White paper · Draft · October 2026

Open neural interfaces

Neural interfaces for everyone, everywhere: open, buildable and affordable enough to become ubiquitous, so that high-quality brain data from everyday life becomes abundant and free to access.

Kristian Droste, founder

Summary

Neural interfaces for everyone, everywhere. The mission is not open hardware for its own sake. It is open, accessible, buildable and affordable hardware that becomes ubiquitous, so that high-quality neural data from people’s everyday lives becomes abundant. That abundance speeds up neurotechnology and brain health applications that are useful and improve the lives of all humans.

The work starts with wearable, research-grade EEG, makes it smaller and concealable, then does the same for wearable OPM-MEG.

Apps, for fun and for research, run on those devices. All are free for anyone who donates their data, many are free for everyone, and some require a small subscription to keep data private. Voluntarily shared data is open and free to access. On it, the foundation builds, trains, distills and open-sources foundation neural models, and better models make better apps and devices, which bring more data.

Our data privacy stance

Brain data, like all physiological data, is personal, and it stays private by default.

Today, the known risks of sharing or publishing personal brain data are very low. Even so, we believe every person should decide for themselves whether, and when, to share it.

• Sharing is voluntary, and anyone can stop at any time.
• Shared data is de-identified and open to all.
• Private data is never shared or used to train models.

Data already published cannot be recalled, and the foundation keeps reassessing the risks as models improve.

1. Discovery is bottlenecked

Neuroscience studies the brain through a handful of expensive instruments, and most of what they record ends up in small, separate datasets.

Price is the barrier, not physics. A 2016 study found that open EEG hardware gave readings very close to a medical-grade amplifier, and named hardware price as a limit on adoption (Frey, 2016). Builders of everyday brain-computer interfaces report the same gap: poor access to research-grade neural data, and few easy, comfortable sensors (Knierim et al., 2021).

The data is small. One of the largest open clinical EEG collections held 16,986 sessions from 10,874 people when it was described in 2016 (Obeid and Picone, 2016). An open EEG foundation model presented at ICLR 2024 was pre-trained on about 2,500 hours of EEG from around 20 datasets (Jiang et al., 2024). That is less than half the waking hours of one person in one year, at 16 hours a day.

2. Abundance, and one metric

No one can add hours to a lab’s scanner, but anyone who wants a research-grade instrument can be given one.

For neuroscience, abundance means:

• Recording in everyday life, not only in a lab session.
• Long recordings from the same person, over months instead of minutes.
• Studies of thousands of people at once, with no scanner waitlist.
• Enough open data to train open foundation models.

One metric

The foundation’s metric is dollars per research-grade recording hour:

Every choice, from electrodes to enclosures to apps, is judged by whether it lowers that number without lowering data quality. Each device version informs the next, so the hardware improves continuously instead of in rare, large releases. The target is still being set; the aim is an order of magnitude below today’s open hardware.

3. The instrument roadmap

The foundation makes one instrument open and affordable at a time, and moves on only when it passes a gate.

Each step reuses the last: the same open firmware, apps and data formats carry forward.

The third step exists in the lab: OPM-MEG is worn like a helmet with no supercooling, but it still needs background magnetic fields nulled (Boto et al., 2018), and its cost (section 1) keeps it in a few labs.

4. The flywheel

Affordable devices bring people; their apps bring data; the data trains open models; the models make the next devices and apps better.

• Apps drive the turn. They are made for fun and for research. All are free for anyone who donates their data, and many are free for everyone. Some require a small subscription for anyone who keeps their data private. A subscription is never required to use a device and acquire your own brain data from it.
• Data that people donate is published openly and is free for anyone to access, with consent recorded for every recording.
• Models are built, trained, distilled and open-sourced on that data. Distilled models are small enough to run on a phone or on the device itself.
• Funding becomes sustainable through privacy subscriptions, device sales, and private development contracts with governments and enterprises.

5. Open by default

What the foundation makes for the public is published, from the circuit boards to the model weights.

There is precedent for open neural hardware: OpenBCI publishes its board design files and MIT-licensed software (OpenBCI on GitHub). Data that people keep private is never published or used for training.

6. The namespace, in time

Every domain in the neuroscientific namespace has a job, but all of them wait until the core flywheel turns.

The namespace grows from the flywheel, not ahead of it. Each domain opens once there are devices, people and data for it to serve.

7. How it is organized and funded

A small nonprofit team, one metric, one instrument at a time, starting with philanthropic launch funding.

8. Risks and guardrails

The model runs on trust, so the guardrails are part of the product.

References

Every figure comes from a page opened for this paper, as of October 2026.

Instruments, cost and data

• Pedersen et al. (2022). Wearable OPM-MEG: a changing landscape for epilepsy. Epilepsia.
• Boto et al. (2018). Moving magnetoencephalography towards real-world applications with a wearable system. Nature.
• OpenBCI. Store and GitHub.
• BioSemi. Price list.
• Frey, J. (2016). Comparison of an open-hardware electroencephalography amplifier with medical grade device in brain-computer interface applications.
• Knierim, Berger and Reali (2021). Open-source concealed EEG data collection for brain-computer interfaces.
• Obeid and Picone (2016). The Temple University Hospital EEG data corpus. Frontiers in Neuroscience.
• Jiang, Zhao and Lu (2024). Large brain model for learning generic representations with tremendous EEG data in BCI. ICLR.

Rules

• ArentFox Schiff. California and Colorado establish protections for neural data.
• California Attorney General. California Consumer Privacy Act.
• FDA (2026). General wellness: policy for low risk devices.

Execution strategy

The budget, the team and how it works together are published separately, as the execution strategy. Read the execution strategy →