Switch to OpenAI-compatible chat API with Qwen2.5-7B

Fix HF API path: use /hf-inference/models/ on new router

Fix HF API: switch to router.huggingface.co (old endpoint is 410)

Debug: surface HF API error in chat response

Add version number (v0.7) to status bar

Fix LLM: use Mistral v0.2 (warm) and full Willam persona prompt

Replace Kokoro with edge-tts: zero model downloads, instant startup

Fix model download: upgrade to Kokoro v1.0 with correct release URLs

Fix model files: Use voices.json instead of .bin

Fix VOICES_URL: Correct filename in GitHub release

Fix SyntaxError: use triple quotes for prompt

Build fix: Pin numpy and use Python 3.10

Robust build: Switch to full python image

Fix metadata colors

Deploy Willam-viv-peg