Build order-book depth
from day one.
Vinu Liquidity is market-making software built for Vinu Exchange deployments. It generates two-sided quotes for spot markets using reference prices from 13 venues. Its model adjusts pricing and order sizes, applies cost floors, and hedges inventory through OKX, Bybit, or Binance perpetuals when risk thresholds are reached, within limits set by your team.
Liquidity you own, not liquidity you rent.
Vinu Liquidity runs within the Vinu Exchange stack, enabling two-sided quoting as new spot markets launch. Exchange operators control inventory and loss limits, while token projects can support post-listing market activity through reference-based pricing and configurable risk controls.
New pair · ACME/USDT
Listed today on your own order book
Spread
12 bps
Inventory
+1,299
Depth on every pair from launch
No separate market-maker contract required. Define your inventory and loss limits while the engine manages two-sided quotes within those boundaries. Quoting can become one-sided or pause automatically when risk thresholds are reached.
VNX/USDT · listing day
Quoted off VNC/USDT × price_multiplier
VNC/USDT index
0.04169
× price_multiplier
25
= fair value
1.0423
Two-sided presence · day 1
97.5%
Spread
18 bps
Profile
DepthBootstrap
A market that stays quoted after listing
Generate bids and offers around a reference price derived from your chosen market. Configurable spread caps and inventory limits help maintain controlled quoting as trading activity develops, with automatic adjustments when market conditions change.
Six services, one closed loop.
Reference prices enter the system, quotes are generated, fills return, and inventory is hedged through an event-driven workflow. Each Rust service has a defined responsibility and its own database schema, making individual processing stages easier to monitor and failures easier to isolate.
oms-service
Order management
Compares the desired ladder with live orders and only touches what changed, so resting quotes keep their queue position. It turns engine trades into fills.
Hover or tap a service to see what it does
Hover or tap a service to see what it does
↺ Inventory and hedge cost flow back to the quoter
cex_accounting.inventory · cex_hedger.cost → quoter-service
Six services, one Postgres
One cex_liquidity database, with its own schema and migrations for each service.
Event bus, at-least-once
A duplicate-fill guard per market makes a redelivered fill harmless.
Rust, fixed-point money
Fixed-point decimal for every price, size and PnL value, and no unsafe code.
One price you can trust, built from many.
Each market uses a reference price derived from 13 venues through 17 feed adapters. Outliers are filtered, stale feeds are excluded, and a weighted median is published at a controlled rate. A confidence score helps operators assess the reliability of the reference price.
17 feed adapters, one catalogue
Sources are picked per market in the operator console
WebSocket5 native adapters
subscribe payload + JSON pointers
HTTP polling12 providers · 500 ms default, 100 ms minimum
Watchdog
30 s silence
Backoff
500 ms → 30 s
Breaker
5 failures
Config
hot reload
BTC-USDT index
Weighted median of fresh, in-band sources
Index
64,257.3
OKX
WS
64,254.6
64,259.7
Bybit
WS
64,235.4
64,241.8
Coinbase
WS
64,255.7
64,263.4
Kraken
WS
65,502.3
65,512.8
Binance.US
WS
64,252.7
64,266.8
BtcTurk
HTTP
64,212.6
64,243.4
Kraken is 194 bps off the median · trimmed
Trim band
150 bps
Stale after
5 s
Publish
≥ 50 ms
Heartbeat
500 ms
Volatility and confidence
EWMA in bps per 100 ms, rate-independent
fast > 2× slow · regime widening on
Confidence
fresh share × e^(−dispersion / 50 bps)
Fresh
6 / 6
Dispersion
15.6 bps
no quotes below 20%
Model-driven quoting with a built-in cost floor.
The GLFT optimal market-making model determines spreads and inventory skew, with continuous calibration based on executed trades and exchange market data. Bounded adjustments help manage adverse selection, while a cost floor covering maker fees, hedging costs, and a safety margin takes precedence over configured spread caps.
