A living neural organoid resting on a microelectrode array
Quantum Organoid Intelligence

Intelligence you
grow, not train.

NeuroKai cultures living human neural tissue directly onto silicon, then gives it something to think about. Deploy code to real neurons on the NK-1 — or reach a thousand of them remotely through Vivarium. No lab required.

Request platform access Read the science
Substrate liveCultures online 2,148Spikes / sec 1.42MMedian viability 147 dWatts per node 0.31Closed loops running 96 Substrate liveCultures online 2,148Spikes / sec 1.42MMedian viability 147 dWatts per node 0.31Closed loops running 96
105×
Less energy per inference than a comparable silicon model
147 d
Median culture viability under continuous perfusion
2,148
Organoid nodes addressable across the Vivarium fleet
24 / 7
Remote electrophysiology with sub-10 ms loop latency
01 — The substrate

Four billion years of
engineering, already done.

A neuron is self-programming, energy-frugal, and learns from a handful of examples. Digital models spend enormous resources approximating one. We start with the real thing.

01 / CULTURE

Grown, not fabricated

Human iPSC-derived neural organoids are differentiated in-house and seated onto our microelectrode arrays. Each node is roughly half a millimetre across and carries eight bidirectional channels — enough to speak and listen at once.

02 / PERFUSION

A body for the tissue

A closed microfluidic loop delivers neuronal medium continuously, holding temperature, pH, and osmolarity inside a narrow band. Removing the human hand from the incubator is what took viability from hours to months.

03 / STIMULUS

Reward, not backprop

There are no weights to update. Learning is shaped the way it is in a brain: structured electrical stimulation paired with timed neuromodulator release, delivered as predictable signal for desired behaviour and noise for everything else.

Fluorescence microscopy of a dense cultured neural network
Fig. 1 — Self-organised culture at 21 DIV · 40× confocal
02 — Orchestration

The wetware is the body.
Kai is what inhabits it.

Living tissue drifts. Cultures mature, excitability shifts, nodes age out. A fixed control program cannot hold a moving substrate — so we don't use one.

Quantum Kai, in the loop

NeuroKai runs on Quantum Kai, the Quantum Agentic Intelligence developed by ALIENTELLIGENCE. Kai holds persistent memory of every culture in the fleet — how a given node responded last week, which stimulation heuristics took, when a network began to fade — and orchestrates a fleet of specialised agents against that memory.

It plans the experiment, dispatches the agents, reads the spikes, and re-plans when the tissue changes its mind. No retraining cycle. No overnight queue.

  • 01
    Observe30 kHz continuous read across every active channel.
  • 02
    AttributeKai relates today's activity to months of prior state for that exact culture.
  • 03
    ActSpecialised agents issue stimulation, perfusion, and uncaging commands in parallel.
  • 04
    Re-planOutcome folds back into memory; the protocol rewrites itself mid-run.
# nk · attach --node vv-0417
kai ▸ binding orchestrator to substrate…
✓ node vv-0417 · 8 ch · 61 d in vitro
✓ memory recalled · 43 prior sessions
kai ▸ spawning 6 agents · goal locked
├ electrophysiologist · reading
├ plasticity-tuner · shaping
└ viability-warden · watching
✓ baseline 3.1 Hz · burst idx 0.22
kai ▸ excitability up 8% vs. session 42
kai ▸ re-planning stimulus amplitude…
✓ closed loop active · 6.4 ms latency
kai ▸ "It's learning faster than last week."
03 — Platform

Two ways to reach
a living network.

Bring the substrate into your lab, or leave the biology to us and connect over the wire. Identical API either way.

The NK-1 biocomputing device
Hardware

NK-1

$38,000 · ships Q1

A sealed, self-supporting biological computer for the bench. Life support, perfusion, recording, and compute all live inside the enclosure — plug in power and ethernet and you have a culture on your desk.

  • 32 addressable organoid nodes, 256 channels
  • On-device Kai runtime — runs fully air-gapped
  • Six-month consumable cycle, hot-swappable cartridges
  • USB and GPIO passthrough for cameras and actuators
Configure an NK-1
Interior of the Vivarium biocomputing facility
Hosted

Vivarium

from $900 / month

Our facility, your experiments. Reserve nodes across the fleet and run them around the clock from a notebook. We keep the tissue alive; you keep the science.

  • Reserved or burst access to 2,000+ live nodes
  • Python SDK, notebooks, and streaming spike API
  • Full raw electrophysiology retained and exportable
  • Free tier for accredited academic research
Request a Vivarium seat
04 — Specification

The numbers
behind the tissue.

Figures are fleet medians across the trailing ninety days. Live telemetry is available to seat holders.

Microelectrode array chip with four tissue wells
Fig. 2 — NK-MEA v4 · four wells, eight channels each
ParameterValue
Node typeHuman iPSC-derived cortical organoid, ~0.5 mm
Channels per node8 bidirectional, stimulate and record
Sampling rate30 kHz per channel, continuous
Closed-loop latency6.4 ms median, 11 ms p99
Median viability147 days · best observed 214
Power per node0.31 W including life support
PerfusionClosed microfluidic loop, 15 µL/min
NeuromodulationDopamine and serotonin, UV-uncaged, per-well
InterfacePython SDK · REST · streaming websocket
OrchestrationALIENTELLIGENCE Quantum Kai, on-device or hosted
Data retentionFull raw waveform, indefinite, exportable
05 — Applications

What a living
computer is for.

A / PHARMA

Human data, no animal

Assay a compound against functioning human neural tissue and watch what it does to learning itself — not just to cell viability. Ethically superior to animal models and more predictive of the human response.

B / NEUROLOGY

Disease, in motion

Model epilepsy, Alzheimer's, and Rett syndrome in patient-derived tissue, then perturb it continuously for months. Mechanisms that only appear over long horizons become observable.

C / MACHINE LEARNING

Learning at 0.31 watts

Benchmark reinforcement learning against a substrate that adapts in real time on a rounding error of the energy. For sample-efficiency research, biology is still the state of the art.

06 — Position

We take the hard
question seriously.

Our cultures are small, undifferentiated relative to a brain, and possess no sensory apparatus, no body, and no capacity for suffering that current neuroscience can identify. We believe that is true. We also believe it is exactly the kind of claim that should be verified by people who do not work here.

So NeuroKai operates under an external review board with authority to halt any protocol. All tissue derives from consenting adult donors under documented IRB approval. We publish culture provenance, age limits, and termination criteria alongside every dataset, and we support the development of binding sector-wide standards before, not after, they become urgent.

Our governance framework and donor consent policy are published in full and updated quarterly.

Now accepting the founding research cohort

Bring us a question
silicon can't answer.

We onboard a small number of labs each quarter and work alongside them closely. Tell us what you'd run, and we'll tell you honestly whether the substrate is ready for it.

Request access Download the spec sheet