Skip to content
neuroflow0.2.31 · docs

← neuroflow

/neuroflow:brain-build

Assemble a computational brain model — neuron models, network topology, connectivity, and simulation framework setup.

/brain-build covers the full model construction pipeline, from choosing a neuron model and simulation framework to defining network topology and connectivity rules.

When to use it

  • You want to build a new spiking network, mean-field model, or whole-brain model from scratch
  • You are continuing development of an existing model
  • You need help choosing between simulation frameworks (NEURON, Brian2, NetPyNE, NEST, tvb-library)
  • You want to define neuron types, synapse models, and connectivity matrices

What it does

Claude determines whether this is a new model or an existing one, then guides you through the full construction pipeline.

If starting a new model

Claude asks:

  1. What brain region or circuit? (cortex, hippocampus, basal ganglia, full brain, custom)
  2. What level of abstraction? (single compartment, multi-compartment, point neuron, mean-field / population model, spiking network)
  3. What neuron model? (Hodgkin-Huxley, integrate-and-fire, AdEx, Izhikevich, custom)
  4. What simulation framework? (NEURON, Brian2, NetPyNE, NEST, tvb-library, custom)
  5. How many neurons / populations?
  6. What connectivity structure? (random, topographic, small-world, empirical connectome, custom)
  7. What inputs drive the model? (current injection, Poisson noise, sensory input, empirical LFP/EEG)
  8. What outputs are needed? (spike trains, membrane potentials, LFP, population firing rate)

If continuing an existing model

Claude reads .neuroflow/brain-build/flow.md and any existing model spec to understand where things left off, then asks what you want to work on next.

Steps

  1. Write model-spec.md — target circuit, abstraction level, neuron model, connectivity rules, input/output description, and definition of a successful model
  2. Plan the implementation: compartment definitions, synapse types, connectivity matrices, parameter ranges
  3. Build the model iteratively — write code, run a minimal test (single neuron fires, small network connects), refine
  4. Apply domain best practices for the chosen framework (NEURON .hoc/.py, Brian2 NeuronGroup/Synapses, NetPyNE netParams/simConfig, etc.)
  5. Pass the smoke test before scaling up — smoke_test.py runs a short simulation, checks its metrics in code (no NaN, not silent, no runaway rates, plus your own *_ok checks) and writes smoke-record.json with a hash of the model code

Model code is saved to your models/ folder (or the path set in .neuroflow/brain-build/flow.md). Specs and notes go to .neuroflow/brain-build/.

Smoke record

The smoke record goes stale as soon as the model code changes. /brain-run and /brain-optimize check it before a full run or sweep and ask for a fresh smoke test when it is stale, failing or missing.

Files read and written

Direction Files
Reads .neuroflow/project_config.md, .neuroflow/flow.md, .neuroflow/brain-build/flow.md
Writes .neuroflow/brain-build/, .neuroflow/brain-build/flow.md, .neuroflow/brain-build/smoke-record.json, .neuroflow/sessions/YYYY-MM-DD.md, models/ (code output)
  • /brain-optimize — run a parameter search or fit the model to data after building
  • /brain-run — launch a full simulation run of the assembled model