FastMM

FastMM is a low-latency market-making engine. Read the documentation.

https://github.com/user-attachments/assets/dcd130c5-28fe-482f-a757-d0fc6550c5b9

Features

• Strategies in C++ or Python. Python hooks marked @fastmm.hot compile to machine code with Numba.
• One strategy file for backtest and live. Replays are deterministic, and the simulated exchange models queue position.
• Account pools. Several sub-accounts trade behind one venue, and orders spread across them.
• Gateway. One fastmm-gateway holds the exchange connections; several strategy processes attach to it and can stop or restart without dropping them.
• Live control. fastmm-ctl changes a running session's parameters without a restart.
• A/B tests. Parameter sets run side by side, split by instrument, by time block or both, and are compared on markouts and net with paired intervals.
• Low-latency tuning. Core pinning, busy polling, and AF_XDP or DPDK receive for UDP multicast feeds. Order entry runs on the kernel TCP stack, which OpenOnload or XLIO can replace through LD_PRELOAD.

Quickstart

The package on PyPI is fastmm-engine, imported as fastmm; pip install fastmm installs a different project.

python3 -m venv .venv && . .venv/bin/activate
pip install "fastmm-engine[hot]"
fastmm init my-mm && cd my-mm
python backtest.py

This runs a strategy on a simulated market, with no account and no API keys (Linux x86-64, Python 3.10+). On Debian and Ubuntu, sudo apt install python3-venv first; on macOS or Windows, run it in Docker. The quick start continues to a live session against the local simulated exchange; C++ strategies start from the C++ quick start.

A strategy

import fastmm
from fastmm import Param


class TouchMM(fastmm.Strategy):
    half_spread_bps = Param(0.01, min=0.0, doc="half spread, bps")
    quote_qty = Param(0.002, min=0.0, doc="size per side")

    @fastmm.hot  # compiled to machine code with Numba
    def on_book(self, ctx, book):
        if not book.valid:
            ctx.pull()
            return
        half = book.mid * self.half_spread_bps * 1e-4
        ctx.clear()
        ctx.bid(book.mid - half, self.quote_qty)
        ctx.ask(book.mid + half, self.quote_qty)
        ctx.keep_passive()

Exchanges

Binance (spot and USDⓈ-M futures), OKX, Bybit, Gate, Hyperliquid, Lighter, Deribit, Gemini and Coinbase. A Nasdaq connector (TotalView-ITCH market data, OUCH 5.0 orders) trades only against the bundled simulator, fastmm-sim-itch.

If you are running FastMM on Binance, try bndesk, a real-time dashboard for your accounts.

License

MIT