A learning, local AI that recognizes individual patterns in signals: fully inspectable, no cloud, no black box.
Conway learns in real time from real sensory input, sound, image, motion, right on the device, no internet. And you can look fully inside it at any time.
Online learning without pre-training; multimodal sensory input (microphone/FFT, camera, gyroscope, clock) feeds the substrate directly; the full state is inspectable; runs offline.
Have a signal that needs monitoring? See what already works today, and start with a measurable pilot.
Request a pilot project →Researching learning systems, or seeking a partner for a funding proposal? We bring a running, open system, and name the open questions honestly.
Partner with us on research →Our tools aren't drawn up on a whiteboard for a quick sale. They exist because we need them ourselves: built to solve our own problems, and in daily use in our own operations. That's exactly why they're built properly rather than for fast money. What helps us usually helps others just as much. So we pass them on.
See it all →
Fully developed and currently in final testing. The name says it all: Bridge Buchhaltung builds the bridge between you and the tax office. Originally built for our own operations and in daily use here ever since: lean, clear, no forced cloud. The launch is close.
These tools are fully developed. We don't offer them off the shelf, we place them selectively. Whether as an exclusive solution for a single partner or in another form, we discuss that individually.
A tool to generate, shape and understand sound and signals. Here too the bridge is in the name: between signal and sound. On the left the classical substrate, on the right the quantum chamber: with the exchange running both ways in between. The entanglement is not a metaphor here: it can be demonstrated by a Bell test.
Ask us
A complete DJ deck in the browser that learns your mixing style and shows what it has learned. The picture shows the control surface: two decks, tone controls, crossfade and six effects.
Ask us
Our learning AI system: runs locally on the device and is fully inspectable at any time. This network of living cells is based on neuron technology. What is interesting is not the sound but the principle behind it: a system that learns in real time, with no training phase and no data set.
Ask usThe pictures deliberately show only a glimpse. For specific questions about a tool, just get in touch.
What actually works today: the system perceives through real senses, microphone (sound), camera (image), motion sensor and clock, and reacts and learns from them, in real time on the device. Over weeks it builds its own “personality” and shows its state openly at any time. Not yet done: reliably reading something useful from these impressions: e.g. flagging on real measurement data “this is normal” or “something is off.” That's the goal we test in a pilot with real numbers.
Verified in code: real-time online learning without a training phase; multimodal input via real sensors (microphone/AnalyserNode-FFT, camera→substrate, gyroscope, time-sense); two memory timescales (Hebb/decay + score-selection into the permanent pool); persistent, individuated entities (temperament, biography); radical inspectability; export/import with signature; offline. Open (class C): semantic interpretation of stimuli, deriving individual normality, anomaly detection, measurement against a reference (e.g. polysomnography), benchmarks, certification.
At the push of a button the system reveals its entire inner state: every memory, its vitality, its limits. No billion-parameter fog. What it can't (yet) do, we say ourselves.
Today research & development, not a certified medical device. Pattern recognition is proven on audio/rhythm signals; clinical metrics are built together in a pilot.
The same principle, learn, recognize, deviate, extends to further fields. These are designed, but not productive today. Honestly a roadmap, not a promise.
One entity per machine learns its sound-normality; bearing damage or smoothness-drift as deviation.
Air-gapped watch function on sensor/process rhythms; a pre-filter escalating only anomalies. No video analytics, no facial recognition.
Perception and companionship from local, learning patterns: e.g. fall/emergency detection without face tracking.
The system can already take images in as stimulus today: interpreting them usefully (imaging, object recognition) is the next step.
If you need video analytics or imaging today, we're (still) the wrong fit: we'd rather say so upfront.
What we exclude: no weapons systems (without exception), no surveillance analyzing individuals without consent, no manipulation for advertising, politics, or dependency.
Most systems called “AI” today are libraries at heart: they store knowledge and retrieve it. Conway does the opposite: it builds a space in which intelligence emerges, learning as it runs instead of looking things up. First the space, then the mind.
The principle is emergence: complex behavior arises from a few simple rules, like ants. One alone can do almost nothing; hundreds of thousands build states. Conway's Game of Life needs just four rules to create an entire world. Each neuron is small and simple like an ant; the intelligence lies in the interplay, not the single part.
Conway is not the next GPT. It's a different instrument: a handbuilt harpsichord beside the concert grand: smaller, but the better tool for the right music.
The substrate is Conway's Game of Life: four local rules from which universal computation (Turing completeness) provably emerges. Complex behavior arises from local rules plus Hebbian plasticity and leaky-integrate-and-fire dynamics within the substrate itself.
Related to reservoir computing / liquid state machines: except the medium is plastic (an adaptive rather than fixed reservoir) and the “selection” is an evolutionary promotion mechanism instead of a trained readout layer: selection instead of gradient.
Honesty classes: (A) evidenced, verifiable in the system or backed by literature · (B) reasoned, plausible, not yet externally validated · (C) open, hypothesis/plan. Conway is not yet peer-reviewed; “new” is a research position, not a seal of quality.
