Brains learn from a few examples on about 20 watts. Labs have already shown that cultured neurons and brain organoids adapt to closed-loop tasks, from a simplified game of Pong to recognizing speakers.
The organoid programming platform
Compute that grows.
Design an experiment as code. Run it on living human neurons in our lab. Get back data you can reproduce. Organoid Programming is building the platform for organoid intelligence, so you can work with biological compute without building a wet lab.
Move your pointer across the culture to stimulate it. Click to make it fire.
Living neurons can learn. Working with them shouldn't take a lab.
Every experiment is still a one-off: a bespoke rig, custom code and months of setup. Results are hard to reproduce, and most researchers who could push the field forward never get access to living tissue.
The platform for organoid programming: protocols written as code, run remotely on quality-checked human brain organoids, with every result versioned, traceable and exported in open formats.
- Now
Access and rigor. Reproducible experiments on living neurons for neuroscience, drug discovery and AI research.
- Next
A new kind of computer. Learning systems that are sample-efficient and energy-efficient by design, built on biology.
Intelligence has an energy problem.
Every leap in AI has been paid for in electricity, and demand is still climbing.
Data centers used about 415 TWh of electricity in 2024, around 1.5% of the world's total, and AI is the fastest-growing reason that number is rising.
Meanwhile, the most capable learning system we know of runs on roughly the power of a dim light bulb. It learns from a handful of examples, keeps learning for a lifetime and repairs itself as it goes.
Neuromorphic chips try to make silicon behave more like neurons. We start from the other end: make living neurons programmable, and give researchers the tools to find out what they can do.
Nature already built the processor.
- Dendrites receive
- Soma integrates
- Synapses learn
- Axon transmits
86 bnneurons in an adult human brain
~20 Wto run all of them
Continuallearning: synapses strengthen and weaken with use, for life
Neurons don't separate memory from compute. Every synapse both stores a weight and does arithmetic, in parallel, at the same place. That's the property chip designers have chased for decades with neuromorphic hardware. Biology has it by default.
Grown, not fabricated.
Day zero: a few thousand stem cells. A few months later: a self-organized, firing network.
A brain organoid starts as a few thousand stem cells and organizes itself into living tissue.
Adult cells, donated with consent, are reprogrammed into induced pluripotent stem cells. Given the right signals, they fold into a 3D sphere that follows the same developmental script as the embryonic cortex: progenitor rosettes form, neurons migrate outwards, and layers appear.
Within weeks the neurons connect. Within months the tissue produces spontaneous, synchronised bursts of electrical activity, which is the raw material for computation.
-
Day 0
Stem cells
Induced pluripotent stem cells are aggregated into small clusters.
-
Week 1–2
Neural induction
Signaling cues steer the cells toward a neural fate.
-
Week 3–5
Rosettes
Progenitor zones form, like the developing cortex.
-
Week 6–12
Layers
Neurons migrate outwards and cortical-like layers emerge.
-
Months
Network activity
Spontaneous, synchronised firing appears across the tissue.
Learning to listen.
Beneath the tissue sits a grid of electrodes, sampling tens of thousands of times a second.
To hear what the tissue is doing, we place it on a microelectrode array: a glass chip with a grid of electrodes thinner than a hair.
Each electrode picks up the voltage spikes of nearby neurons thousands of times a second. Spike sorting turns that noisy signal into a precise record of which cells fired and when.
The same electrodes can also deliver tiny pulses of current. That makes the array a two-way channel, and that's what turns a recording into an interface.
Learning to speak.
Move your pointer across the culture. Every pass is a stimulus, and every click a volley.
Then we talk back. Any input, whether a sensor reading, a game state or a sound, can be written into the tissue as a pattern of stimulation.
The choice of electrodes, the timing and the frequency of the pulses carry the information. The network's activity on a separate set of electrodes is read out and decoded into an action.
Try it: move your pointer over the culture, then click. You're acting as a stimulation electrode.
The tissue is the model.
Feedback in, adaptation out. No weights file, only synapses.
Close the loop and something remarkable happens: the network adapts.
In 2022, a team showed that cultured neurons on an electrode array improved at a simplified game of Pong within minutes of play. Responses that hit the ball were followed by predictable stimulation, and misses by unpredictable noise. In 2023, another group used a brain organoid as a physical reservoir to recognize speakers from audio.
There was no backpropagation and no weights file. The learning lived in the synapses.
These results are early and narrow, but they are real. The science works; the tooling to repeat and build on it barely exists. That's the gap we're closing.
Sources: Kagan et al., Neuron (2022); Cai et al., Nature Electronics (2023).
From one-off rig to repeatable program.
Write a protocol once. Run it on living tissue in our lab. Get back data anyone can verify. Organoid Programming turns the hardest part of organoid intelligence, the infrastructure, into something you call from Python.
