Live simulation / JavaScript CPU / No CUDA required

A small brain.
A world you can change.

Place obstacles. Tune the connection between world and brain. Watch neural activity steer the player.

Loading toy circuit…

The default model is a hand-designed, 12-neuron demo.

0 ms · frame 0

Preparing the local simulation…

Keep an experiment. Continue it later.

Checkpoints preserve the brain, arena and current settings. Recordings include your obstacle edits, gain changes and silencing events; verification recomputes the run.

Everything runs locally after model loading. Imports stay on your device. Up to 100 seconds per recording; export and reset to start another. These JSON formats currently replay in JavaScript, not Python.

Build on this

This is a reference environment adapter built on the same FlyBrain JavaScript CPU runtime as the circuit lab. Its default fixed-step behavior is checked against Python Session and DodgeArena. Downloading a model is optional.

Generate an editable Python equivalent
flybrain init my-fly --template dodge
cd my-fly
python app.py --output runs/first

Integration, timing and replay guide → · Adapter source

What is modeled?

The toy circuit is artificial. The real option uses 313 cells from MaleCNS v1.0 with measured connectivity, assumed LIF dynamics and engineered input/output mappings. GF activity is assigned to steering for this demo; that is not its natural behavioral meaning. There is no trained policy, rescue controller or claim of biological fidelity.

Model provenance and assumptions · Recorded comparisons and sound · Contribute an adapter