#FAQ
#Where should I start as a new user?
Start with Getting Started, then run the sample workflow in First Translation. Use a short file first so you can check quality, formatting, and performance before translating a large project.
#What is CatScribe best for?
Translation projects that need review and consistency: books, long documents, subtitle files, and repeated terminology. It is designed for workflows where you translate, edit, refine, and export — not for translating a single sentence once (though Quick Translate covers that too).
#Is CatScribe free? What does "Beta" mean?
CatScribe is currently in beta — the app labels itself "Beta (v1) · in development". During the beta, features the interface marks as Pro (for example attaching multiple glossary groups to one job, chunk score recalculation, and video export with subtitles) are shown as "Unlocked free during Beta". A cloud tier ("Catscribe Cloud") appears in the app as a "Coming soon" / "Pro" teaser and is not yet available.
#Which platforms are supported?
Windows (direct-download installer or Microsoft Store) and Linux (.deb package). macOS is not available yet — it is planned for later. See Installation.
#Can I run CatScribe without internet?
Yes, once your engine's language packs or models are installed. Local engines (Argos, LibreTranslate, MarianMT, NLLB) translate entirely on your device. What may need internet once: Argos language packs (50–100 MB per pair — some releases pre-seed English→Portuguese), MarianMT/NLLB model downloads from Hugging Face (200–600 MB), and optional add-ons like COMET (~1–2 GB). After that, translation runs fully offline. See Offline Translation.
#What file formats can I translate?
- Documents (File Translation): PDF, DOCX, EPUB — up to 50 MB per file.
- Subtitles (Subtitle Translation): SRT and VTT, UTF-8 encoded — up to 5 MB. ASS files are not supported.
- Video (Subtitle Translation): MP4, MKV, AVI, MOV, WebM — up to 500 MB, used for subtitle extraction, preview, and export.
- Quick Translate: pasted text up to 8,192 characters.
These are hard limits enforced at upload, not recommendations.
#What output formats can I export?
For documents: "Same as input", "Plain Text (.txt)", "Word Document (.doc/.docx)", or "EPUB (.epub)". There is no separate PDF output option — you get a translated PDF by translating a PDF with "Same as input". Subtitles export back to their own format (SRT or VTT).
#Which languages are supported?
It depends on the engine. Argos supports 30+ languages via downloadable packs (each ~50–100 MB). NLLB supports 200+ languages through one multilingual model. MarianMT has dedicated models for a few pairs (English↔Spanish, English↔French) and falls back to NLLB for everything else. Google (Free) covers its usual wide range online.
#Which engine should I choose?
Argos is the recommended default — the in-app guide calls it "Fast, offline translation. Install once, then translate without Docker or internet." NLLB is the quality-first local option for complex or literary text, at the cost of more RAM and slower speed. Google (Free) is convenient online but "May be rate-limited for heavy usage". If you don't want to decide, turn on Auto Mode and pick a quality level. Full comparison in Translation Engines.
#Why don't I see Argos in the provider list?
You are in Simple mode, the default interface mode. Simple mode groups the local engines (Argos, LibreTranslate, MarianMT, NLLB) into a single "Offline Translation" option. Switch the top-bar toggle to Advanced mode to see every engine individually. See Simple vs Advanced Mode.
#What is the difference between Offline Translation and Argos?
"Offline Translation" is the Simple-mode option that stands in for all local engines; Argos is one specific engine (the default). When you pick Offline Translation, the app runs whichever engine is set as your Preferred Offline Provider — Argos unless you change it. The selector shows "Currently using" with the resolved engine's name.
#How do I change the offline provider?
Open Settings → Providers → "Preferred Offline Provider" and choose Argos, LibreTranslate, MarianMT, or NLLB. The help text explains it exactly: "Engine that runs whenever "Offline Translation" is selected in Simple mode." Jobs that already exist keep the engine they started with.
#What are Ollama and LM Studio? Do I need them?
