SOUL-E replaces the ML team for forecasting, end to end. It designs, tests, deploys and keeps improving its own forecasting models, for any business, asset or data.
Tested on unseen dataPre-registered evaluationRuns on consumer hardware
SOUL-E · one unit · BTCChampion kept
Existing modelTested · beaten
Existing modelTested · beaten
New, by SOUL-EBeats the best · kept
Its forecast, in plain words“Wider range this week: volatility has risen six days in a row.”
Forecasting skill1.9–2.4× the best rival
RetrainingChallenger vs champion
Benchmarked against open forecasting models from
Google TimesFM 2.5Amazon Chronos-2Amazon Chronos-BoltSalesforce Moirai 2.0Datadog Toto 2.0NXAI TiRex-2AutoGluon-TimeSeriesLightGBMXGBoostCatBoostGoogle TimesFM 2.5Amazon Chronos-2Amazon Chronos-BoltSalesforce Moirai 2.0Datadog Toto 2.0NXAI TiRex-2AutoGluon-TimeSeriesLightGBMXGBoostCatBoost
The platform
One engine replaces the ML team.
Give SOUL-E data and a question. It finds the best forecasting models that already exist, invents new ones to beat them, proves every candidate on data it has never seen, and rebuilds the winner as the world changes.
Tier 0 · DataAny time-stamped data, plus the decision it should serve
FinancePrices and volumes
EnergyGrid prices, load, weather
FreightLane volumes and rates
RetailSales, stock, lead times
YoursExported files or a feed
Tier 1 · EngineCore
SOUL-ESelf-Optimizing Universal Learner – Engine
01Find the best existing models, from every lab, and test them the same way.
02Invent new model designs that beat them, for this business and data.
03Prove every candidate on unseen data; a rule set in advance picks the winner.
04Improve as data shifts: a challenger must beat the champion to replace it.
Model-agnosticEach new model from Google, Amazon or open source becomes its next starting point.
Decision-trainedChosen on your decision: ranking for traders, revenue for batteries.
CompoundingEvery model and result is kept, so each new field starts ahead.
Low computeSpecialised models that train on ordinary hardware.
Tier 2 · ForecastsDelivered with their accuracy, explained in plain language
Asset rankingsTomorrow's likely winners and losers
Risk rangesHow far a portfolio could move
Bid signalsWhen a battery should charge and sell
Demand forecastsWhat to order, and when
API, MCP, productsWherever the decision is made
Security and governance
Customer data stays privateModels never leave our cloudForecasts hashed before the outcomeIndependent scoringSovereign cloud region
1.0×More accurateThe forecasting skill of the best rival model, on unseen stocks and crypto.
0.00 hFaster to buildTo train the final models, on a laptop processor.
$0Cheaper to runOf consumer hardware ran the whole search. One GPU behind a rival model costs US$25,000–40,000.
Small tests, on purpose. A live, independently scored record is next.
Products
Built on SOUL-E.
Finance first: tools for traders and the platforms they trade on, in beta today. Energy, enterprise licences and an API for AI agents follow.
Quant-grade strategy testing, without codeDescribe a strategy in plain English. Every variation is validated on unseen data and scored for overfitting and bias before it becomes a bot.
A new field changes the inputs and the outputs, not the engine.
NowFinancial marketsAsset rankings, risk ranges and strategy backtests for traders, platforms and funds.Read more →
NextEnergyPrice forecasts and bid signals for grid batteries, optimisers and power traders.Read more →
ComingCommodities and freightPrice, spread, rate and capacity forecasts for desks, brokers and 3PLs.Read more →
ComingRetail and platformsDemand and supply forecasts for stores, marketplaces and their planners.Read more →
For developers and AI agents
An analytical brain your agents can call.
Agents that plan, price and order need forecasts. Send SOUL-E your data and a question through the API or MCP, and get forecasts back with their accuracy. The models stay in our cloud.
1from cyaura import SoulE
23 soule = SoulE(api_key="…")
4 job = soule.forecast(
5 data="sales_daily.csv",
6 target="units", horizon="14d",
7 decision="reorder", # trained on your decision8 )
9 job.champion # the model kept, scored on unseen data10 job.explain() # why, in plain words
Trust
AI you can check, not just trust.
CheckableEvery forecast, on the recordLive forecasts are hashed and time-stamped before the outcome is known, scored by rules registered in advance, and checked by an independent party.
PrivateYour data stays yoursEach customer gets models built for its own data. SOUL-E learns from every result without sharing anyone's data. Models, code and weights stay in our cloud.
Non-custodialNever holding anyone's moneyOur finance products use read-only connections and trade-only keys. They make the data clear and never tell people what to do.