## Implementation Complete ### Root Cause The transcription endpoint called `local_stt.transcribe(audio_content)` without passing `chunk_duration`, causing the entire audio file to be loaded into memory at once. ### Solution Added automatic chunking for large audio files: - Files >10MB are automatically processed in 10-minute chunks with 15-second overlap - Configurable via environment variables: - `LIBRECHAT_AUDIO_CHUNK_DURATION` (default: 600 seconds) - `LIBRECHAT_AUDIO_CHUNK_SIZE_THRESHOLD` (default: 10MB) ### Changes - `routes.py`: Added chunk detection logic and configuration constants - `test_transcriptions.py`: Added TestTranscriptionChunking class with tests for both small and large files ### Testing - All 36 STT tests pass - Chunking logic verified through mock tests - Leverages existing parakeet-mlx chunked processing support Note: Testing against the actual long audio files (gateway paths) was not performed as remote file access was not available.

Try this one for an even longer file: /home/gateway/YouTube/videos/JKrLyZngHIY/audio.transcription.mp3