- FIXED: translation with LM Studio failed for every model. The app sent a JSON-mode flag that LM Studio rejects (HTTP 400); it is gone now, so translation works with the models you already have loaded.
- FIXED: errors during translation said "see console" — but the app has no console. You now get a clear message with the server's actual response, and everything is written to logs\app.log (Settings > About > Open log folder).
- IMPROVED: instructions for LM Studio models are now sent in English in the system role, so small models no longer translate the instructions themselves.
- NEW: "LLM output token limit" setting (default 4096) so long screens are no longer cut off. Note: the model's context length is set in LM Studio itself.
Thank you to the player who reported this with full logs — that is exactly what made the fix possible.
