Rondine v0.1.1

• Add platform-specific engine installation guidance when no runnable engine is detected
• Make rondine setup automatically select and verify engines compatible with the host
• Refuse to start rondine serve when the selected engine is missing or incompatible
• Detect Rondine-managed llama.cpp, MLX-LM, and vLLM installations
• Improve setup error reporting and document the engine installation workflow

Rondine v0.1.0

• Initial Rondine control-plane CLI (Python 3.11+)
• Catalog presets: Qwen3.6, Gemma 4, DeepSeek-V4-Flash, GLM-5.2
• Engines: llama.cpp (GGUF), MLX-LM (Apple Silicon), vLLM (DGX Spark / NVFP4)
• Discrete NVIDIA GPU targets (cuda-8…cuda-80) with VRAM-based fit
• Commands: doctor, models, suggest, plan, setup, pull, serve, preset, stop, verify
• Coding profile + docs/coding.md + docs/engine-tuning.md
• Cluster inventory helpers (cluster init/doctor/plan/serve dry-run)
• Memory strategies: resident, supported CUDA hybrid (RAM+VRAM / --cpu-moe), and explicit experimental mmap SSD paging (--memory-mode, --allow-oversize)
• Disk free-space detection in doctor and a pre-download guard for sharded GGUFs
• Official curl / PowerShell installers with GitHub Release checksum verification
• Documentation site at rondine.dev (GitHub Pages)