# LM Studio Migration Notes

## Summary

This repo originally translated subtitle chunks through a Google Translate scraper wired directly into `src/engines.py`. The translation backend is now replaced with a dedicated LM Studio client that talks to an OpenAI-compatible `/v1/chat/completions` endpoint.

## New Runtime Defaults

- `LM_STUDIO_BASE_URL=http://127.0.0.1:1234/v1`
- `LM_STUDIO_API_KEY=lm-studio`
- `LM_STUDIO_MODEL=gemma-3-4b-it`
- `--translation-backend lmstudio`

## Commands Used In This Checkout

```powershell
uv venv --clear --python "C:\pinokio\bin\miniconda\python.exe" .venv
uv pip install --python .venv\Scripts\python.exe -r requirements.txt pytest
```

Validation commands:

```powershell
.venv\Scripts\python.exe -m pytest
.venv\Scripts\python.exe main.py --help
.venv\Scripts\python.exe -c "from src.translation import TranslationConfig, LMStudioTranslator; print(TranslationConfig.from_env().model)"
```

## Files Touched

- `main.py`
- `requirements.txt`
- `README.md`
- `src/engines.py`
- `src/translation.py`
- `tests/conftest.py`
- `tests/test_main_cli.py`
- `tests/test_translation.py`

## Notes

- Translation remains segment-by-segment for deterministic subtitle ordering.
- The CLI now supports `--lmstudio-base-url` and `--lmstudio-model`.
- Parser/help now loads before heavy runtime imports, which makes `main.py --help` more reliable.
- `src/googlev4.py` was removed from the active codebase because LM Studio is now the only supported translation backend.