Built against informed flow
Each layer widens or tilts the quote by a bounded amount
Microprice fair value
clamped to oracle
OFI tilt
±8 bps
Flow-toxicity widening
up to 20 bps
Inventory-velocity meter
2 s window
Latency cushion
vol·√(p95 / 100 ms)
Confidence widening
up to 10 bps
Regime widening
+5 bps
Confidence gate
no quotes < 20%
BTC-USDT ladder
3 levels a side · 5 bps step · post-only
Full optimal spread · PnL first
microprice, clamped to oracle
Inventory
+7.31 BTC
Bid side adds inventory · size 20%· taper floor 20%
Spread
27.6 bps
Skew
−6.6 bps
Size / level
−8%
Step
5 bps
Why this spread
ProfitGlft half-spread, built up layer by layer
Cost floor
floor bindsMaker fee2.0
Hedge cost4.5
Safety1.0
Toxicity
0.00
Toxicity widening +0.4
Model is under cost · the quote sits on the floor
Orders that keep their place in the queue.
The order manager compares the target quote ladder with existing orders, retaining unchanged orders to preserve queue position where possible. Each rejection type triggers a defined response, from requoting and backoff to cooldowns or safe mode, preventing uncontrolled retry loops during volatile market conditions.
Minimal-churn reconcile
BTC-USDT · 3 levels a side
Index
64,229.5
Ask L3
btc-usdt:ask:3
64,311.8
64,296.3 · 0.12
Keep+2.4
≤ 3 bps
Ask L2
btc-usdt:ask:2
64,279.7
64,264.2 · 0.10
Keep+2.4
≤ 3 bps
Ask L1
btc-usdt:ask:1
64,247.6
64,232.1 · 0.08
Keep+2.4
≤ 3 bps
Bid L1
btc-usdt:bid:1
64,197.4
64,197.4 · 0.08
Create+4.6
> 3 bps
Bid L2
btc-usdt:bid:2
64,165.3
64,165.3 · 0.10
Create+4.6
> 3 bps
Bid L3
btc-usdt:bid:3
64,133.2
64,133.2 · 0.12
Create+4.6
> 3 bps
Stable id per market, side, level · keeps queue
Kept
3
Cancelled
3
Created
3
Queue kept
73%
Reject-aware reactions
Every reject class has its own response
Reject burst
20 / 20 in 5 s
- Requote
Filter
20 filter rejects in 1.2 s
- Back off
Rate limit
retry_after 500 ms
- 1 s cooldown
Post-only would take
ask L1
- Reconcile from stream
Timeout
no ack within 3 s
- Safe mode
Balance
or open-order cap
- Halt
Risk state · KYC · AML
account level
Inventory hedging that follows your risk limits.
The hedging engine aggregates inventory across markets into a net delta for each underlying asset. Exposure within configured bands is left unchanged. When a threshold is crossed, bounded IOC orders are placed sequentially through OKX, Bybit, or Binance USDT perpetuals, with safeguards against duplicate submissions.
Delta band and IOC ladder
BTC · net of all spot markets · Bybit BTCUSDT perp
Net delta
+146 USDT
Intent · h-btc-0412
Sell 0.0095 BTC
- IOC 5Partial
64,147.9
h-btc-0412-1
- IOC 15Partial
64,083.7
h-btc-0412-2
- IOC 50Waiting
63,859.1
h-btc-0412-3
Partial fill · next rung after 1 s
Venue router
All-in hedge cost, BTC perps
OKX
healthy · margin 41%
8.1 bpsBybit
healthy · margin 38%
6.1 bpsBinance
half-open · margin 44%
7.9 bps
Pin a venue · optional fallback to auto
Margin and funding gates
Bybit · cross margin buffer
Margin buffer
14.7%
Hedge cost
7.82 bps
- Taker fee5.00
- Half-spread0.51
- Impact0.70
- Adverse funding0.61
- Slippage reserve1.00
To quoter as cost floor
published every 1 s
No fresh estimate → 10 bps
Risk controls built into every layer.
Each market operates with its own state machine and configurable loss, inventory, and drawdown limits. Portfolio-wide controls monitor aggregate exposure across markets and hedge positions. When a portfolio limit is breached, the engine halts quoting and initiates position flattening. The halt persists across restarts until an authorized operator re-enables trading.
