Source: Drakkar-Software/OctoBot README and CONTRIBUTING. Researched 2026-11-04. The official guides and architecture pages returned 403/404, so most internals are unverified. Treat this file as a short outline, not a design source.
What is documented#
- Two shapes: classic ("manual") OctoBot with backtesting, Telegram and TradingView, and node mode, the default in the latest release, which is a backend for the new UI and the mobile app.
- Exchanges go through CCXT (more than 15).
- Strategy and exchange-specific code lives in the separate OctoBot-Tentacles repository; the generic core refuses per-coin logic. New dependencies need an issue or an optional import.
- Trading modes: Grid (many buy and sell orders at intervals), DCA (scheduled buys to lower the average cost), TradingView (alert driven), Index (baskets), ChatGPT/AI (LLM opinion, Ollama supported), and market making in a separate project.
- Paper trading is described as risk-free, and a built-in backtesting engine simulates a portfolio on historical data. Backtesting exists only in manual OctoBot, not in node mode.
Unverified#
Evaluator and strategy composition, fill model, fees, slippage, risk controls, order timeouts, recovery, persistence, configuration and secrets handling.
Recommendations for vukem-trading#
- Separate the trading mode (how orders are placed: grid, DCA) from the strategy or evaluator (what the signal is).
- Keep a plugin boundary for exchange and strategy code so the core stays generic.
- Mirror the headless engine plus separate UI/API split; this project already runs the web and the engine as distinct processes.
- Label paper trading clearly as its own mode and verify its fill and fee behavior before trusting results.
- If an LLM is ever added, it proposes validated drafts only; it never signs or sends orders (see the brainstorming decisions).