Monte-Neo

Fast local research for trading strategies on Apple Silicon
Fee-aware next-bar economics · Monte Carlo · paper OMS
MIT · Python 3.11+ · Metal / MLX / Numba

Current: v0.15.1 · Changelog · Docs index · FAQ

What this is (and is not)

Job: on Apple Silicon macOS, build and verify trading-domain strategies very quickly with fee-aware next-bar economics you can re-check (export API + golden vectors).

Lanes: research bar (primary speed path) · Monte Carlo research · paper OMS (validation).

Not a goal: replace full event-driven production / live-bot platforms. Research sweep throughput is not an OMS event-loop claim.

Why Monte-Neo

Install

From git (today — free):

pip install "git+https://github.com/NeoZorK/Monte-Neo.git"
# Apple Silicon Metal / MLX:
pip install "monte-neo[apple] @ git+https://github.com/NeoZorK/Monte-Neo.git"

In-repo (recommended for contributors):

git clone https://github.com/NeoZorK/Monte-Neo.git
cd Monte-Neo
uv sync --extra apple   # on Apple Silicon; omit --extra apple on Linux/CI

PyPI: packaging is prepared ([apple] extras, PEP 440 wheel version). First upload waits on TestPyPI + maintainer OK — see PACKAGING.md.

Requirements: Python 3.11+. Best experience on Apple Silicon macOS. Numba CPU paths work more broadly; Metal/MLX are the [apple] extra.

Quick start

uv sync
uv run monte-neo
uv run pytest tests -n auto

Research export (recommended)

from monte_neo.backtest import (
    ExecutionModel,
    export_sma_sweep,
    synthetic_ohlcv,
)

ohlc = synthetic_ohlcv(100_000, seed=42)
model = ExecutionModel(commission_bps=5.0, slippage_bps=5.0, warmup_bars=50)
out = export_sma_sweep(
    ohlc["open"], ohlc["high"], ohlc["low"], ohlc["close"],
    combos=16,
    model=model,
    device="auto",  # Metal when safe; else cpu_numba (see fallback_reason)
)
print(out["device"], out.get("fallback_reason"), out["combos"])

Single bar backtest

from monte_neo.backtest import (
    ExecutionModel,
    frame_to_ohlc,
    run_bar_backtest,
    sma_signal,
    synthetic_ohlcv,
)

ohlc = frame_to_ohlc(synthetic_ohlcv(5_000, seed=42))
model = ExecutionModel(
    commission_bps=5.0,
    slippage_bps=5.0,
    size_fraction=0.25,
    sl_pct=1.0,
    tp_pct=2.0,
)
sig = sma_signal(ohlc["close"], fast=10, slow=40)
out = run_bar_backtest(
    ohlc["open"], ohlc["high"], ohlc["low"], ohlc["close"], sig, model=model
)
print(out["total_return"], out["metrics"]["max_drawdown"], len(out["trades"]))

More: docs/guides/quick-start.md · backtest engine · FAQ

How to use (research workflow)

1. Load or synthesize OHLCV (synthetic_ohlcv / your frame → frame_to_ohlc).
2. Set an ExecutionModel (fees, SL/TP, sessions, side mode).
3. Sweep with export_sma_sweep / export_batch, or a single export_single.
4. Check device, signal_device, and fallback_reason when using auto.
5. Optional depth: equity_stride, journal, plan_research_bytes / memory on exports.
6. Paper OMS (monte_neo.oms) only when you need event-lane validation — not for sweep cps claims.

Devices: auto · metal · cpu_numba (and MLX where signal paths allow). Oversized Metal jobs demote to Numba instead of hanging (v0.14.1+).

Features (honest)

Project structure

Monte-Neo/
├── src/monte_neo/
│   ├── backtest/      # Research bar engine + export
│   ├── oms/           # Paper OMS + accel
│   ├── core/          # Generator / Metal bridges
│   ├── data/          # Market data downloaders
│   ├── monte_carlo/   # MC methods
│   ├── metrics/       # Trading metrics
│   ├── cli/           # Interactive CLI
│   └── visualization/
├── tests/
├── docs/
└── docker/

Screenshots / demos

More under docs/assets/. Runnable script: examples/export_sma_sweep_quickstart.py.

License

MIT — see LICENSE.

Public repository: https://github.com/NeoZorK/Monte-Neo