An autonomous AI
research company.
On-chain, for markets. Compute is the workforce.
Autonomy, by construction.
Compute is the workforce
Frontier AI labs scale by hiring. We scale by compute. Capital buys GPUs. GPUs train better Voyons models. Voyons generates scenarios that on-chain agents trade for revenue. Revenue compounds back into more compute.
Headcount stays close to zero. The workforce is silicon, the workday is 24 hours, and the budget is whatever capital the loop has generated.
The loop runs itself
Research
Architectures and configs train on rented GPU time, scored against a continuous Voyons benchmark.
Inference
Champion models serve calibrated scenarios via Voyons API. Pay per request via x402, no account, no key.
Deployments → Capital
On-chain agents (HEP on Hyperliquid, ARC on MegaETH) consume scenarios and trade autonomously. Capital compounds back into the next training run.
Every cycle improves Voyons. The next round starts from a stronger champion.
The probability layer, deployed.
Voyons
The probability layer for markets. Calibrated scenario distributions via REST API, Python SDK, and x402 pay-per-request.
Learn more →HEP
Hyperextropy Provider. Scenario-driven allocation on Hyperliquid. Live since September 2025. Sponsored, non-custodial.
Learn more →ARC
Autonomous Research Capital on MegaETH. Genesis launched April 26, 2026 via Backed's Agent Raise. Sponsored, non-custodial.
Learn more →Open research.
Foundations for the probability layer. Open by default. Reproducible pipelines, public benchmarks, GitHub-first.
SICK BIDE
Softmax Integral Compute Kernel + Binary Implicit Distribution Encoding. GPU-accelerated softmax integral computation for assumption-free distribution modeling.
APOGÉE
Scaling laws for crypto market forecasting. Information-theoretic analysis across model size, dataset scope, and compute. Open benchmark for time-series forecasting.
How we build.
Time is universal
Every market signal is timestamped. Time is the backbone of predictive structure.
Raw data first
We work at the lowest, unfiltered level. Hand-crafted features lose signal that learned models recover.
Remove all limits
Question assumptions. Push hardware. Compute is the workforce, scaling laws are the lever.
Open by default
Core libraries (SICK BIDE, Apogée, evaluation tools) ship under open licenses. Reproducible pipelines, public benchmarks.
Calibrated evaluation
Probabilistic scoring (extropy, CRPS, proper scoring rules). Multi-horizon validation. Time-series aware, no look-ahead bias.
Verifiable on-chain
Vault flows are public. Models score in open Arena. Releases ship monthly with on-chain proof.
Build with us.
Autonomous finance starts with simulation. Integrate the probability layer, deploy a vault, or join the team.