Market state machine
ETH-USDT · one state engine per market
Trigger
Oracle, venue, params healthy
Transitions
- 14:02:07.750HALT → OFFOnly an operator re-enable leaves HALT · re-enabled → OFF → WARMUP
- 14:02:10.250OFF → WARMUPOperator enables · warm-up checks start
- 14:02:12.750WARMUP → ONOracle, venue, params healthy · ladder posted
Burst rule
> 5 fills / 2 s
Stale index
> 1 s → pulled
Safe mode
cancel · widen
Leaving HALT
operator only
Burst rule
> 5 fills / 2 s
Stale index
> 1 s → pulled
Safe mode
cancel · widen
Leaving HALT
operator only
Per-market guardrails
ETH-USDT · checked on every fill and mark
Mark-to-market loss
186 / 1,000 USDT
at the limit → HALT
Inventory
6.51 / 10 ETH
past 1× → one-sided · past 2× → HALT
Drawdown from peak
139 / 600 USDT
at the limit → HALT
Drawdown velocity · 30 s
41 / 250 USDT
at the limit → HALT
Live quotes
Long past max inventory, the market only sells until it is back inside.
Portfolio caps & kill-switch
All markets · evaluated every 250 ms
Spot MtM
−2,415
Perp MtM
+1,245
Fees
−323
Hedge cost
−164
Global max loss
1,657 / 20,000
Global max drawdown
3,272 / 5,000
Gross notional
192,921 / 500,000
Net notional · ETH
20,271 / 50,000
Loss, drawdown or gross notional cap → global halt
Every market → HALT
12 markets · resting quotes cancelled
Spot flattened
reduce-only IOC · 50 bps through
Perp legs closed
reduce-only · OKX · Bybit · Binance
Halt persisted
survives restarts · resume is an operator action
13
reference price venues
3
perp venues for hedging
50 ms
fastest index cadence
0 bps
quoted below the cost floor
Proven in simulation before it quotes.
The deterministic backtester exercises production quoting and order-reconciliation logic against seeded synthetic markets, modelling queue position, latency, partial fills, and toxic order flow. Scenario tests cover venue outages, margin shortfalls, and service restarts, while the full test suite runs in CI on every pull request. These tests validate system behaviour under simulated conditions, not live trading performance.
Profile A/B · same market path
Synthetic seeded market · illustrative output
- Seed
- 7
- Sim time
- 10 min
- Step
- 100 ms
- Re-quote
- 250 ms
- Latency
- 20 ms
- Toxic flow
- 15%
Production code under test
Modelled
- Quoted spread18.4 bps10.2 bps
- Two-sided presence97.1 %99.6 %
- Fill ratio0.180.31
- Spread capturerelative
- Adverse-selection costrelative
- PnL net of hedgingrelative
- Max drawdownrelative
- Cancel / place churnrelative
Money KPIs are shown relative to each other. Every figure here comes from a synthetic market, not live trading.
Scenario suite
Mock perp venues · deterministic
Partial fills escalate the IOC ladder until flat
partial_fills_escalate_the_ladder_until_flat_then_clear
Timeout resolved by client id, never resent
timeout_is_resolved_by_client_id_and_never_resent
Venue down → widen → halt → operator clear
venue_down_widens_then_halts_by_age_and_needs_operator_clear
Margin shortfall switches the router
margin_shortfall_switches_the_auto_router_to_the_other_venue
Funding spike widens the band, never past the cap
funding_spike_widens_the_band_but_never_past_the_unhedged_cap
Global flatten closes the perp leg reduce-only
global_flatten_closes_the_perp_leg_reduce_only
Oracle → quoter → OMS → accounting
oracle_to_quoter_to_oms_to_accounting
Restart without duplicate orders
restart_reconciliation_avoids_duplicate_orders
Loss breach halts through the state engine
loss_breach_halts_via_state_engine
CI gate
Every pull request and push to main
- ok
Formatting
$ cargo fmt --all --check
- ok
Lints, zero warnings
$ cargo clippy --workspace --all-targets -- -D warnings
- ok
Unit, scenario and integration tests
$ cargo test --workspace
- ok
Service contracts stay identical
$ diff control_internal.proto ×4
- ok
Backtest A/B scenarios
$ cargo run -p backtest --example ab_profiles
Guaranteed in code
#![forbid(unsafe_code)]
in every library crate
Fixed-point decimal
every price, size and PnL value
Quoter invariants
bid below ask · never across the touch · no negative size
Runs on your servers, under your control.
Vinu Liquidity is self-hosted software, not a managed cloud service. Deploy signed container images from your own registry using a Helm release on your Kubernetes cluster. Authentication, secret management, monitoring, and upgrades remain under your organization's control.
