av/harbor v0.2.23

v0.2.23 - Speaches + future + fixes

on GitHub

Speaches

Calling your LLM is now easier than ever.

# Start the service
harbor up speaches

faster-whisper-server is now a more abstract project called speaches which is now supported by Harbor, unlike the previous iteration - this one supports both TTS and STT at the same time, so you only need one service to call your LLMs now.

Out of the box, it'll use Systran/faster-distil-whisper-large-v3 for STT and new cool hexgrad/Kokoro-82M for the TTS, both will be pre-configured for use with Open WebUI Audio settings.

Misc

• Harbor agents scaffold
• Harbor Boost:
  • New experimental recpl module to test recursive GUI planning
  • Boost now detects more tasks from new Open WebUI versions (to avoid running expensive workflows for them)
• kobold - more reasonable defaults
• General - adding generic DO_NOT_TRACK env vars for multiple services to disable tracking out of the box