A factory for the decision layers AI is missing.
Hokusai is a protocol for building shared, incentive-aligned models that don't exist or sit captured inside labs. The engineers who use these models earn a stake in the improvements they create.
How It Works
Data Submission
Submit high-quality, privacy-conscious data to improve AI models in your area of expertise.
Performance Evaluation
Your data is evaluated based on how much it improves model performance using transparent metrics.
Earn Rewards
Rewards are distributed based on performance improvement. Rewards can be in tokens that represent an ownership stake in the model or directly in USDC.
How Much Can I Earn?
Each model has a different performance metric and different rewards based on their economic value. Under the hood, Hokusai creates tokens for each model and rewards represent an ownership stake in the model.
Hokusai uses DeltaOne as a unit of measurement for performance improvement. For example, improving a model's AUROC from 0.81 to 0.84 would represent 3 DeltaOne units.
Each DeltaOne unit mints a predetermined amount of tokens, creating a direct link between value creation and rewards. Tokens can be converted to USDC at any time if crypto gives you the ick.
For Decision-Layer Builders
If you've identified a routing, selection, or optimization problem that currently lives in internal scripts or gets absorbed by labs, Hokusai gives you the primitives to launch it as a shared model with DeltaOne measurement, bonding-curve incentives, and on-chain attribution from day one.
DeltaOne and bonding curves
Contribute data that trains smarter AI and earn rewards.
Fair. Transparent. No middlemen.
Each model has a different performance metric and different rewards based on their economic value. Under the hood, Hokusai creates tokens for each model and rewards represent an ownership stake in the model.
Hokusai uses DeltaOne as a unit of measurement for performance improvement. For example, improving a model's AUROC from 0.81 to 0.84 would represent 3 DeltaOne units.
Each DeltaOne unit mints a predetermined amount of tokens, creating a direct link between value creation and rewards. Tokens can be converted to USDC at any time if crypto gives you the ick.
How does it work?
Hokusai creates tokens for each model. Tokens are created for performance improvements. Hokusai uses DeltaOne as a unit of measurement for performance improvement. For example, improving a model's AUROC from 0.81 to 0.84 would represent 3 DeltaOne units.
Hokusai creates a bonding curve for each model. The amount each data supplier earns for a DeltaOne improvement depends on the amount of USDC available. Create strong incentives for better performance.
This creates strong incentives for data suppliers to contribute high-quality data and for model developers to improve performance.
Core protocol properties
Trustless Protocol
Automatically measures and rewards data contributions based on their impact on model performance.
Earned Ownership
Earn tokens proportional to your data's measured impact on model performance.
Direct Revenue
No middlemen. Your data's impact directly translates to your earnings.
Privacy First
Your data remains private while still contributing to model improvements.
Reference apps
The protocol is already broad enough to support multiple model categories. The coding task router leads here, alongside the existing healthcare, finance, and language examples already visible in the Hokusai model directory.
Routing / Multi-Model Coding
Coding Task Router
The first decision layer on Hokusai: a shared router for multi-model coding harnesses that improves through real routing outcomes.
Open routerImaging / Radiology
Chest X-Ray Classifier
A reference model showing how the protocol rewards measurable performance lifts on clinically relevant benchmarks.
Explore modelsFinance / Trading
Crypto Trading Prediction
A market-facing example of shared model improvement, transparent metrics, and on-chain contributor incentives.
Explore modelsNLP / Healthcare
Medical Text Analysis
Evidence that the same protocol primitives apply across specialized domains where better data compounds model performance.
Explore modelsBuild on the protocol
Got a decision layer worth sharing?
If you've identified a routing, selection, or optimization problem that today gets solved in scripts or captured by labs, Hokusai's primitives DeltaOne measurement, bonding-curve incentives, on-chain attribution give you a path to build it as a shared, owned model. Talk to us.