Installation
One Helm release in your own cluster and namespace
Release path
- ✓Signed release
container images from Vinu
- ✓Your registry
mirrored, pinned by digest
- ✓helm upgrade
on your schedule, by your team
Your cluster · namespace vinu-liquidity
mTLS between every service
oracle-service
reference prices
1.8.0
ready
quoter-service
quote ladders
1.8.0
ready
oms-service
order reconcile
1.8.0
ready
accounting-service
risk and PnL
1.8.0
ready
hedger-service
perp hedging
1.8.0
ready
control-api
rules and registries
1.8.0
ready
dashboard
liquidity.ops.example.com
1.8.0
ready
Your systems
- Identity provider
OIDC or SAML 2.0
- Secret store
venue and database credentials
- Prometheus
/live · /ready · /metrics
- Audit log
every allow and deny
Delivery
container images
Install
one Helm release
Schema
migrates itself
Upgrades
when you decide
Operator sign-in
Your IdP over OIDC or SAML 2.0 · no local passwords
selin.kaya@your-exchange.com
IdP · OIDC- IdP groups
- liq-operators
- Dashboard role
- Market operator
Permissions
- liquidity:market:read
- liquidity:market:manage
- liquidity:market:operate
- liquidity:ruleset:propose
Roles come only from IdP groups. Leave the IdP and dashboard access ends with it.
Secrets and audit
References in config · decisions on record
values.yaml · references only
- hedge.okx.apiKeyread at runtime
secretRef: secretstore://liquidity/hedge-okx
- hedge.bybit.apiKeyread at runtime
secretRef: secretstore://liquidity/hedge-bybit
- database.passwordread at runtime
secretRef: secretstore://liquidity/db
Never written to env vars or the database
Audit log
- ALLOW14:21:06
liquidity:market:operate
ETH-USDT · enable
- ALLOW14:20:19
liquidity:market:read
risk overview
- ALLOW14:19:32
liquidity:ruleset:propose
ruleset v14 · to quorum
A guided tour of the operator dashboard, with SSO, roles, audit log and release settings.
Open the dashboard demoRun it in your cloud, or in your own data centre.
Vinu Liquidity runs on any conformant Kubernetes cluster. Terraform modules orchestrated with Terragrunt provision the cluster, database, and secret store, while a Helm release deploys the market-making services. Choose a supported cloud provider or an on-premises environment, configure the deployment variables, and run the same stack within infrastructure you control.
AWS
EKS · Secrets Manager · RDS
Huawei Cloud
CCE · DEW · RDS
Azure
AKS · Key Vault · PostgreSQL
Google Cloud
GKE · Secret Manager · Cloud SQL
On-prem
Kubernetes · Vault · PostgreSQL
Infrastructure as code
Terraform modules · Terragrunt
$ terragrunt run-all apply
target = huawei · provider huaweicloud/huaweicloud
- ✓networkhuaweicloud_vpc.main
- ✓kuberneteshuaweicloud_cce_cluster.liquidity
- ✓databasehuaweicloud_rds_instance.liquidity
- ✓secretshuaweicloud_csms_secret.venues
- ✓liquidityhelm_release.liquidity
Apply complete · Huawei Cloud · 0 changed elsewhere
Repository layout
Same modules, any target
- modules/# provider-neutral
- network · kubernetes · database
- secrets · liquidity · observability
- live/
- root.hcl# remote state, provider
- prod/
- env.hcltarget = "huawei"
- kubernetes/terragrunt.hcl
- liquidity/terragrunt.hcl
- staging/
Switching to Huawei Cloud changes one variable; the modules and the chart stay the same.
Your own engine, or a market-making firm?
External market makers can provide capital, trading expertise, and liquidity support. A self-hosted market-making engine gives your team direct control over quoting strategies, inventory, and operational data. The right approach depends on your capital requirements, available resources, and preferred level of control. Here's how the two models can differ.
Side by side
What you give up, and what you keep
Topic
External market maker
typical arrangement
Vinu Liquidity
running on your cluster
Inventory
External market maker
May involve provider capital, token loans, or other inventory arrangements defined by contract.
Vinu Liquidity
Quotes from your own inventory. No token loans, no options to grant.
Custody
External market maker
Asset custody and account control depend on the provider's operating model and agreement.
