several cards, change a finished subtitle, run one external command, supply a source subtitle, accept a webhook, notify you when a translation ends, and run a scheduled task.
• Plex, Jellyfin, and Emby stay built in. A plugin server runs after them.
• The developer guide is in Lingarr.Docs/developers/Plugins.md. samples/StyleSample is a working set of plugins, all off until enabled.
• International quotes follows the target language. The Quote style dropdown can overwrite that country with another dialogue style.
• A plugin's failure policy says what Lingarr does: continue after an error, stop this file and keep earlier edits, or fail the translation and keep the old file.
• Strip HTML is on. Lingarr removes tags such as font, bold, and italics before translation, so the written subtitle stays plain text.
• Translation services and Provider Health list only translators. Other plugins stay on the Plugins page.
• When a provider cancels a job, Lingarr tries again. The default is 5 tries, 2 hours apart. 0 tries means no retry. A wait below 1 hour waits 1 hour. The status shows Cancelled (1/5) for that retry.
• Line retries stay inside the current job and wait seconds. Each later try waits longer by the growth factor. That wait is separate from the hours between full job retries.
• Settings → Translation → Setup has one Provider panel beside Languages. Drag a row to change the fallback order. Address, key, model, and account fields sit under the selected row. Microsoft Translate has no fields of its own.
• Open Translation Setup scrolls to that provider and rings the row.
• The dashboard chart hover shows the day, that day's count, and then the average. Both numbers are bold. The day is written as Sep 23.
• Translation detail pages show Revise with AI and Subtitle Quality side by side on wide screens.
• Settings, provider, and subtitle guidance was revised for clarity.
• Settings → Translation → Advanced can use TypeSafe Jev or OpenAI Luna Decisions for the same two subtitle checks. Luna uses the OpenAI translation key. Jev keeps its TypeSafe key. Existing installs stay on Jev, and both checks stay off until enabled.
• A subtitle file with no language code is treated as unknown.
• Creating a user requires a signed-in account once authentication is on. Onboarding cannot be completed again after it is finished.

Lingarr Next 1.1.5 [All The Subs]

Lingarr can turn a picture or caption track into a text subtitle, ask Bazarr for a missing source file, and queue a new movie or episode from Jellyfin or Emby.

• An English picture subtitle or closed caption inside a video can be turned into a text file before translation. Blu-ray PGS, DVD VobSub, DVB, and DivX XSUB are read with seconv and Tesseract. Captions and teletext are read with ccextractor. Both tools are included in the image.
• Settings → Translation → Subtitles turns each tool and each format on or off. They start on. Picture conversion is a last resort: Lingarr prefers a text subtitle, including one from Bazarr, and converts a picture track only after 72 hours with no text subtitle. The wait can be changed.
• A library scan finds files that have a picture or caption track and no text subtitle, then prepares them for translation. The scan is heavy on a large library and works through one file at a time.
• An OCR subtitle is named with an ocr tag, such as movie.en.ocr.srt.
• When a movie or episode has no source subtitle, Lingarr can ask Bazarr for the highest scored match in the source language and then translate it. The switch, address, API key, and minimum score are on Settings → Connections. The switch starts off. Forced subtitles are not used. By default Lingarr extracts an English text track from the video first and asks Bazarr only if that subtitle is still missing.
• A missed Bazarr search is tried again every 12 hours and stops after 168 hours. Both times can be changed.
• Lingarr keeps searching Bazarr for a subtitle it OCRed, and a downloaded subtitle replaces that OCR file. The switch starts on.
• Jellyfin Item Added and Emby library.new queue a new movie or episode the same way a new Plex item does. The Webhook card lists Radarr, Sonarr, Plex, Jellyfin, and Emby under one sentence.
• System logs open on the newest lines. Older lines load as you scroll up. A warning or error includes a short next step. Agents can read GET /api/logs and wait for the next error with GET /api/logs/wait.
• Moving between settings sections and tabs loads the page again.
• A connection test says "Bazarr connection succeeded." or "Plex connection succeeded."
• Migrations M0035 through M0040 run at startup.

Lingarr Next 1.1.4 [Resume and Classify]

A cancelled translation can continue, and TypeSafe Jev can keep cues, credits, and bad results out of the file.

