#Auto Mode
Auto Mode is a per-job execution preset in File Translation and Subtitle Translation. Turn it ON, pick a quality level, and CatScribe selects the provider, models, chunk size, and AI refinement layers for that job — using only engines and models that are actually installed on your machine.
Auto Mode is distinct from the global Simple vs Advanced Mode interface toggle: Auto Mode configures one job; Simple/Advanced changes what the interface shows everywhere.
#The Control
- The "Auto Mode" segmented toggle shows OFF / ON and defaults to OFF on both screens.
- An info button opens the "About Auto Mode" explainer: "Choose OFF for manual setup, or choose ON and then pick the quality level that matches your speed and quality goals."
- With Auto Mode ON, File Translation collapses to four fields: Quality, Source Language, Target Language, and Output Format. The provider selector, glossary block, chunk size control, and AI refinement section are hidden — CatScribe decides those.
- Subtitle Translation with Auto Mode ON shows only Quality, Source, and Target.
- A "Recommended setup" line suggests the quality level that best matches your detected hardware.
#Quality Levels
File Translation offers Fast / Balanced / Maximum; Subtitle Translation labels the top level Max Quality. The in-app descriptions:
| Level | In-app description |
|---|---|
| OFF - Manual mode | "You control provider, model, and advanced settings directly. Best when you want full manual control." |
| ON + Fast | "Prioritizes speed and lower resource usage. Good for quick drafts or large files when turnaround matters most." |
| ON + Balanced | "Balances quality and performance. Recommended for most translations." |
| ON + Max Quality | "Uses higher-quality processing with more resource usage. Best for final output when quality is the top priority." |
#What Each Level Actually Configures
Auto Mode "selects providers, models, chunk size, and LLM layers based on quality level and available resources," and it only ever uses installed providers and models.
| Decision | Fast | Balanced | Maximum |
|---|---|---|---|
| Engine preference (first installed engine wins) | Argos, then MarianMT | Argos, MarianMT, NLLB | NLLB, MarianMT, Argos |
| Chunk size (characters) | 4000 | 3000 | 2000 |
| Subtitle context window (entries per chunk) | 5 | 7 | 10 |
| AI refinement layers | None | Validation + Rewrite | Validation + Rewrite + Technical |
| AI orchestration | — | Single shared model for all stages | Separate model per stage |
File Translation shows the resolved plan as a pipeline badge before you start:
Fast -> "Base translation only"
Balanced -> "Validation + Rewrite"
Maximum -> "Validation + Rewrite + Technical"
Notes on model selection:
- For the AI layers, Auto Mode picks models from size-bounded preference lists appropriate to the stage and your hardware (for example, mid-size instruction models for validation on Balanced, larger models for rewrite on Maximum), with a keyword-based fallback when a preferred model is not installed.
- The AI provider follows your preferred LLM provider first, then falls back in the order Ollama → LM Studio → Claude.
- On Fast, AI layers are hard-disabled — even leftover refinement flags from a previous manual setup are stripped.
#What Auto Mode Does Not Do
- It does not install anything. If no local engine is installed at all, File Translation opens the local-engine installation dialog and Subtitle Translation asks you to install a translation provider first.
- It does not skip review. Auto Mode output is still a draft — review it in the CAT Editor.
- It does not manage terminology. Attach glossary groups before enabling Auto Mode if you want term enforcement (Auto Mode hides the glossary block but does not create glossaries for you) — see Glossaries.
#Choosing A Level
| Situation | Recommended |
|---|---|
| First translation, unfamiliar file type | Balanced |
| Very large file, draft quality is fine | Fast |
| Final-quality pass on important content | Maximum (needs a local LLM — Ollama or LM Studio with a model loaded, or a Claude key) |
| Limited RAM / older hardware | Fast, or follow the "Recommended setup" hint |
| You want to tune chunk size or compare engines | Auto Mode OFF — see Translation Engines |
Maximum uses more RAM and takes noticeably longer because every chunk can pass through three AI stages. On CPU-only machines, expect Balanced to be the practical ceiling for long books — see Performance Guidelines.
#Recommended First Run
- Turn Auto Mode ON.
- Select source and target languages (and Output Format for files).
- Choose Balanced — or whatever the "Recommended setup" line suggests.
- Translate a short sample first (a chapter, not the whole book).
- Review the output in the CAT Editor; check names and terminology.
- Adjust the quality level, add glossary terms, then run the full document.
#When To Turn It Off
Turn Auto Mode OFF when you need direct control:
- Selecting one specific engine for a job (for example forcing NLLB for a rare language pair).
- Tuning "Maximum part size" (chunk size) yourself.
- Configuring the AI refinement stages and per-stage models manually, including the Orchestration choice (Single-model / Multi-model).
- Troubleshooting a specific output problem engine by engine.
Manual setup exposes all of these controls — with more of them visible in Advanced interface mode. See Simple vs Advanced Mode.