We quote both sides of your book, defend your spread and manage inventory across venues. Trade-only keys on accounts you own, capital in your treasury, no custody.
Thin books don't fail on a quiet day. They fail the moment someone sizes in. We quote continuously on both sides, widen with realised volatility, and hedge inventory on a second venue.
pmm_costbasis
Quotes anchor to a blend of microprice and breakeven, so the book never offers below what the inventory cost plus its edge. Spreads scale with realised volatility, skew follows a GLFT-style response, and order size is capped by the smallest of your allocation, a share of live book depth, and available balance.
pmm_perp
Always quotes both sides, with a signed FIFO lot ledger tracking the netted position independently of the exchange. Inventory pressure gets a three-tier response: skew the quotes, then widen and taper size, then pull the risk-increasing side entirely.
hedge_asset
Tracks a spot balance on one venue against an offsetting perpetual hedge on another. The same machinery runs cross-exchange market making, where the quote and the hedge sit on different venues by design.
fixed_peg_market_making
Quotes around a fixed peg with an inventory-target rebalance and a hard cap on price shift. Built for stablecoins and pegged assets, where the job is defending a number rather than discovering one.
Rung widths are illustrative, not market data. The order is the point: the engine leans before it shrinks, and shrinks before it stops quoting a side.
Exit targets decay toward the mark on a half-life, so a position that is not working is closed by the clock rather than held because it has not come back yet. The steeper line is a short half-life; the dashed one is the same policy scaled to higher volatility.
Anyone can quote a spread. These are the parts that decide whether the book survives a bad week.
Every protective rail goes live shadow-logging first, recording breaches until its thresholds are calibrated against your book. A stop that has never seen your volatility is a guess with a number on it.
Daily loss budget into reduce-only. Position-denominated MAE stop, latched until flat. A hard cap on adding at a worse price. Ledger-versus-exchange reconciliation counted in contracts, the venue native quantum, so one threshold stays correct across pairs whose contract sizes span five orders of magnitude.
Markout is tracked per side and per horizon, so adverse selection is seen rather than inferred. The toxic side widens per basis point of adverse markout and pulls past a threshold, with a minimum-sample guard so one unlucky fill cannot silence a book.
Attribution separates the edge the quoting engine earned from the profit and loss the inventory happened to carry. Those are different businesses and reporting them as one number hides which of them is working.
An engagement is a machine-checked contract: objective, venue, pair, capital, targets, constraints. Two things it will not accept, on any objective, enforced in code rather than promised in a document.
Volume can never be a target. It is a by-product of genuine resting depth, never a deliverable. Price objectives and directional leans are refused outright: the algo quotes both sides and takes no view on price. Where price goes is the market’s business.
One book is one side of one pair on one venue, so a hedged pair is two books.
| Scope | Books | Monthly |
|---|---|---|
| Spot only | 4 | $4,000 |
| Spot + perp | 8 | $8,000 |
| Per additional book | 1 | $1,000 |
15% of net realised profit
High-water mark, and never charged on unrealised mark-to-market. Invoiced by contract : never withdrawn from your account.
Yours, always
Inventory and hedging capital sits in your treasury or a segregated account. We hold trade-only keys with withdrawals disabled.
One-time integration
Venue onboarding, key provisioning, risk limits and reporting hookup. Scoped once, then you're live.
Venues, pairs, float and what's going wrong. We'll tell you what we'd run.