DevByteAI

torch.load() calls in mtl_tts.py were missing map_location parameter, causing RuntimeError when loading CUDA-saved models on Mac M1/MPS devices. Fixes #85

fix: Support Python 3.12 by relaxing numpy version constraint

2

## Summary Relax numpy version constraint from `>=1.24.0,=1.24.0,

Add dicta-onnx dependency to fix Hebrew TTS support. Previously Hebrew text processing failed due to missing dependency.

fix(template-transform): use base64 encoding for Jinja2 templates to fix #26818

1

## Summary Fixes #26818 - Textarea pre-fill values in Template node were not displaying correctly when using Jinja2 variables. ## Problem When creating HTML forms in Template node with Jinja2...

## Summary Fixes #4388 This PR fixes the incorrect `transcription_delay` metric calculation when using STT-based turn detection (e.g., Deepgram Flux). --- ## Problem When using STT turn detection mode, the...

Fixes #4243 :Prevent immediate speech interruption in pause mode

1

## Summary Fixes #4243 This PR fixes the phantom VAD activity issue that caused unwanted interruptions when using STT turn detection with `resume_false_interruption=True`. --- ## Problem When using STT turn...

## Summary Fixes #34382 This PR fixes the streaming token leakage bug where internal model calls made by middleware (like `SummarizationMiddleware`, `LLMToolSelectorMiddleware`, and `ToolEmulatorMiddleware`) were contaminating the parent agent's output...

## Summary Fixes #34490 This PR fixes the `TypeError: Type is not msgpack serializable: Send` error that occurs when using `ShellToolMiddleware` with `InMemorySaver` checkpointer. --- ## Problem When combining `ShellToolMiddleware`...