StepTodayWith Organoid Programming
Getting startedBuild a wet lab, a culture pipeline and an electrophysiology rigReserve a quality-checked culture through the API
Experiment designScripts tied to one rig in one labVersioned protocols in Python, portable across cultures
Closed loopCustom real-time code, rebuilt for every studyA managed runtime for stimulation, recording and feedback
ReproducibilityMethods sections and goodwillEvery run linked to culture lineage, protocol hash and raw data
DataProprietary formats on scattered drivesOpen NWB files with provenance and consent attached
Organoid cloud
Living compute on demand. Reserve quality-checked organoids on high-density arrays and watch viability, firing rate and bursting in real time.
Protocol SDK
Experiments as code. Define stimulation patterns, encoders and decoders in Python, review them like software and rerun them on any culture.
Closed-loop runtime
Train living networks. Millisecond loops between tissue and your feedback logic, for reinforcement-style training and reservoir computing.
Experiment ledger
Results you can defend. Every run is tied to culture lineage, media changes and protocol hash, ready for review, audit or replication.
Open data
No lock-in. Raw and spike-sorted recordings in Neurodata Without Borders format, ready for the analysis tools you already use.
Provenance built in
Responsible by default. Consent and provenance travel with every cell line, and usage policies are enforced at the API.
from organoid import Client, protocols
client = Client()
culture = client.cultures.reserve(region="cortex", min_age_days=60)
# Encode a game state as stimulation on "sensory" electrodes
encoder = protocols.place_code(culture.mea.region("sensory"))
decoder = protocols.rate_decoder(culture.mea.region("motor"))
with culture.session(hz=20) as s:
for state in env:
action = decoder(s.stimulate(encoder(state)).spikes)
s.feedback("predictable" if env.reward(action) else "noise")
Built for teams who want living neurons without the lab
Neuroscience labs
Run plasticity and network experiments without building an electrophysiology rig.
Drug discovery
Measure how compounds change human neural network function, not just cell survival.
AI research
Explore energy-efficient, sample-efficient learning in a fundamentally different substrate.
Disease modeling
Compare network behaviour in organoids from different donors to study neurological conditions.
Living tissue deserves a higher standard.
Working with human neural tissue raises real questions, and we'd rather face them early than after the fact. The field's own roadmap calls for ethics to be embedded from the start, and these are the principles we build by.
- 01
Consent and provenance
We work only with cell lines whose donors gave informed consent for this kind of research, and that provenance stays attached to every experiment.
- 02
Honest claims
Organoids are not brains, and responsive tissue is not a mind. We describe what our systems do in measurable terms and never overstate it.
- 03
Independent oversight
We seek review from bioethicists outside the company, and we publish our policies on how cultures are used and retired.
- 04
Open science
We share methods and data formats so others can check, reproduce and challenge our results.
Further reading: The Baltimore declaration toward the exploration of organoid intelligence, Frontiers in Science (2023).
Standing on serious science.
Organoid intelligence is young, and its foundations are public. These papers, from other research groups, shaped how we think.
-
In vitro neurons learn and exhibit sentience when embodied in a simulated game-world
Kagan et al. Cultured neurons improved at a Pong-like task under closed-loop feedback.
-
Organoid intelligence (OI): the new frontier in biocomputing and intelligence-in-a-dish
Smirnova et al. The field's roadmap for scaling, interfaces and embedded ethics.
-
Brain organoid reservoir computing for artificial intelligence
Cai et al. An organoid used as a physical reservoir for speech recognition.
-
Open and remotely accessible Neuroplatform for research in wetware computing
Jordan et al. Remote, around-the-clock access to organoids on electrode arrays.
Go deeper into biocomputing.
Long-form, fully sourced guides to how organoid intelligence works, and a glossary of the terms you'll meet along the way.
Start here
What is organoid intelligence? The complete guide
Organoids, electrode arrays, closed loops and the landmark experiments, in one place.
Read the guide TechnologyBiocomputing vs silicon: the energy question
Why a 20-watt brain keeps engineers up at night, and what the honest caveats are.
Read article ResearchHow neurons learn in a dish
Closed-loop training, DishBrain and the free energy principle, explained carefully.
Read articleThis needs people who speak two languages.
Biocomputing sits between the bench and the terminal. We want to hear from people who are fluent in one and curious about the other.
- Stem-cell & organoid biology
- Electrophysiology & MEA engineering
- Computational neuroscience
- Machine learning
- Real-time & distributed systems
- Developer tools & APIs
- Bioethics & governance
Send a short note about what you've built to careers@organoidprogramming.com.
Epilogue
Join the first cohort.
We're opening the platform to a small first group of neuroscience labs, drug-discovery teams and AI researchers. Tell us what you'd like to run on living neurons, and we'll follow up as places open.