• A cancelled translation keeps the lines it already finished. The next run for that same movie or episode continues when those lines score at least 90. The switch starts on. The threshold sits under it in Settings → Translation → Advanced.
• Under Cancelled, the list shows progress and quality, such as 55% · Qual: 98%.
• A completed movie or episode that still has its subtitle files gets a Quality line even when the request did not store line rows.
• Settings → Translation → Advanced has a Jev block for a TypeSafe API key. Two uses start off: skip sound cues and credits, and drop a result that is not a translation.
• The Languages panel on Translation → Setup is one wide row, with Source and Target side by side.

Lingarr Next 1.1.3 [Imports]

The public repository is https://github.com/apoapostolov/lingarr-next.

The daily translation check asks Plex, Radarr, and Sonarr which files are new, and it skips a title that already has its languages.

• Lingarr does not open a folder that already has the source language and every target language. It opens a folder only for a new import that is still missing one. A full drive walk runs only when Plex, Radarr, and Sonarr are all unconfigured and the library disk scan switch is on.
• Translation automation runs once a day at 02:00 UTC, before Sunday housekeeping at 03:00 and statistics at 05:00. An existing every-3-hours schedule is moved to that daily time.
• Under Quality, a completed translation from a pay-per-token API shows its input and output tokens. Subscription and OAuth providers do not.
• The README and the app header use a transparent mark. The navy tile fills the image.
• Migrations M0031 and M0032 run at startup.

Tag 1.1.3.

Lingarr Next 1.1.2 [Library]

Plex can start a translation when it adds a movie or an episode, and the scheduled walk of every folder stays off until you turn it on.

Library disk scan

• Scheduled translation, housekeeping, and statistics no longer open every movie and episode folder while Scan all folders on a schedule is paused. That switch is on Settings → Automation and is paused by default. Plex, Radarr, and Sonarr webhooks still translate a new item. A folder that was already checked is skipped until its folder time changes.

Plex

• A Plex webhook for a newly added movie or episode can start a translation. A movie is matched by its tmdb, imdb, or tvdb id. An episode is matched by the show, season, and episode number, or by the show id from Plex when the title differs. Lingarr translates only when a source subtitle file is already present and a target subtitle is not.
• Movies and episodes each have a switch on the Plex card. Both are on.
• Migrations M0028, M0029, and M0030 run at startup.

The image line remains lingarr-next. This patch is tag 1.1.2.

Lingarr Next 1.1.1 [Plex]

Sign in to Plex from Settings. After a translation, Lingarr can make that subtitle the one Plex plays.

Plex

• Settings → Connections → Media servers can sign in with the Plex PIN page, or with a local server address and token. The token stays encrypted.
• Sign out stays signed out when PLEX_TOKEN is set on the server. A sign-in from the card is the connection used after that, including across restarts.
• A finished movie or episode translation can tell Plex to select the new subtitle for one configured language. The item is matched by file path, by a tmdb, imdb, or tvdb id in the file name, or by the Plex title, original title, and slug.
• If a refresh does not list a sidecar such as name.bg.srt, Lingarr uploads that file onto the matched item and then selects it.

Upstream fixes

• USE_BATCH_TRANSLATION can be set from the environment.
• Saving an AI service no longer turns batch translation off for a service that supports batch.
• A translation job skips a request that is already completed or cancelled, and a queued duplicate is removed before resume.
• JOB_TIMEOUT_MINUTES sets how long a silent SQLite Hangfire job waits before retry. The default is 30.
• Content-API line translations strip subtitle markup when that setting is on.

Upgrade

Migrations M0026 and M0027 run at startup. Image line remains lingarr-next.

Lingarr Next 1.1.0 [Keep House]

Lingarr Next 1.1.0 [Keep House]

[1.1.0] - 2026-08-13 — Keep House

Weekly sidecar housekeeping, Mistral, Revise with AI, a Dashboard that opens
from cache, and the portable backend fixes from upstream 1.3.0.

Dashboard

• The Dashboard now opens instantly. The last successfully loaded statistics,
  activity, and translation history are shown immediately from a local cache,
  then refresh in the background. When fresh data arrives, only the changed
  numbers animate into place, so the page never blanks out while loading.
• Added subtle "Updating… / Updated Xm ago" indicators next to each Dashboard
  section so the refresh state is always visible without blocking the view.
• Reduced Dashboard load cost on the server. The activity summary is now
  memoized and served stale-while-revalidate, so repeat loads and tab switches
  no longer re-run the full set of translation queries every time.
• Added short-lived browser caching for the Dashboard and Statistics reads so
  repeat navigations skip the network round-trip entirely.
• Reordered the All-time Totals cards to Files processed, Lines
  translated, Characters translated.