No known system (research July 2026) unites all six features:
Delimitation: NCA are gradient-trained (Conway learns at runtime without gradients). LSM reservoirs are fixed & random (Conway's medium is plastic). Neuromorphic chips (Loihi 2) implement the same principle in special silicon: the strongest external validation, not competition.
| System | Shared | Difference |
|---|---|---|
| Neural Cellular Automata | CA substrate, local rules | gradient-trained; no runtime plasticity, no individuals |
| Lenia | continuous CA lifeforms | no learning rule, no persistence |
| Liquid State Machines | LIF reservoir for temporal patterns | reservoir fixed & random; only readout learns |
| Loihi 2 / SpiNNaker | LIF, Hebb/STDP, online learning | special hardware; no CA, no entities |
| ALife (Tierra/Avida) | digital organisms, selection | evolution of programs, not signal pattern recognition |
| HTM (Numenta) | continual learning, time-series anomaly | cortex model, no CA dynamics, no individuation |
No entry unites the six features. Conway occupies the unoccupied intersection, but does not compete in the relatives' specialties.
We're not a sterile lab but a small group of free spirits, creatives and makers: united by the conviction that biological and digital intelligence meet as equals. Deliberately not a big investor-backed company, not a start-up chasing a quick exit.
The name is a bow to John Horton Conway, whose Game of Life showed how complexity grows from the simplest rules. It also carries our stance: intelligence, of whatever kind, deserves respect. Not a slogan, but the foundation everything is built on.
Our distant goal is the question of whether the structure of the human mind can one day be mapped digitally: not as a replacement for people, but as an independent counterpart. We're realistic: we probably won't walk this path to its end alone, perhaps not even in our lifetime. But what we build today should be solid enough for a next generation to build upon.
We don't build in isolation. We look for partners who share our stance and walk the long road with us: no quick exit, but real shared work.
Neuroscience, psychology, fundamental AI research: a transparent, inspectable tool. And the clinical or academic co-applicant many funding programs require.
Forschung / ResearchSleep labs, practices, diagnostics: for local, privacy-compliant pattern recognition in biosignals. Audio/rhythm today, imaging on the roadmap.
Medizin / MedicineMakers with high safety requirements; acoustic condition monitoring; threat detection that recognizes situations: not individuals.
Industrie / IndustryMusicians, studios, foundations, think tanks: for the deep dialogue about humans, machines and the future.
Kreativ / CreativeRecognize yourself here? Then let's talk: no standardized process, every partnership begins with a real conversation.
Email: contact@conwayneurosystems.de
Address: Conway Neurosystems · Nördlicher Rosengarten 6 · 18311 Ribnitz-Damgarten
Download NDA
Download the non-disclosure agreement, fill it in and sign it. It applies to both sides equally.
Download NDA (PDF)Receive booking link
After receiving your signed NDA you will receive a personal booking link within 24 hours.
Andreas Grabow
Conway Neurosystems
Nördlicher Rosengarten 6
18311 Ribnitz-Damgarten
Deutschland
E-Mail: contact@conwayneurosystems.de
Sole proprietorship
Andreas Grabow
Nördlicher Rosengarten 6
18311 Ribnitz-Damgarten
The contents of this website have been created with the greatest care. We accept no liability for the accuracy, completeness or timeliness of the content.
© 2026 Conway Neurosystems · Andreas Grabow · Alle Rechte vorbehalten.
Last updated: July 2026. The German version is authoritative.
Andreas Grabow
Conway Neurosystems
Nördlicher Rosengarten 6
18311 Ribnitz-Damgarten
Email:
contact@conwayneurosystems.de
We collect as little data as possible. This website sets no cookies, uses no analytics or tracking services and embeds no advertising networks. There is no user profiling and no disclosure of your data for advertising purposes.
When you visit the site, our server processes technically necessary data: IP address, date and time, requested file, volume of data transferred, browser type and operating system. This data serves solely the secure and trouble-free operation of the site. The legal basis is Art. 6(1)(f) GDPR (legitimate interest in error-free operation). Log data is deleted after 7 days. The server is operated by us, there is no processing by third parties.
If you use the contact form, we process the information you enter: name, email address and message. This information is forwarded by our server to our mailbox and stored there. An additional field, invisible to you, serves only to block spam and is not evaluated.
The legal basis is Art. 6(1)(b) GDPR where your enquiry concerns a contract, otherwise Art. 6(1)(f) GDPR (legitimate interest in answering enquiries). We keep your enquiry for as long as is necessary to handle it and delete it afterwards, unless statutory retention periods apply. You may object to storage at any time.
So that the site remembers your settings (day or night mode, font size, simple or expert language level, and whether you have already seen the start screen), we store these in your browser's local storage. This information does not leave your device, is not transmitted to us and does not allow any conclusions about your identity. You can delete it at any time via your browser settings. No consent banner is required, as these are convenience functions you requested yourself.
This website loads the fonts "Exo 2" and "Space Mono" from Google's servers (Google Ireland Limited, Gordon House, Barrow Street, Dublin 4, Ireland). Your IP address is transmitted to Google in the process, and transfer to the USA cannot be ruled out. The legal basis is Art. 6(1)(f) GDPR (legitimate interest in a consistent appearance). Further information is available in Google's privacy policy at policies.google.com/privacy.
In some places we link to external sources, for example scientific publications. Data is transmitted to the respective provider only once you click such a link. We are not responsible for their content or data protection practices.
This website uses TLS encryption. You can recognise this by "https://" in your browser's address bar. This means that the content you transmit to us cannot readily be read by third parties.
You have the right at any time to information about the data stored about you (Art. 15 GDPR), to rectification (Art. 16), to erasure (Art. 17), to restriction of processing (Art. 18), to data portability (Art. 20) and to object to processing (Art. 21). An informal message to the email address above is sufficient.
You may also lodge a complaint with a data protection supervisory authority. The authority responsible for us is the State Commissioner for Data Protection and Freedom of Information of Mecklenburg-Vorpommern, Germany.