WickraStreaming-first technical indicators.

Install in 30 seconds ​

pip install wickra

A streaming RSI in every language ​

import wickra as ta

rsi = ta.RSI(14)
for price in live_feed:
    v = rsi.update(price)
    if v is not None and v > 70:
        print('overbought', v)

Requirements ​

The minimum supported version per language — prebuilt packages need only the runtime; building a binding from source needs the extra toolchain noted below.

Go (cgo) and R need a C compiler; Java runs with --enable-native-access. Full detail — runtime vs. build-from-source, per-language notes — is on the Requirements page in the docs.

Why streaming-first matters ​

A batch-only library recomputes its full indicator over every historical bar each time a new tick arrives. With a 5 000-bar history that's 5 001 work instead of 1. Wickra holds the indicator state in a struct and advances it by one update per tick — the cost stays flat as your history grows.

The streaming gap widens linearly with history length — see the benchmark page for the full table.

How the benchmark is measured ​

The per-tick numbers come from the same compare_libraries script the benchmark page runs: each library is handed an identical generated price series, warmed up over 5 000 bars, and then timed advancing one tick at a time. The Wickra figure is measured through the Python binding, so the small PyO3 boundary cost is already included rather than hidden in the bare Rust kernel. Reproduced on a Windows 11 / AMD Ryzen 9 9950X machine with Rust 1.92 in release profile — read the values as relative speedups on identical input, not as an absolute performance contract, since CPU, memory clock, and runtime versions all move the absolutes.

Who streaming-first is for ​

The same indicator object serves three workflows without a code change. For live trading, the optional Binance Spot WebSocket adapter pushes ticks straight into an indicator that updates in constant time, so per-tick latency stays flat even after a session has run for hours. For backtesting, you can replay a full history through that very same struct and trust that batch and streaming produce identical output — the equivalence is pinned by reference-value tests. For research, the Rust core and the Python, Node.js, WASM, C, C++, C#, Go, Java and R bindings all share one implementation, so a notebook prototype and a production service compute the exact same numbers.

The Wickra ecosystem ​

Wickra is the core of a family of twenty-three products, one repository each under wickra-lib. Every one of them is a Rust core with a CLI and the same ten-language binding surface as the core — released to the same registries from one pipeline, and checked byte-for-byte across all ten languages by a golden corpus in every repository. Each product installs by its own name (pip install wickra-<name>, npm install wickra-<name>, cargo install wickra-<name>, …) and has its own site.

Data — market data in, market data replayed, market data synthesised ​

Research — backtest, screen, search, train ​

Trust — prove, verify, benchmark and gate what a backtest claims ​

Surface — what a trader looks at and talks to ​

Edge — the core compiled for the browser, the binary, the chip ​

Try it without installing anything: all 514 indicators over a real Binance feed at live.wickra.org, the backtester compiled to WebAssembly at backtest-live.wickra.org, and one strategy spec side by side in Python, Rust, JS and Go at playground.wickra.org — zero backend, all of them.

The full indicator catalogue ​

514 indicators across twenty-four families. Every one is implemented once in Rust, re-exported by every binding, and pinned by reference-value tests.

See the full per-indicator deep-dives in the docs →