v1.3.0 — Local LLM & Offline AI

What's New

Local LLM Support

• Download and run Qwen 2.5-7B (or any GGUF model) entirely offline
• Zero-Dependency Download: Pure Python stdlib + shell/curl fallback — no pip install needed
• Smart GGUF selection: auto-picks best single-file quantization (q3_k_m > q2_k > ...)
• Built-in health monitoring, warmup progress, and crash recovery

Code-Enforced Group Chat Routing

• Group chat routing now deterministic (code decides who speaks to whom)
• LLM only generates content — identical conversation flow across cloud and local models
• User speaks → pets discuss (2-3 rounds) → wrap up to user

Improvements

• pet-llm-server.py: Shell/Python polyglot — works with /bin/sh even without Python
• pet-group-chat.py: Local LLM support with short prompt optimized for 7B models
• pet-auto-reply.py: Auto-detect local vs cloud provider
• Settings UI: New Local LLM tab with model download, server start/stop, health status

Technical

• Supports macOS 12+ (macOS 13+ ships with built-in Python 3)
• Bundled llama-server for Apple Silicon acceleration (Metal GPU)
• ~3.5GB default model download (Qwen 2.5-7B GGUF q3_k_m)