av/harbor v0.5.5

v0.5.5 - Speech-to-Speech

on GitHub

Speech-to-Speech (s2s)

Local speech-to-speech pipeline (VAD, STT, LLM, TTS) wrapping Hugging Face's speech-to-speech project, wired to Harbor backends automatically.

harbor up s2s

Misc

• Workspace and config files now stay owned by your host user across 20+ services (no more root-owned files in bind mounts).
• Dify upgraded to the 1.x stack and DeerFlow migrated to upstream v2.
• ComfyUI moved to a maintained image with working ROCm support.
• Large repair sweep restoring dozens of services on current upstream images, including Open WebUI web search, OpenHands, Perplexica, Hermes, Nexa, Bolt, Airweave, Kotaemon, and Khoj.
• Cached GGUF models are now discovered correctly across llama.cpp, ik_llama.cpp, Ollama, llama-swap, LocalAI, and others.
• harbor url and harbor open now work for multi-port and host-networked services.
• New built-in landing page and Harbor CLI companion services for agent containers.
• hf CLI 1.x cache checks and xet downloads repaired.

Full Changelog: v0.5.4...v0.5.5