Setup: macOS (Apple GPU)¶
Use this guide when you run the provider client natively on a Mac with Apple Silicon. For example, flockadoo-flock1 on a MacBook. You need 48 GB of unified memory to run approved vLLM Metal models (BF16 7B-class). Intel Macs are not supported.
Quick install (recommended)¶
- Get your provider API key — see Sign up and create a flock.
- Run the install script:
The setup wizard offers vLLM Metal or llama.cpp. It installs the one you pick if it is missing, offers to replace a vLLM Metal release older than 0.30.0 (see Supported engine versions), and shows which models fit this Mac. vLLM Metal serves on http://127.0.0.1:8000. llama.cpp installs into ~/.local/share/flockadoo/llama.cpp and serves on http://127.0.0.1:8081. Re-run anytime:
The script places flockadoo in /usr/local/bin.
- Start the client:
Default config: ~/.config/flockadoo/config.yaml. Logs: ~/.local/state/flockadoo/log.
Apple GPU monitoring¶
On macOS the client reports basic GPU information (model name and core count). It does not report GPU memory or utilization, so load from other processes does not set the health-report busy flag. A request in flight still does (default one at a time). See How it works.
On Linux with NVIDIA, the health report also marks the GPU busy when utilization is above 20%. See Setup: Linux.
Manual install from source¶
- Go: Install Go 1.22 or higher.
- Clone and build:
- Run
flockadoo setup, then:
The client opens an outbound connection to Flockadoo — no ngrok or open firewall ports required.