Banbot — event-driven multi-market trading bot

An open-source algorithmic trading bot for crypto markets, built around an event-driven engine that runs the same strategy code for both backtesting and live trading. It is the execution layer of this wiki’s founding trio — where Kronos predicts and TensorTrade learns a policy, banbot is the engine that actually places and manages orders. T3: a vendor/author’s own project README — capable and concrete, but the performance claims are self-reported and unbenchmarked by a third party.

What it is

• Multi-everything by design. Banbot manages multiple exchange accounts, trading pairs (symbols), strategies, and timeframes simultaneously in one unified framework — they combine independently rather than being locked into rigid pairings.
• Exchanges. Binance, OKX, and Bybit, across spot and futures markets.
• One codebase, two modes. The single strategy implementation is used for both backtesting and live deployment, so a backtested strategy ships to production without a rewrite.
• Web UI. Strategy development, backtesting, and deployment happen through a browser interface — no IDE required.

Architecture & performance

• Event-driven model. Market data is processed as a stream of events, which the project frames as preventing lookahead bias (a backtest can’t peek at future bars) while leaving strategy implementation flexible. This is the defining design choice — see event-driven-trading for why event-driven backtests are the bias-safe norm.
• Speed claim. “Backtest 1 year of klines in seconds” — the headline performance pitch, attributed to the Go core.
• Hyperparameter optimization. Built-in strategy tuning supporting multiple algorithms: Bayesian / TPE (model-based), random, and CMA-ES (model-free) variants — the same optimizers documented as subjects in ../optimization-algorithms-wiki; here they’re a means to tune a strategy (see strategy-optimization for the trading-side mapping), a clean illustration of this wiki’s scope boundary.

Stack & license

• Go (~72%) for the core engine; Svelte (~20%) + TypeScript (~7%) for the web UI; some Python. By banbox.
• Dual-licensed: GNU AGPLv3 plus a commercial option.

Why it matters here

Banbot completes the prediction → strategy → execution stack. It is also the source that tipped the financial-ml cluster to three and triggered this spoke’s spin-out. Its event-driven, no-lookahead backtesting echoes the exact design concern TensorTrade addresses from the RL side, and its embedded CMA-ES/TPE optimizers tie the trading domain back to the metaheuristics spoke.

Related

backtesting · strategy-optimization · event-driven-trading · algorithmic-trading · tensortrade · kronos-financial-foundation-model · banbox · synthesis