The Workstation

The whole lab.
On your computer.

Research Associates get the complete source code: every experiment, the full connectome model and the instruments. Run it on Mac, Windows or Linux.

Get the source codeSetup guide

The workstation running the double-slit experiment, with neural activity panels

Runs on Mac, Windows and Linux.

Download the source code, install three prerequisites and run setup. It downloads the connectome data and opens the workstation in your browser.

Source code access is included with Research Associate on Patreon, $10+ a month. Become a Research Associate

Everything the recordings are made with.

Every experiment

All 68 experiments in the registry, each with its cameras, controls and timeline.

The connectome

166,700 neurons stepping live, with spike rasters, population readouts and the reconstructed anatomy viewer.

Your own records

Measured trials saved on your machine, and experiment packages you can export, share and import.

Experiment Builder’s Guide

Design your own experiments.

The workstation comes with detailed, step-by-step instructions for building your own experiments: setting up a scene, placing the fly, writing the timeline, choosing cameras, recording the neural response, and packaging the result to share with other Research Associates.

Included as a guide on this site and as a PDF.

Tech specs

Simulation
flybrain 0.1.0 · leaky integrate-and-fire network · 20 ms step · runs on the CPU
Connectome
MaleCNS v1.0 · 166,700 neurons · 25,582,938 directed connections · fixed weights
Body
NeuroMechFly v2 · micro-CT derived · 69 segments · three levels of detail
Anatomy viewer
626 reconstructed neurons · 3,896 soma positions
Experiments
68 in the registry, with new experiments added for Research Associates
Rendering
Scientific, Topology, Clay study and Fully rendered modes · switchable quality and level of detail
Records
Trials stored locally in SQLite · versioned experiment packages for export and import
Documentation
Experiment Builder’s Guide: step-by-step instructions for building and sharing your own experiments · web and PDF
Prerequisites
Node.js 22 or later · uv, which installs Python 3.12 · FFmpeg with ffprobe
Requirements (estimated)
macOS 13 or later, Apple silicon or Intel · Windows 10 or 11, 64-bit · 64-bit Linux (Ubuntu 22.04 or equivalent) · 4-core CPU minimum, 8 cores recommended · 16 GB RAM · a WebGL 2 browser; a discrete GPU is recommended for Fully rendered mode · about 5 GB free disk space
Delivery
Source code access for Research Associates on Patreon, $10+ a month · runs locally after a guided setup that downloads the connectome data

Hardware requirements are estimates. Setup guide

$10 a month.

Research Associates get the workstation’s source code, new experiments as they are made, and a vote on what the fly does next.

$10/month

Research Associate · Patreon

Become a Research Associate

Or do it yourself

Everything it runs on is public.

Every dataset and package the workstation uses is listed here. Turning them into a working lab is a substantial engineering project.

Data and models

Software

Python 3.12

Web

Tools

What assembling it involves

  1. Download the MaleCNS flat connectome and convert it into a sparse weight matrix for 166,700 neurons and 25,582,938 directed connections.
  2. Run it as a spiking network in 20 ms steps, and map each kind of stimulus onto the sensory neurons it should drive: visual, olfactory, gustatory, mechanosensory, thermal and hygrosensory.
  3. Render what the fly sees from its own head on every frame and encode it as retinal input, kept separate from anything the display shows.
  4. Fetch the skeletons of 626 reconstructed neurons, simplify them for WebGL, and map recorded spikes onto them.
  5. Rig the 69-segment NeuroMechFly body, build three levels of detail, and write the gait, posing and inverse kinematics.
  6. Build each experiment: geometry, supporting flies, cameras, timeline, input mappings and a versioned package format. The registry has 68.
  7. Store measured trials in SQLite, prepare media with FFmpeg, and keep the backend generic enough to load any experiment.
  8. Write the interface: instrument panels, spike rasters, the compound-eye inspector, and four switchable rendering modes.

Expect to need fluency in Python, TypeScript and WebGL, and working knowledge of spiking network models and connectome data. Or skip all of it: Research Associates get the finished source code for $10 a month. Become a Research Associate