FLOCKINATOR
Menu

BUILD / WITHOUT HARDWARE

Start with
the experiment.

Reproduce the flagship communication benchmark on your computer. No panels, account credentials or model API calls are required.

01 / GET THE SOURCE

Use a public checkout.

Clone the public repository and enter its directory. Use uv to run the Python tools. 

git clone https://github.com/guruswami-ai/flockinator-public.git
cd flockinator-public

Record the commit you tested with git rev-parse HEAD. Results and commands should always be interpreted against that revision.

02 / REPRODUCE

Run the canonical check.

uv run python tools/reproduce/run.py --verify

This reruns seeds 0–15 and compares the resulting scientific content with tools/reproduce/expected.json. A successful check prints VERIFY OK and a SHA256 digest.

To save an inspectable result and summary:

uv run python tools/reproduce/run.py \
  --out /tmp/flockinator-repro

The output describes teams of four agents, with success defined as at least three choosing the correct sector. It is a host benchmark, not a test of sixteen physical devices.

03 / EXPLORE

Open the companion locally.

uv run python -m http.server 8811 \
  --bind 127.0.0.1 --directory cloudflare/site

Open localhost:8811 to explore the checked-in benchmark and replay. The static site does not provision nodes or connect to your hardware.

When you are ready to make it physical, begin with one panel and verify it before scaling up.

CONTINUE READINGBring one panel to life