They are separate programs you install yourself, not part of CatScribe. Both run large language models locally; CatScribe detects them and uses their models for the AI refinement layers, "Improve with AI" in the CAT editor, and the optional AI Help Chat. They are optional — plain machine translation works without them. CatScribe offers an "Install Ollama Automatically" button on the Providers screen (restart your computer afterwards so it is detected); LM Studio is a manual install, and you must start its Local Server — opening the LM Studio app alone is not enough.
#Do I need Docker?
Only for one engine: LibreTranslate runs as a local Docker container (CatScribe checks http://localhost:5001 and provides the exact docker run command in Settings → Integrations). Argos, MarianMT, and NLLB need no Docker at all.
#How does CatScribe score translation quality?
Two mechanisms, shown in the CAT editor and the subtitle quality report:
- COMET — an optional neural quality-estimation model ("Enable COMET quality scoring" in Settings, ~1–2 GB download, off by default). Scored 0–100; bands are ≥85 Excellent, ≥70 High, ≥50 Medium, ≥30 Low.
- BLEU — computed only when you provide a human reference translation in the same target language ("N/A (reference required)" otherwise).
Chunk score tooltips also show glossary accuracy, combined as "Score = (2×Quality + Glossary) / 3". Very short segments are flagged ("⚠ Short segment") because their estimates are less reliable.
#Can I fix one bad sentence without retranslating everything?
Yes — that is the core of the workflow. Edit the chunk in the CAT Editor, or use the "Chunk Override" panel's "Re-run chunk" to retranslate just that chunk with a different engine (Chunk Overrides). For recurring terms, add a glossary entry instead.
#Can CatScribe handle video, or only subtitle files?
More than just files. In Subtitle Translation you can upload a video (MP4, MKV, AVI, MOV, WebM): CatScribe extracts embedded subtitle tracks ("Use This Subtitle"), and if none exist it can optionally detect burned-in subtitles with AI ("CatScribe can attempt to detect subtitles burned into the video using AI." — a ~250 MB optional install). You get an in-app player with a subtitle overlay, per-cue editing with CPS readability checks, and export back to SRT/VTT. "Download video with subtitles" produces a video with an embedded subtitle track (a selectable soft track, not pixels burned into the image). See Translating Subtitles.
#Where is my data stored?
Everything is local: jobs, chunks, and glossaries live in a local database inside the app's data folder; translated files go to your configured "Output Directory"; downloaded models are cached locally. Settings offers automatic backups (default folder Documents/CatScribe/Backups, every 24 hours, keeping up to 20) plus "Backup now" and restore.
#How do updates work?
Direct-installer builds: Settings shows "Current version" and a "Check for updates" button. Microsoft Store builds update through the Store — the app states "Updates are provided through Microsoft Store."
#Why is the app still running after I close the window?
By default, clicking X hides CatScribe to the system tray so background jobs and engines keep running ("Close to tray instead of exiting when I click X", on by default). Use Tray → Quit Application to fully exit, or change the behavior in Settings.
#Is there a comics/manga (visual) translation mode?
Not in current builds. A visual translation workspace for comics, manga, and image-rich PDFs exists as an in-development beta, but it is hidden by default — the app reports "Visual Translation is not available in this build." Watch release notes for when it ships.
#Why do PDFs sometimes behave differently than EPUB or DOCX?
PDFs store visual layout, not clean reading order — columns, headers, footnotes, and scanned pages all complicate extraction. File Translation has a "PDF Translation Mode" selector ("Auto (recommended)", "Simple text", "High Fidelity (preserves layout and tables)", "Fast (OCR for scanned PDFs)", "Via DOCX (structure + images)") to handle these cases. If you have the same book as EPUB or DOCX, prefer it. See Working With EPUB, DOCX, and PDF.
#Why is my translation quality inconsistent?
Usually missing glossary constraints. Add names and fixed terms in Glossaries, attach the group to the job, review affected chunks, and re-run only the sections that need correction. See Maintaining Consistency.
#Where can I fix formatting problems, merged words, or failed chunks?
Start with Common Issues & Fixes for output problems and Troubleshooting for engine and setup problems, then Editing Without Breaking Formatting for editing safely.