Kronos

Type-safe backtesting and live trading for crypto algo strategies

Table of Contents

Algorithmic crypto trading lives or dies on backtests. Most of the tools that run them are slow, imprecise, non-deterministic, and force engineers to rewrite the same strategy from scratch every time they deploy on a new exchange. We got tired of watching that pattern, so we built Kronos: a deterministic backtesting and live-trading engine where strategies iterate at the pace of an idea, not a deploy cycle. Type-safe Go end to end, an exchange-agnostic SDK, and a hosted engine that runs up to 100× faster than the conventional tools.

Hosted platform is in beta, gated behind sign-up. The CLI is out under MIT, and the SDK is public in the same GitHub org.

How Kronos came to be

Kronos is a Holdex build. No external founder, no third-party team — the same engineers we'd put on a co-build with someone else's company are writing this one, except the customer is us. We're the trading desk that wants the tool, the engineering team that builds it, and the support layer for the early adopters using it.

Two layers go to market. The hosted beta is for market makers and serious algorithmic trading desks that want the managed surface — direct exchange data feeds, custom order-book integration, the risk plumbing wired in. The open-source side is two repositories, both Go, both MIT: kronos (the framework) and kronos-sdk (the strategy SDK).

What Kronos does

Strategies compile into Go plugins that hot-reload without restarting the framework. Each one runs in its own isolated process. Lifecycle is HTTP, with real-time monitoring over Unix sockets and HTTP APIs. Bubble Tea and Lipgloss handle the terminal UI, Cobra runs the CLI, Fx wires dependencies, Ginkgo runs the tests. Decimal arithmetic is financial-grade, so float drift doesn't quietly poison results. RSI, MACD, EMA, SMA, and Bollinger Bands ship in the box. Multi-timeframe analysis works out of the gate, and so does cross-exchange logic like arbitrage detection.

Interop is type-safe Go all the way down, with optional escape hatches: ML inference via gRPC, ONNX Runtime in-process, or HTTP APIs to models in any language. The promise the docs make is plain: "same code works in backtesting and live trading. No environment-specific logic. No adapter layers."

Exchange coverage skews toward perpetuals. Hyperliquid perps are stable. Bybit and Paradex sit in the active build lane. Binance is supported in the SDK for backtest and data. The hosted platform supports major exchanges, with custom adapters delivered inside 48 hours.

The hosted version uses the same engine and adds a managed layer on top. Simulation runs against real exchange order-book data, with automatic historical retrieval and latency-and-depth modelling so the backtest fills behave the way live fills will. Data ingestion is hybrid WebSocket + REST, sourced directly from exchanges or from vetted data partners. Custom order-book integration is supported for desks that want to pipe in their own feeds.

Risk management goes deeper than the standard stops. Automated stop-loss, trailing stops, and position sizing are table stakes. On top of those: real-time order-book-based mitigation, configurable circuit breakers, order rate limiting, and a pre-trade policy engine that runs every check you'd want before a fill. Reporting bundles the standard metrics (VaR, drawdown, P&L, Sharpe) with an interactive trade visualiser and replay controls (pause, rewind, fast-forward through any session).

Build status

MetricSnapshot
StatusBeta hosted platform; MIT-licensed framework; public SDK releases at v0.0.7
Repositoriesgithub.com/backtesting-org (kronos, kronos-sdk, live-trading, documentation)
Performance claimUp to 100x faster than conventional backtesting tools
Stable connectorsHyperliquid perps
In developmentBybit, Paradex, Binance backtesting/data
Product directionWebSocket monitoring, paper trading, notifications, and broader exchange coverage

Kronos is what happens when Holdex turns an internal pain into infrastructure: the trading desk gets a faster loop, and the engineering team turns that loop into a reusable platform.

The hosted platform is open to early adopters now, with special terms for the desks that ship strategies on it during beta. If you're running an algo book and the existing tooling is the bottleneck, that's the conversation we want to have.

Channels

Have a suggestion?Edit this page