Skip to content

Setup: Windows

Use this guide when you run the provider client natively on 64-bit Windows. For example, flockadoo-flock1 on a Windows PC. The setup wizard installs llama.cpp. Native vLLM is not supported.

You need an NVIDIA GPU with at least 24 GB of VRAM, and 32 GB of RAM plus a fixed 64 GB page file (or 64 GB of RAM).

  1. Get your provider API key — see Sign up and create a flock.
  2. In PowerShell:
irm https://raw.githubusercontent.com/flockadoo/flockadoo/main/scripts/windows-install.ps1 | iex

The script installs flockadoo to %LOCALAPPDATA%\flockadoo\bin, runs flockadoo setup, and adds that folder to your user PATH. It does not require Administrator. Re-run the wizard anytime:

flockadoo setup

The wizard downloads llama.cpp (CUDA) into %LOCALAPPDATA%\flockadoo\llama.cpp and serves on http://127.0.0.1:8081. First-run download can take several minutes.

Info

Windows setup only lists catalog models that have a llama.cpp build. Fitting VRAM is not enough — encoder and embedding models such as baai/bge-m3 stay off Windows. See Approved model builds.

  1. Start the client from Start Menu → Flockadoo → Flockadoo Client, or from a new terminal:
flockadoo

On Windows the client runs in the notification area by default. The terminal waits until the local dashboard is up, then the prompt returns. Closing that window does not stop the client.

Right-click the Flockadoo icon and choose Open dashboard to see live status in your browser — session, models, GPU, and recent requests. Open config and Open logs open those files. Choose Quit on that menu when you want to stop the client.

To use the terminal dashboard instead:

flockadoo --console

Default config: %USERPROFILE%\.config\flockadoo\config.yaml. Logs: %USERPROFILE%\.local\state\flockadoo\log.

If startup fails with Invalid provider API key, copy the key from My Flocks → Settings into api_key and restart. See FAQ.

If SmartScreen says Windows protected your PC, choose More info → Run anyway. The GitHub binary is not code-signed.

GPU monitoring

Install the NVIDIA driver so nvidia-smi is on PATH. You need 24 GB of VRAM, and 32 GB of RAM plus a fixed 64 GB page file (or 64 GB of RAM). Then the health report includes GPU name, VRAM, and utilization. Utilization above 20% marks the GPU busy in that report. GET /api/busy and a 503 follow inference slots. See How it works. Without nvidia-smi the client still runs; GPU fields are empty, the health report does not mark the GPU busy from utilization, and the wizard will not offer llama.cpp.

flockadoo compatibility uses the same nvidia-smi data, or system RAM if no NVIDIA GPU is present.

Manual install

  1. Download flockadoo-windows-amd64.zip from GitHub Releases.
  2. Place flockadoo.exe on PATH.
  3. Run flockadoo setup.
  4. Run flockadoo.

The client opens an outbound connection to Flockadoo — no ngrok or open firewall ports required.