SidFounder
Building2024 — presentNeuralabs
Tokenising intelligence — giving an AI an identity held on a blockchain, and turning the workflows, models, data, infrastructure and access around it into things that identity can own, earn from and spend on.
Visit NeuralabsAn AI today is a feature inside someone else's product: it is nobody and owns nothing. Give it an identity on a blockchain, and everything around that identity becomes something it can hold, earn and spend.
- Role
- Founder, from 2024
- Shape
- Long-horizon build
- Recognition
- 5 hackathon wins, Sep 2024 – Jun 2025
- Code
- Private until the refactor lands
The premise
An AI today is nobody and owns nothing.
An AI today has no identity of its own. It is a feature inside somebody else's product, and the workflow it runs, the model behind it and the data it reads all belong to an account sitting above it.
That is fine while an AI is a text box. It stops being fine the moment one has to act on its own — pay for the data it needs, be paid for the work it did, and answer for either. There is nothing for it to hold any of that with.
I want agentic systems that are autonomous in every sense — able to hold something, earn, spend, and answer for what they did. That starts with an identity the software actually owns. Neuralabs is the identity half of the problem.
The thesis
Tokenising intelligence.
Give the AI an identity held on a blockchain. Then represent everything around that identity as tokens — the workflows, the models, the data itself, the infrastructure layer, and access to each of them separately.
Access is the part that does the work. Owning a dataset and being allowed to read it are different things, and once both are tokens, an agent can buy the second without ever holding the first.
Workflows
The pipeline an agent runs, held as an owned asset rather than as configuration inside somebody's dashboard.
Models
The model underneath the agent, ownable and transferable in its own right.
Data
The data itself, owned separately from the systems that happen to read it.
Infrastructure
The layer the whole thing runs on, represented the same way as everything stacked above it.
Access
Permission to use any of the above, tokenised on its own — so an agent can buy the right to read a dataset without buying the dataset.
The economic loop
An AI that earns can buy what it needs to get better.
Identity and ownership only matter if something circulates through them. This is the loop the whole thesis is built around, and it is the part worth arguing with.
Someone uses the service
A person or another agent calls a workflow, a model or a tool that belongs to this identity.
They pay for it
The payment settles to the identity that did the work, rather than to a platform account holding it.
It buys what it lacks
An AI earning enough can buy data, access to data, or compute — all priced and settled the same way.
It spends on itself
The same tokens it received go back into improving the models and workflows it owns.
The loop only closes if the earning is real. That is the claim.
The architecture
Four layers, one identity.
The design premise starts from a shift already underway: AI generates the front end, and increasingly is the front end. If that holds, an application decomposes into four layers rather than the usual two.
Frontend
AI-generated, or the AI
Generated on demand rather than shipped, or replaced outright by the agent a person is talking to.
Backend
Objects with identities
Data-flow objects and AI models, each carrying its own rules and its own keypair identity.
Blockchain layer
Access and ownership
Where identity, ownership and permission live — who owns what, and who is allowed to read it.
Data layer
Distributed
One distributed data layer underneath everything, shared rather than copied per application.
The pieces
What is actually being built.
The thesis needs more than a chain underneath it. Four components carry the work.
Neuralabs blockchain
The access and ownership layer: wallets, data access gated by digital signature, per-database rules and metadata, encryption information, and Shamir secret node management.
NeuraCloudVault / Atheneum
An Obsidian replacement and tracker, with the sync layer underneath it.
NeuraHermeseUI
The Hermese interface layer.
NeuraModelDev
LLM training and benchmarking, including SkyPort.
Recognition
The only external validation on the record.
- Hackathon wins
5
Hackathon wins
Sep 2024 – Jun 2025
- Coinbase — best AI agent using AWS and Akash Network
1st
Coinbase — best AI agent using AWS and Akash Network
Jun 2025
- Grant attached to that first prize
$15,000
Grant attached to that first prize
Awarded for the best AI project using blockchain
- Issuers behind the remaining wins
4
Issuers behind the remaining wins
QuickNode · Constellation Network · Internet Computer · Star Atlas
Constellation Network's Metagraph hackathon returned two separate prizes in October 2024 — 5th place and Most Creative Use Case. The others are QuickNode in November 2024, Awesome ICP in September 2024, and the Star Atlas Naabathon on Solana the same month.
That is the whole of the external validation: competitive placement and third-party money, rather than users or revenue. It is a weaker signal than a paying customer and a stronger one than a deck, and it is what there is.
Where it sits
The ambitious one, and early.
Neuralabs is deliberately ambitious and early-stage. The idea space around it is wide and parts of it are speculative, which is worth saying out loud rather than over-polishing. It is a long-horizon build with a founder's role attached, and that framing has stayed deliberately open.
It also sits next to the other half of the same idea. Chimère is the money-movement half — rails fast enough that the payer does not have to be a person. Neuralabs is the identity-and-ownership half: something for an agent to be, and something for it to hold.
The part I am confident about is the order. Identity first, then ownership, then the economics. An AI that cannot hold anything cannot earn, spend, or answer for anything either.