flysim

runs a MaleCNS-constrained Drosophila brain and body simulator from the shell

TLDR

Run the lightweight engineering demo (no connectome download)

$ flysim run eon-demo --seed [1] --headless

Render a completed run to video

$ flysim render [runs/run-id]

Download the starter MaleCNS tables

$ flysim data sync --profile starter

Checksum the locked dataset

$ flysim data validate

Build the runtime aggregate graph

$ flysim data import-aggregate

Serve the browser arena without the full CNS

$ flysim web serve --mode preview

Estimate sparse-graph memory at several scales

$ flysim benchmark neural --scales [0.01] [0.1] [1.0]

SYNOPSIS

flysim command [subcommand] [options]

flysim is the command-line entrypoint of MaleCNS Virtual Fly (the Fly.exe project): a scientifically explicit simulator that can execute the traced universe of the released *Drosophila* male CNS connectome (165,122 neurons and about 25.6 million edges) inside a NeuroMechFly / MuJoCo body.The Python package is malecns-flysim; the console script is flysim. Dataset files are not shipped in the repository. flysim data sync pulls checksum-locked HHMI Janelia artifacts into $FLYSIM_DATA_ROOT, import-aggregate builds the sparse runtime graph, and evidence-grade commands refuse to start from a dirty git worktree.eon-demo is an engineering scaffold and neural-bypass control: it runs before FlyGym, CUDA, or MaleCNS data are installed. Full-connectome work needs an NVIDIA GPU, a native GeNN/PyGeNN build, and FlyGym. The production environment is Linux; Windows is supported through WSL2.

PARAMETERS

--seed N

RNG seed for a run (default 1 on eon-demo / eon-malecns).

--headless

Run without a display. On Linux this typically selects OSMesa for MuJoCo.

--render

Produce an MP4 as part of the run (checksum recorded in the run manifest).

--graph PATH

Path to an imported MaleCNS aggregate graph. Required for eon-malecns.

--root PATH

Data root. Defaults to $FLYSIM_DATA_ROOT when that variable is set.

--profile NAME

Dataset profile for data sync / status / validate: metadata, starter, or full.

--mode preview|full-cns

web serve mode. preview is labelled no-CNS; full-cns executes the real graph.

--allow-dirty-tree

Record a non-evidence-grade run from an uncommitted git worktree.

--fps N

Frames per second for render (default 30).

SUBCOMMANDS

data subcommand

Acquire, validate, and import the MaleCNS dataset. Common subcommands: sync, validate, status, import-aggregate.

run scenario

Execute a named simulation. eon-demo is the labelled engineering storyboard (no dataset). eon-malecns runs the traced connectome on GPU. full-vnc-walk stays gated until its motor decoder passes.

render run-directory

Rasterise a recorded run to video after the simulation has finished.

validate run-directory

Check a run directory against the project's validation contracts.

web serve

Host the interactive multi-fly browser arena.

benchmark kind

Memory estimates and measured CUDA/graph-load checks (neural, multi-fly, circuit, …).

evidence subcommand

Build or validate immutable evidence bundles (build, build-v0, validate).

showcase subcommand

Package cinematic / Eon-class demonstration artifacts.

stage2 subcommand

Fitted-dynamics readiness, holdouts, and exit gates for Stage 2 physiology work.

CAVEATS

Pre-alpha research software (GPL-2.0-or-later). Full-graph runs are slow relative to biological time (documented swarm demos run at a small fraction of real time on a consumer GPU). full-vnc-walk errors out until the motor-decoder interface gate passes. Rendering in WSL2 is software rasterisation; CUDA compute is a separate path. No command prints a validation tier as a hardcoded literal: eon-malecns resolves the project tier from evidence bundles at run time.

RESOURCES

Source code · Documentation