Your AI trading team that still loses money, but now reproducibly

What it does

Lumibot is a Python algorithmic trading framework that wraps backtesting, paper trading, and live broker execution behind the same strategy code. The newer twist is a built-in agent runtime: you can assemble “AI trading teams” where researcher, bull, bear, and trader agents pass context and arguments through a strategy loop, with only designated agents allowed to submit orders.

The interesting bit

The framework treats AI agents as first-class citizens inside a traditional quant loop, not as chatbots bolted onto a dashboard. You can lock agents to read-only roles (allow_trading=False), pipe SEC filings and FRED macro data into their context, and run the entire multi-agent debate through a backtest first. The same Strategy class runs on Yahoo historical data or Alpaca live orders by swapping the runner.

Key highlights

• Deterministic Python strategies and LLM-agent strategies coexist in the same abstraction
• Built-in agent tools: market state, DuckDB queries, technical indicators, SEC fundamentals, FRED macro, Alpaca news, local memory
• Multi-agent patterns: research → bull/bear debate → risk-aware trader, or Citadel-style sector pods, or Buffett-style value desks
• One code path for backtest, paper, and live; broker swap is a runner change, not a rewrite
• Hosted cloud option (BotSpot) with bundled data, scheduling, and MCP integration for coding agents

Caveats

• The README’s performance tear sheet uses a 2026 date range, which appears to be simulated future data for demonstration; the project explicitly warns that “backtests are not expected future performance”
• AI agent execution depends on external LLM keys (Gemini by default); missing keys halt the strategy with a printed error
• The framework is tightly coupled to a commercial hosted offering (BotSpot) for convenient data and deployment

Verdict

Worth a look if you want to experiment with multi-agent LLM strategies in a structured quant loop with real broker hooks. Skip it if you need battle-tested execution infrastructure or are allergic to framework-plus-cloud ecosystems.