Language bindings

All bindings expose the same core model: strategy callbacks, order emission, position queries. What each one carries around that model differs, and the table below says where.

Language	Guide	Reference
Python	Python	API reference
Node.js	Node.js	API reference
Codon	Codon	API reference
JavaScript (embedded)	JavaScript	API reference
C API	C API	reference

What each binding carries

Capability	Python	Node	Codon	JavaScript (QuickJS)
Strategy callbacks — trade, book, bar	yes	yes	yes	yes
BacktestRunner — event-driven replay (run_csv, run_tape, run_ohlcv, run_bars)	yes	yes	yes	no
Engine + SignalBuilder — backtest a pre-built signal list	yes	yes	yes	yes
SimulatedExecutor, BacktestResult	yes	yes	yes	yes
DataReader, DataWriter	yes	yes	yes	yes
Bar aggregators — time, tick, volume, range, Renko, Heikin-Ashi	yes	yes	yes	yes

The embedded JavaScript binding has no BacktestRunner: the event-driven replay entry points are unreachable from QuickJS, and a strategy backtests there by building a signal list and running it through Engine. Everything else in the table is reachable from all four.

Which one to use

Python is the easiest starting point for backtesting — BacktestRunner for event-driven strategies, Engine for pre-built signal lists. Node.js makes more sense if your infrastructure is already JS.

Codon has nearly identical syntax to Python but compiles to native code. Start with Python, switch to Codon if you need it.

The embedded JS binding runs QuickJS inside the C++ process — no separate Node.js runtime. Useful for scripted rules and backtesting where spinning up an external runtime isn't practical.

For other languages, use the C API.

AI-agent companion

lrvx-mcp is a Model Context Protocol server. AI coding agents (Cursor, Claude Code, Cline) spawn it locally and ask it about lrvx before generating code. It exposes tools that resolve a symbol across surfaces (lookup_symbol), enumerate one surface's exports (list_bindings), return a starter strategy class that parses + validates (scaffold_strategy), pull example code from the docs corpus (get_example), and search the docs (docs_search). The package version is bumped in lockstep with lrvx and the npm package, so the agent's view of the surface matches what you installed. Setup is in the package README.

lookup_symbol and list_bindings cover six surfaces — cpp, capi, python, node, codon, quickjs — down to individual methods, reported as Owner.method. C++ is listed first in a resolution, because it is the engine and every other surface wraps it; the row carries the header to include. So SimulatedExecutor resolves in all of them, BacktestConfig resolves in C++ alone, and run_csv resolves as BacktestRunner.run_csv.

Note that capi is not C++: it is the flat extern "C" ABI built for FFI consumers, so BacktestRunner appears there as the opaque handle LrvxBacktestRunnerHandle. Ask for cpp when you mean the engine's own classes.