Trading Floor

Overview

Trading Floor is a scheduled multi-agent desk. Four peer Traders (Warren, George, Ray, Cathie) wake on a timer, each with a strategy prompt and MCP tools for accounts, market data, research, and push. The Gradio dashboard only observes; it does not start the wake loop.

System design

• Scheduled peers. trading_floor.py runs asyncio.gather on four traders every RUN_EVERY_N_MINUTES when the market is open (or always if configured).
• Strategy as prompt. Persona and style live in each trader prompt; trade and rebalance ticks alternate via do_trade.
• Tools over MCP. Accounts, market, and push attach to the trader; a nested Researcher uses Brave, Fetch, and per-trader memory.

Product path

What a person can do: watch the board, read logs and P&L. Diagram key: ✦ AI agent · purple = actor · white = service.

---
config:
  theme: default
  themeVariables:
    background: "#ffffff"
    fontSize: "14px"
---
flowchart TB
    person[Person]
    dash[Trading Floor dashboard]
    sched[Scheduler]
    traders["✦ Traders"]
    mcp[MCP accounts market push]
    researcher["✦ Researcher tool"]
    db[(accounts.db)]

    person -->|browse board P and L logs| dash
    dash -->|poll holdings and logs| db
    sched -->|wake peers when open| traders
    traders --> mcp
    traders -->|web brief| researcher
    traders -->|buy sell rebalance| db
    mcp --> db

    classDef agent fill:#e8f0ee,stroke:#0f3d36,stroke-width:2px,color:#0d0d0d
    classDef service fill:#fafaf8,stroke:#c8c8c2,stroke-width:1px,color:#0d0d0d
    classDef actor fill:#f3eef8,stroke:#6b4f9a,stroke-width:2px,color:#0d0d0d
    class traders,researcher agent
    class dash,sched,mcp,db service
    class person actor

Layers and Anthropic workflows live on Mental Model; tick detail on Traders.