Model catalog

Small install.
Explicit downloads.

The SDK includes versioned source URLs and metadata. It never downloads biological data during import or toy loading.

Inspect first, then download
from flybrain import list_models, model_info, fetch_model

print([(m["id"], m["status"]) for m in list_models()])
print(model_info("flywire-v783"))

# Fetch just the ~1.1 MB neuron-ID file.
paths = fetch_model("flywire-v783", assets=["neuron_ids"])
print(paths["neuron_ids"])

Available sources

IDStatusAssets
toyRunnable synthetic demo12 neurons, 12 edges. Bundled, MIT.
male-cns-escape-v1Runnable experimental subgraph313 neurons, 20,607 edges. 3.8 MB, CC BY 4.0.
male-cns-v1.0Raw dataConnections 1.05 GB; annotations 14.5 MB; transmitter predictions 43.3 MB.
flywire-v783Raw dataConnections 852 MB; neuron IDs 1.1 MB.

Official sources: MaleCNS downloads and FlyWire v783 archive. Their CC-BY-4.0 terms remain separate from the SDK's MIT license.

Raw wiring is not yet a runnable brain. Raw source files still need conversion and parameter/mapping choices. The escape model provides one reproducible selection with explicit assumptions; it is not a calibrated complete fly.

Cache and integrity

Specify assets to choose files and cache_dir to choose storage. The default is ~/.cache/flybrain-sdk/. Transfers stream to temporary files and verify size and available upstream checksums. Each downloaded asset gets a SHA256 cache receipt. MaleCNS SHA256 pins were computed from official-URL downloads by this project; they are not publisher-issued signatures.

Help add a runnable model

The first real slice is available with a reproducible recipe, source attribution, software tests and measured CPU runtime. Contributions can add another circuit, improve calibration or provide an application adapter. The catalog grows through reviewed pull requests.

Read the concrete integration plan →