Most brain-computer interface work is assembled from parts: one tool to read the recording, a stack of scripts to filter and epoch it, a notebook to train the decoder, a spreadsheet for the results, and a paper describing all of it in prose. Every seam is a place where a parameter gets lost and a result stops being reproducible.
BCILattice is a desktop application that covers those stages in one project, and BCINexus is the platform a finished project is published to. Analysis runs locally — recordings do not leave your machine unless you deliberately publish them — and every step records the parameters it ran with, so the study that gets published is the study that was actually run.
What one study covers
A study in BCILattice is a single project holding the recordings, every processing stage, the trained models and the report. Stages are named and inspectable; nothing is a black box and nothing overwrites the original recording.
| Stage | What happens |
|---|---|
| Import | EEG, fNIRS, EMG and EOG files — single file or whole-folder multi-subject import — with label columns confirmed and quality warnings shown before anything is added. |
| Preprocessing | Bandpass, notch, resampling, referencing and ICA for EEG; optical density, Beer-Lambert conversion, motion correction and short-channel regression for fNIRS; rectification and envelope extraction for EMG. |
| Analysis | Eleven reviewable steps from ICA review through epoching to feature extraction, each reading a named stage and writing a named stage. |
| ML Suite | A node canvas of 119 blocks: classical models, deep architectures, cross-validation, domain adaptation, evaluation. |
| Results | Metrics, per-fold detail, group statistics, and a validation tab that states which comparisons the design actually supports. |
| Publish | The study, its datasets and its models go to BCINexus for review, citation and reuse — or stay entirely local. |
Local-first, and that is a design decision
Training runs against your own CPU or NVIDIA GPU. Recordings, intermediate stages and trained weights live in the project folder on your disk. There is no requirement to upload a dataset to make the software work, which is what makes it usable for recordings that are not permitted to leave the institution in the first place.
Publishing is a separate, explicit action. When you take it, you choose what goes: the study record alone, or the study with its dataset and model weights attached.
Runs offline
Import, preprocessing, analysis and training need no network connection. The account and the community layer do; the science does not.
Your hardware
NVIDIA CUDA 11.8–12.4 is fully supported for deep models. A GPU is optional — classical pipelines train on CPU in seconds.
Windows, macOS, Linux
Windows 10 21H2+ and Windows 11, macOS 12 through 15, and current Ubuntu, Debian, Fedora, RHEL and Arch releases.
What publishing adds
A published study on BCINexus is not a PDF. It carries the pipeline that produced it, so a reader sees the actual filter cutoffs, the actual split strategy and the actual fold scores rather than a methods paragraph describing them.
Studies can be submitted for community review, cited, and forked as the starting point for new work. Datasets and trained models are first-class objects with their own pages, so a model someone else trained can be downloaded and evaluated against your own recordings.
Who this is built for
Labs running human-subject studies
Multi-subject import, grouped splits that keep a subject out of both train and test, and per-group statistics on the results.
Graduate students
The free plan covers real work: two studies, the full analysis suite, and 118 of the 119 ML blocks.
Engineers building decoders
A block canvas for the pipeline, custom Python blocks where the canvas runs out, and export paths for what you build.
Hardware makers
File import, an LSL bridge, or a native SDK integration so recordings from your device land in a complete workflow.
Questions
Is BCILattice free?
There is a free plan covering two studies with the full analysis suite and 118 of the 119 ML blocks. Paid plans raise the study and storage limits and add team, clinical and enterprise features. The free plan is not a time-limited trial.
Do my recordings get uploaded?
No. Import, preprocessing, analysis and training all run on your machine, and the project stays in a folder you choose. Data reaches BCINexus only when you explicitly publish a study, dataset or model.
Which signals are supported?
EEG, fNIRS, EMG and EOG, and more than one modality can live in the same study. Steps read the sampling rate from the file rather than assuming an EEG-style rate, which matters for a roughly 10 Hz fNIRS recording.
Do I need to know Python?
No. The analysis suite and the ML canvas are both graphical. Python is available where you want it — custom blocks execute code you supply, and the results notebook runs against the study in-process — but a complete study can be run without writing any.
Can it fit alongside my existing MNE or MATLAB workflow?
Yes. BCILattice imports the formats those workflows produce — EDF, BDF, GDF, FIF, VHDR, SET, CNT, SNIRF and tabular files — so it can take over from any point in an existing pipeline rather than replacing it wholesale.
Read the reference
Related
BCI Analysis Software
Eleven steps sit between a conditioned recording and a labelled design matrix. In BCILattice each one shows you its result before you commit it, and records what it ran with afterwards.
Reproducible BCI Research
A methods section is a summary written from memory weeks after the run. A study that carries its own stages, parameters and fold scores is the run itself.
Neuroscience Research Workflow
The study is the unit of work: recordings, stages, models, results and report in one project that a colleague can open and understand without you narrating it.
EEG Analysis Software
ICA review with components you actually label, channel scoring you confirm, epoching with rejection you can see, and a time-frequency view that computes once and derives seven views from it.