How they work

  • StockMarketBot
  • CryptoMarketBot
  1. Watch: code reads about 150 news sources and live prices

  2. Look up: the agent wakes and checks prices, news and history

  3. Decide: the agent trades, a risk engine checks every order

The marketbots Docker stack, on the owner's server, runs both dashboards, their MongoDB, the stock engine, the crypto engine, the local LLM, web search, the shared scraper and the history store. Its connections:

  • Browser to Dashboards, one per bot: HTTPS
  • Dashboards, one per bot to MongoDB, a separate database per bot: Read data, write sign-ins
  • Browser to Stock engine: Engine WebSocket stream
  • Browser to Crypto engine: Engine WebSocket stream
  • Stock engine to MongoDB, a separate database per bot: Own trading state
  • Crypto engine to MongoDB, a separate database per bot: Own trading state
  • Stock engine to Local LLM, Qwen 3.6 on two GPUs, for the agents, news summaries and research
  • Crypto engine to Local LLM, Qwen 3.6 on two GPUs, for the agents, news summaries and research
  • Stock engine to History store, price bars and news archive
  • Crypto engine to History store, price bars and news archive
  • Stock engine to Web search, SearXNG
  • Crypto engine to Web search, SearXNG
  • Stock engine to Shared scraper
  • Crypto engine to Shared scraper
  • Stock engine to the Alpaca paper account and the Kraken paper wallet
  • Crypto engine to the Alpaca paper account and the Kraken paper wallet
  • Stock engine to news sources
  • Crypto engine to news sources
  • Web search, SearXNG to search engines and article pages
  • Shared scraper to search engines and article pages

Code in each engine watches the news and live prices all the time and wakes the bot's agent when something matters. The agent runs on the local model: it looks things up in the paper account, the history store and its own database, then decides. A risk engine checks every order in code before it reaches the paper account, and every step streams live.