Media library

• Fixed empty, unlabeled subtitle language pills appearing in the Movies list.
  Subtitles with no resolvable language no longer render an empty badge,
  matching the behaviour already used for TV episodes.

Library housekeeping

• Weekly job that renames orphan subtitle sidecars so they match the movie or
  episode file name (eng/bul included). Destination collisions are skipped.
• Optional weekly extract of one English text track from a video when no
  matching sidecar exists. Image-based subs are left alone. Capped per run.
• After either change, Jellyfin and Plex are asked to re-read that folder when
  a URL and token are configured. The job is on the Schedule page.

Translation providers

• Added Mistral AI as a first-class provider with encrypted API key storage,
  live model discovery, instruction profiles, and fallback-chain support.

AI revision

• Completed translations can be sent back through the first chat model in the
  chain. The job rereads source and target, rewrites only lines that change,
  and runs the existing quality checks again. The action lives on the
  translation detail page as Revise with AI. Scrapers such as Microsoft
  cannot revise a file on their own.

Upstream 1.3.0 ports

• Microsoft Translator now splits lines over 1000 characters and recombines the
  translated chunks, preserving current retry, timeout, and jitter behavior.
• API key generation now runs after onboarding completes and is authorized.
  Password updates use the same hasher as user creation.
• Settings JSON now accepts numbers written as strings, which unblocks threshold
  storage.
• Plugin provider keys are registered in lowercase so lookups stay consistent.
• Webhook URLs include the configured base path.
• The batch-translation toggle follows the chain's primary provider, including
  OpenRouter, Z.ai, OpenCode Go, Qwen, and xAI.
• Local AI batch translation retries when the model returns unparsable JSON.
• PostgreSQL timestamp conversion and FluentMigrator postgresql conditions
  match upstream 1.3.0. SQLite rollback of the include/exclude rename is
  idempotent.

Maintenance

• Took the safe 1.3.0 dependency train: ASP.NET/EF/Sqlite 10.0.10, Hangfire
  1.8.24, Test SDK 18.8.1, plus axios 1.19, Vue 3.5.40, Vite 8.1.5,
  and vue-tsc 3.3.8. Pinia 4, Node 26 types, Tailwind 4.3, and oxlint/oxfmt
  were left alone.

Release and compatibility notes

• Migrations M0024 (Mistral + revise) and M0025 (housekeeping settings)
  apply automatically at startup.
• New settings: subtitle_naming_enabled, subtitle_extract_enabled,
  subtitle_maintenance_schedule, subtitle_extract_max_per_run. Extract
  stays off in the database default; turn it on in settings or compose.
• The server image now ships ffmpeg so extract can run inside the container.
• Bedroom image line remains lingarr-bedroom:*. Do not point compose at
  official GHCR.

Lingarr Next 1.0.0

Lingarr Next 1.0.0

1.0.0 - 2026-07-29

This is the first consolidated release of Lingarr Next, the separate fork
maintained by Apostol Apostolov.

Dashboard and translation confidence

• Added a redesigned operational Dashboard centered on useful, recent
  information instead of lifetime-only totals.
• Restored Lingarr Next's readable historical translation graph inside Recent
  Activity.
• Added a configurable recent-activity window, normally 48 hours.
• Added a natural-language summary of completed subtitle files, quality results,
  provider contribution, active work, fallbacks, and provider availability.
• Added input-token, output-token, and approximate-cost reporting when metered
  OpenRouter, OpenAI, DeepSeek, or Anthropic translations occur in the selected
  period.
• Added a remembered primary-language selector to Media Overview. Movie and TV
  bars now describe current subtitle coverage in that language.
• Added Provider Health with explicit operational states, persistent structured
  outcomes, last-success information, expandable diagnostics, and safe provider
  tests.
• Added observe-only subtitle quality assessments from 0 to 100, including
  versioned line findings for suspicious output, preservation, language,
  formatting, readability, and consistency problems.
• Integrated quality results below Recent Activity and into completed
  translation status without blocking or deleting output.

