Summary

This project provides a persistent memory layer for AI trading agents, enabling them to store decisions, learn from outcomes, and autonomously evolve strategies across sessions. It targets AI agents interacting with financial platforms (MT5, Binance, Alpaca), offering agents the ability to retain and leverage trading history for improved performance.

How It Works

A three-layer architecture, inspired by ACT-R, manages memory. Trades are stored with context. Before new trades, similar past trades are recalled using Outcome-Weighted Memory (OWM), prioritizing successful outcomes. An Evolution Engine analyzes this memory to discover patterns and generate novel strategy hypotheses, validated via Deflated Sharpe Ratio (DSR). This offers structured, outcome-aware memory and automated strategy refinement beyond generic vector stores.

Quick Start & Requirements

Installation is via pip (pip install tradememory-protocol) or Docker (docker compose up -d). A guided setup wizard (tradememory setup) handles platform detection (e.g., Claude Desktop, Cursor), terms acceptance, configuration generation, and health checks. tradememory doctor performs core checks. Links to the tutorial and API reference are available in the documentation.

Highlighted Details

• Includes 15 native MCP (Meta-Communication Protocol) tools for core memory, cognitive functions, and evolution.
• The Evolution Engine demonstrated strategy performance exceeding random baselines, achieving a walk-forward Sharpe Ratio of 3.24.
• Live paper trading of Strategy E is operational on Binance via GitHub Actions.
• Integrates Outcome-Weighted Memory (OWM) and ACT-R cognitive science principles.

Maintenance & Community

No specific details regarding maintainers, community channels (like Discord/Slack), or active sponsorships are provided in the README. The project is listed as "Built by Mnemox."

Licensing & Compatibility

The project is licensed under the MIT license. However, a prominent disclaimer states it is "For educational/research purposes only. Not financial advice," which may impose restrictions on commercial use or integration into closed-source, production trading systems.

Limitations & Caveats

The primary limitation is the explicit restriction to "educational/research purposes only," raising questions about production commercial viability despite the permissive MIT license. Integration with specific AI agent platforms is also required for full functionality.