About Minerva AI

A complete AI trader: Minerva drives two research apps, one backtest bridge, and a composable MetaTrader EA.

We build AI infrastructure for traders who refuse to choose between quality and cost.

The four pieces

Minerva AI is built from four composable pieces that talk to each other in a loop. The chat is the interface — you describe what you want in plain English. Minerva drives the other apps on your machine. EvoAI talks to MetaTrader. The EA does the actual trading.

  • The desktop app — a single workspace where you describe what you want to trade, the system runs the discovery, validation, and deployment, and you can take over manually whenever you want.
  • NN Lab — deep reinforcement learning. Trains DQN, R2D2, GRU on your historical bar data.
  • GA Lab — genetic programming. Evolves readable trading rules from primitives.
  • EvoAI — the bridge. Drives MetaTrader 4 or 5 under Wine and harvests backtest reports.
  • Minerva AI EA — the executor. 14 strategy modules wired together with configurable voting.

Each piece is useful on its own. The chat ties them together but isn’t strictly required.

What Minerva actually does

Minerva interprets your request, picks the right app, sets sensible defaults, runs the workflow, and streams progress back to you. For example:

You: “Train a DQN on EURUSD H1 for 5,000 steps and validate it in MT5.”

Agent: “Training started. Best Sharpe: 1.2, net profit $1,847, 48 trades, max DD 4.2% on the validation set. The DQN picked up on EMA-crossover-with-volume confirmation.”

Every action Minerva takes is auditable. The audit log shows exactly which tools were called, with what inputs, and what happened. You stay in control.

Who it’s for

Minerva AI is built for traders who want an edge without writing a line of code:

  • Discretionary traders who want to backtest a hypothesis without hiring a quant.
  • Quant-curious traders who want the rigour of a research workbench.
  • Trading teams that want a shared workspace for strategy research.
  • Anyone who wants to validate that a strategy actually works in MT4/5 before risking real capital.

Where we’re going

The roadmap is on the project plan. Highlights:

  • Multi-symbol support so one chat session can drive a portfolio of strategies.
  • Self-learning: Minerva re-trains models when live performance drifts from backtest.
  • Bidirectional voice control so you can drive the workflow from the chart.
  • Shared workspaces for trading teams.