Translation providers and fallback chains

• Added first-class OpenRouter support with live model discovery and
  provider-reported pricing where available.
• Made openrouter/free the default and first model, followed by
  openrouter/auto, then the remaining model catalogue.
• Added Z.ai support for the GLM Coding Plan endpoint and its supported GLM
  models.
• Added OpenCode Go support and improved DeepSeek model handling.
• Added a cached model catalogue with manual refresh.
• Replaced provider-only fallback configuration with ordered provider-and-model
  rows.
• Allowed the same provider to appear more than once with different models.
• Preserved compatibility with legacy plain-string and string-array
  SERVICE_TYPE settings.
• Added provider-specific request timeouts, including a longer Microsoft default
  for slow free-translation requests.

AI instruction profiles

• Added separate System Prompt and Context Prompt libraries.
• Added named drafts, immutable published versions, activation, rename,
  duplication, deletion, and version history.
• Added per-provider-row profile assignment with inheritance from active
  defaults.
• Added prompt-capability metadata to plugin manifests.
• Added a comprehensive example translation instruction set covering role,
  priorities, dialogue tone, profanity, forms of address, glossary, captions,
  numbers, readability, ambiguity, and final checks.
• Preserved existing prompt placeholders and added clearer guidance for
  dialogue-context tags.

Settings and interface

• Reorganized Settings into Connections, Translation, Automation, System, and
  Plugins while preserving Lingarr Next's existing cards and control vocabulary.
• Added local Translation tabs for Setup, Subtitles, Prompts, and Advanced.
• Placed Translation Services before Languages and clarified ordered fallback
  editing.
• Added shared tab and settings-shell components for consistent responsive
  navigation.
• Removed anonymous telemetry from Settings, onboarding, scheduled jobs,
  endpoints, persistence, and client services.
• Pointed update checks to the Apostol Apostolov fork.
• Improved toast visibility with the theme's Development-pill styling, a
  restrained spring animation, and protection against notification floods.
• Fixed stale dynamically imported pages after deployments by using safe update
  and asset-path behavior.

Translation list and detail views

• Added the episode name as a smaller second line beneath TV episode filenames.
• Replaced source and target pills with quieter rounded-square language markers.
• Moved completed quality into Status as compact Quality: XX% text.
• Centered Source, Target, Status, Progress, and Completed presentation where
  appropriate.
• Replaced strict completed dates with compact relative times such as 30 m and
  7 d.
• Refined card, badge, and progress colors to remain compatible with every
  Lingarr Next theme.
• Added optional subtitle paths to content-translation API requests and made
  content titles optional.
• Hardened cross-platform basename handling in the Translations list.

Reliability and operations

• Reworked subtitle discovery to avoid unbounded recursive scans, exclude
  trailer folders, and continue supporting conventional subtitle folders.
• Added paged automated-library processing with a durable cursor so large
  libraries make steady progress across runs and restarts.
• Added Hangfire SQLite startup recovery and periodic WAL checkpoint
  maintenance.
• Added transient retry with jitter for Microsoft/GTranslate requests.
• Added early rejection of invalid Sonarr and Radarr identifiers.
• Reduced noisy routine sync and already-translated messages to debug-level
  logging.
• Added safe provider operational retention and health reconciliation.
• Fixed fresh MySQL migrations for provider settings and imported prompt
  profiles by using database-correct quoting for the reserved settings key.
• Kept API keys encrypted and excluded subtitle text, prompts, paths, titles,
  and credentials from provider-health and usage records.

Documentation and testing

• Added fork-specific architecture, testing, reliability, settings, provider,
  fallback-chain, confidence, Dashboard, and prompt-profile documentation.
• Added a live API smoke suite for provider manifests, free translation paths,
  model ordering, content requests, update checks, and health endpoints.
• Added regression coverage for translation chains, fallbacks, model catalogues,
  provider health, quality scoring, prompt profiles, automation, subtitle
  enumeration, Dashboard activity, LLM pricing, and database migrations.
• Added operator guidance that keeps Bedroom deployments on fork-built images
  and treats upstream as a selective import source.

Release and compatibility notes

• Reset the fork's public release history to the independent 1.0.0 baseline.
• Moved the supported public image to
  ghcr.io/apoapostolov/lingarr:1.0.0.
• Database migrations through the LLM-usage and Dashboard-language additions are
  applied automatically at startup.
• Back up the application config and database before switching from an upstream
  image or attempting a downgrade.