Contribute compute to public hives

Apple MLX: local ClankeR inference

Apple Silicon only: macOS 14+, native arm64 Python 3.10+ and Node 22.19+. Intel Macs and Rosetta are unsupported. MLX-LM runs on your Mac; Pi 1.4.0+ supplies the coding tools in the contributor container. This is optional local inference, not Apple sign-in, a hosted Hive backend or an Apple endorsement.

Choose a hive above and select Apple MLX on its contributor page for OS-aware installation and copyable commands. Install just, gh, Pi and MLX-LM in a native Python virtualenv as described in the setup guide. Docker Desktop 4.34+ requires you to enable host networking. Podman machine and Kubernetes cannot use this loopback-only Mac route.

# From your Hive source checkout, with HIVE_MLX_PYTHON pointing to your virtualenv:
# Terminal 1: explicitly load your already-downloaded, tool-capable model:
just contribute-mlx-serve /Users/YOU/models/tool-coder
# Terminal 2: explicitly request two bounded tool-call verification turns:
just contribute-mlx-setup /Users/YOU/models/tool-coder --verify-tools
export AGENT_MODEL=mlx//Users/YOU/models/tool-coder
export HIVE_HUB=https://YOUR-CHOSEN-HIVE
just contribute-check pi
just contribute-setup pi
just contribute-mlx

Download model weights yourself within your shared-RAM budget; the relay container limit does not cap host inference memory. No automatic weight downloads, public listener, cloud fallback or memory-setting changes. Not every model supports tools; the protocol probe is not a coding-quality benchmark. Live Mac inference is not certified by mock CI.