#Translation Engines

CatScribe supports several translation engines — most run entirely on your machine, some are online services. This page lists every engine in current builds, what each is good at, and how models are downloaded and managed.

#The Local MT Engines

#Argos (Local)

• The default Preferred Offline Provider and the app's "FAST & SIMPLE" recommendation: "Fast, offline translation. Install once, then translate without Docker or internet."
• Translates via per-language-pair packs: "Each pack is ~50-100 MB", with 30+ languages available. Packs install from the "Argos Language Packs" dialog with live progress, and are stored locally.
• The Argos runtime installs automatically on first launch (it can also be repaired from Settings → Optional Dependencies → "Argos Core Runtime").
• Internet is needed only to install the runtime and download packs — translation itself is fully offline.

#LibreTranslate (Local)

• A self-hosted translation server. CatScribe checks http://localhost:5001 by default.
• Requires Docker Desktop; Settings → Integrations provides a copyable docker run command and status checks. First container start can take 1–3 minutes.
• Choose it when you already run LibreTranslate for other tools, or want a server-based local engine.

#The Local LLM Providers: Ollama And LM Studio

Ollama and LM Studio are external programs the app integrates with — they are not bundled, and they do not appear in the main translation dropdown. CatScribe uses them for the AI refinement layers, Improve With AI, and the AI Help Chat.

• Ollama — detected at http://localhost:11434. CatScribe offers an in-app installer ("Install Ollama Automatically"; a restart may be needed for detection). Models are downloaded, cancelled, and uninstalled directly from CatScribe's Providers/AI Models screen; Ollama manages the model files itself.
• LM Studio — installed manually; auto-detected at http://localhost:1234/v1. Important nuance from the app: "Opening the LM Studio app does not start its Local Server — start it manually under the Developer tab." CatScribe reports three states: "Ready", "No model loaded", and "Local Server off". LM Studio models appear in CatScribe's selectors but are managed inside LM Studio.

#Model Management: The Providers / AI Models Screen

Open Providers in the sidebar (the screen is titled AI Models):

• "Your Hardware Profile" shows detected RAM, GPU, and "Models Disk Usage", and assigns a tier — "High-End Hardware" (GPU + 16 GB+ RAM), "Mid-Range Hardware" (12 GB+ RAM), or "Low-End Hardware".
• "Recommended Models for Your Hardware" lists Ollama models matched to your tier, from ~637 MB tiny models up to ~40 GB flagship models, each with a "Use for:" note. "Models marked as "Recommended" are optimized for your system configuration."
• Downloads show live progress with a Cancel option; note that interrupted downloads restart from zero (no resume).
• "Installed Models" lists what is on disk; uninstalling an Ollama model frees space ("…you'll need to download it again if needed").

#Choosing An Engine

Whatever you pick: test a short sample before scaling to a full book, and compare meaning accuracy, terminology, formatting, and speed. Engine speed and RAM costs are covered in Performance Guidelines; the offline story is in Offline Translation.