Common Questions about Sublarr

Sublarr is a self-hosted, open-source subtitle manager for anime and media libraries. It runs as a Docker container, integrates with Sonarr, Radarr, Jellyfin and Emby, queries 20+ subtitle providers in parallel, deduplicates via SHA-256 and prefers ASS-formatted subtitles. Optionally translates via a local LLM (Ollama) or cloud AI APIs — translation is experimental.

Honest answer first: if your library is English-language live-action and Bazarr covers it reliably, you probably don't need Sublarr — Bazarr is more mature, has more hands on it, and does that job well. Sublarr earns its place once anime is involved: absolute episode numbering, AniDB resolution, ASS with its styling intact, signs/songs tracks, and the several English tracks per file where picking the wrong one desyncs what you read from what you hear. Likewise if you maintain several languages with their own rules, or if no subtitle exists at all for what you want and you would rather have a machine translation you can inspect than nothing. Details under "Scope & limitations" in the docs. Longer comparison: <a href="/en/vs-bazarr/">Sublarr or Bazarr</a>.

LLM translation is experimental and off by default — quality varies sharply with backend, model and language pair. UI authentication is a single shared password: no user accounts, no roles, no audit trail; for genuine multi-user setups put a reverse proxy with forward-auth in front, whose headers Sublarr reads natively (Authelia, authentik). The scheduler assumes a single primary instance — additional replicas need SUBLARR_SCHEDULER_ROLE=disabled or every recurring job fires twice. There is deliberately no auto-update: the UI tells you a new release exists, you run the upgrade. Docker is the supported path; bare-metal is not tested. And several of the 28 providers need an account or API key — with no keys configured you are running a smaller subset than the number suggests. The project is maintained by one person.

Wire Sonarr to Sublarr via webhook (Settings → Connect → Webhook → URL: http://sublarr:5765/api/v1/webhook/sonarr). When Sonarr imports an episode it triggers a search in Sublarr, which queries the 28 providers in parallel, resolves AniDB for absolute episode numbers, and drops the best match into the library folder. For non-Sonarr libraries use the Wanted scanner (every 4 hours).

Anime releases often bundle several English tracks: fansubs (translating the original Japanese), the dubtitle (transcribing the English dub) and signs/songs-only tracks. If you watch the English dub you want the dubtitle — otherwise what you hear and what you read drift apart. Sublarr's dubtitle detection finds the right track by measuring, not guessing: Tier 1 classifies from metadata and cue density, and when several full-text tracks remain, Tier 2 matches them against the dub audio via Whisper. Sublarr suggests the track and shows the evidence — it never silently changes your files.

Self-hosted means: no monthly cost, no cloud dependency, full control over subtitle quality. Sublarr runs as a Docker container on your LAN, integrates natively with Jellyfin and Emby library scans, and delivers sidecar files next to the media — Plex and other servers pick them up automatically. Versus cloud services: no data sharing, no subscription model, customisable scoring pipeline.

Yes — Sublarr supports 12 translation backends: DeepL, Claude, Gemini, OpenAI, DeepSeek, Mistral, ChatGPT, Google, LibreTranslate, Azure, MyMemory and Ollama (local). Local models run, but are experimental — for production-grade output the cloud backends (DeepL, Claude, Gemini) currently deliver the most reliable results. JA→DE works best in two steps: first fetch the English subtitle (often fan-subbed), then translate EN→DE. Context-windowing preserves speaker consistency.

Yes — Sublarr uses Ollama as a local LLM runner. Recommended models: Qwen2.5 14B (solid all-rounder, ~9 GB) or Llama 3.1 8B (~5 GB). Both run without internet, without data sharing. Local translation is still experimental; if you care about output quality more than offline operation, cloud backends (DeepL, Claude) are currently the more solid choice. Setup: install Ollama on the host, in Sublarr go to Settings → Translation → Backend → Ollama URL (http://host.docker.internal:11434).

Sublarr has a multi-engine sync orchestrator with a fallback chain of ffsubsync (audio-feature based) and alass (alignment based). Enable Auto-Sync in settings and it runs after every download. For large drift (>2s) alass tries first, otherwise ffsubsync. Per-engine audit log shows score and correction. Manual sync with a custom offset works in the waveform editor.

In the Library click a subtitle file → "Edit". The audio waveform shows on top, with a sortable cue list below featuring ASS syntax highlighting. Drag cues to reposition; enable Snap-to-Wave for frame-accurate audio anchors. On save, BOM, line endings, encoding issues and overlapping cues are auto-repaired.

In Sonarr/Radarr under Settings → Connect → +Add → Webhook. URL: http://<sublarr-host>:5765/api/v1/webhook/sonarr (or /radarr). Triggers: On Import, On Upgrade. Sublarr supports multi-instance: add multiple Sonarr/Radarr URLs in Sublarr settings, assign tag-based language profiles (e.g. tag "anime" → DE+EN, tag "movies" → DE only).

Sublarr runs on any Docker-capable platform. Unraid: add ghcr.io/abrechen2/sublarr:latest from Community Apps, map port 5765, set /config and /media volumes. Synology DSM 7 (Container Manager): import docker-compose.yml (see Installation page). QNAP, TrueNAS Scale, Raspberry Pi (ARM64) and cloud providers work identically. RAM usage without translation: ~256 MB; with local LLM: 4–8 GB.

GPL-3.0 guarantees: (1) you can use Sublarr indefinitely, including commercially; (2) the source is and remains open — no hidden telemetry packages or backdoors; (3) anyone can fork if maintenance lapses; (4) modifications must be redistributed under GPL-3.0 if published — no closed-source takeover of your contribution. Practically: no vendor lock-in risk for your library.

By default only provider queries (subtitle search → external APIs like OpenSubtitles, Addic7ed, Subscene). No crash reports, no account creation. Anonymous usage statistics exist but are opt-in and OFF by default — nothing is sent unless you enable it; when enabled it is one ping every 24 hours made of enums, booleans and coarse buckets, and withdrawing consent erases the data already collected (see "Usage Statistics" in the docs). Cloud translation backends (Claude/Gemini/DeepL) send subtitle text to the respective provider — avoidable by choosing Ollama (local). Configuration and database stay in /config on your host.