Trading strategies are living organisms that are born, compete, adapt, and survive in real markets. NOME turns every experiment, extinction, and deployment into memory, so the next generation starts smarter than the last.
Currently evolving across crypto perpetual futures. Forex and equities ecosystems in development.
Quantitative trading is stuck in a craftsman model. A human designs a strategy, backtests it, deploys it, and watches it decay. When it stops working, they start over. This is intelligent design — slow, biased, and fragile.
Markets are ecosystems. They reward adaptation, not rigidity. Platforms today treat strategies as static artifacts. Evonome treats them as living organisms.
A natural biological process, not a software workflow.
New species are born through adversarial debate between AI agents and evolutionary parameter search. Only ideas that survive intellectual natural selection move forward.
Promising candidates undergo controlled experiments. Machine learning layers adapt organisms to shifting market regimes, forging resilience into each organism's genome.
Every experiment is preserved — successes and extinctions alike. No discovery is ever lost. A fossil's DNA can be revived and recombined into future species at any time.
A sealed observation environment. Organisms trade on live market data with no capital at risk. The terrarium reveals whether an organism can breathe outside the lab — or whether it was only alive under controlled conditions.
Validated organisms are released into real markets. They coexist in colonies, compete for capital allocation, cross-pollinate signals, and adapt to changing conditions. The strongest thrive. The failing go extinct.
An invisible memory network connecting every organism, experiment, and decision. Like the fungal networks beneath a forest, it carries discoveries forward, prevents repeated mistakes, and enables cross-pollination between organisms that have never directly interacted.
NOME is the memory layer that turns every experiment, extinction, promotion, and deployment into reusable intelligence for the next generation.
Two AI agents — a creative quant and a risk manager — debate strategy structure through multiple rounds of adversarial critique. Surviving hypotheses enter a bounded evolutionary search: mutation, crossover, and Pareto selection across thousands of parameter combinations. Only statistically robust candidates graduate.
Candidates are enhanced with machine learning layers — gradient-boosted trees, neural networks, and ensemble methods. Walk-forward optimization ensures no future data leakage. Feature importance analysis via SHAP identifies which market signals each organism actually depends on. Statistical hardening tests reject fragile results.
Every experiment — successful or extinct — is preserved as an immutable specimen with its full genome (configuration), metrics, and lineage tree. The archive is searchable by performance characteristics, parameter ranges, and evolutionary ancestry. Any specimen's DNA can be extracted for recombination into new experiments.
Organisms execute against live market data in real-time with zero capital exposure. Performance is monitored against the metrics that earned them graduation. Behavioural drift detection flags organisms whose live patterns diverge from their training characteristics. Only organisms that replicate their lab performance are promoted.
Live organisms are grouped into colonies (portfolios) with dynamic capital allocation. Organisms that consistently perform receive more capital; those under sustained stress are throttled or retired. Cross-pollination allows organisms to share learned signals — a momentum organism's regime detection can inform a mean-reversion organism's entry timing.
A persistent knowledge graph records every discovery, dead end, and decision across the entire platform. Before any new experiment begins, the Mycelium is consulted — surfacing relevant prior work, warning against previously failed approaches, and suggesting recombination opportunities. Institutional knowledge compounds across every generation.
Strategies aren't manually crafted. They emerge through adversarial selection and survive through adaptation. The system discovers what humans can't design.
The Mycelium ensures no lesson is learned twice. Every dead end, every breakthrough, every mutation feeds the next generation. Knowledge compounds, it never resets.
The Ecosystem isn't a list of bots. It's a self-regulating biome where organisms compete for capital, share signals, and evolve together as market conditions shift.
Asset-class agnostic architecture. New ecosystems are plugin deployments, not rewrites.
It's the only system that gets stronger the longer it runs. Evonome doesn't build strategies. It breeds them.
Request Access to the Biome