Vinu Liquidity
The software does not take custody of assets. Spot quotes run on your exchange, while hedging uses your configured external venue accounts.
Visibility
External market maker
Reporting frequency, data access, and transparency depend on the provider and contract.
Vinu Liquidity
Every order, fill and decision in your own console and your own database.
Where orders go
External market maker
Execution and hedging venues depend on the provider's strategy and agreed terms.
Vinu Liquidity
Quotes rest only on your own order book. Outside venues supply prices and perp hedges.
Focus
External market maker
The provider may manage trading strategies across multiple clients and markets.
Vinu Liquidity
An engine dedicated to your markets, tuned by your own team.
Who runs it
External market maker
The firm's traders and capital; you see the outcome.
Vinu Liquidity
Your operators, inside guardrails, a quorum-governed ruleset and a kill-switch.
Terms
External market maker
Bespoke deal terms that are often confidential: loans, options, retainers.
Vinu Liquidity
A software licence. You know what you pay for and what you run.
The left column describes common industry arrangements; terms vary from firm to firm. With an outside firm you also get its capital and staff; with Vinu Liquidity, your team operates the engine.
From first call to live quotes.
The rollout follows five steps, from initial scoping and ruleset design to installation, warm-up, and live quoting. After the scoping call, your team receives a rollout plan, ruleset proposal, and pricing within two business days. Each market goes live only after completing warm-up checks and receiving explicit approval from your operators.
Rollout plan
Your venue · one market at a time
Step 1
Scoping call
Which markets you want covered, how much inventory and loss you will carry, and where you want to hedge.
Step 2
Ruleset proposal
Spreads, caps, price sources and hedge venues in one document, sent with the rollout plan and pricing within two business days.
Step 3
Install on your cluster
Signed images in your registry, one Helm release, sign-in through your identity provider, secrets read from your own store.
Step 4
Warm-up
Each market moves from OFF to WARMUP. Price data, venue health and parameters are checked while your operators review the market.
Step 5
Live with guardrails
The market goes ON with a two-sided ladder. Loss, inventory and drawdown limits are armed and the kill-switch stays with your operators.
Step 1
Scoping call
Which markets you want covered, how much inventory and loss you will carry, and where you want to hedge.
Step 2
Ruleset proposal
Spreads, caps, price sources and hedge venues in one document, sent with the rollout plan and pricing within two business days.
Step 3
Install on your cluster
Signed images in your registry, one Helm release, sign-in through your identity provider, secrets read from your own store.
Step 4
Warm-up
Each market moves from OFF to WARMUP. Price data, venue health and parameters are checked while your operators review the market.
Step 5
Live with guardrails
The market goes ON with a two-sided ladder. Loss, inventory and drawdown limits are armed and the kill-switch stays with your operators.
Market state
Within two business days
- Rollout plan
- Ruleset proposal
- Pricing
Step 4 · Warm-up
- Reference price freshdone
- Venue book and parameters healthydone
- Operators review the marketpending
Questions operators ask before the first call.
Find out how Vinu Liquidity is deployed, where it places orders, who controls assets and credentials, and how operating rules, upgrades, and licensing work.
No. It is closed-source software delivered as signed container images. You pull them into your own private registry and install them with one Helm release on your own Kubernetes cluster. There is no shared, multi-tenant cloud behind it.
Only on venues running the Vinu Exchange stack, where it places orders on your own order books. External exchanges are never quoted on: they serve as reference price sources and, optionally, as perp venues for hedging.
No. The engine runs on your cluster and trades from accounts on your own venue. Perp venue API keys and database credentials stay in your own secret store and are read at runtime; configuration holds only references to them, never the secrets themselves.
Through your own identity provider over SSO, with OIDC or SAML 2.0. There are no local passwords. Roles come from your IdP groups and map to read, manage, operate and ruleset-propose permissions, and every allow and deny decision is written to an audit log.
The engine's operating rules for pricing, quoting, order handling and hedging live in one governed ruleset. A change is proposed as a complete document, approved by quorum, and the proposer cannot vote on their own proposal. Per-market parameters and portfolio caps are changed by operators with the right permission. Both keep a versioned history.
We publish new image tags; you pull them into your registry and run helm upgrade when it suits you. Each service migrates its own database schema during the rollout, so there are no manual database steps.
Vinu Liquidity is licensed as software, not run as a service. We share a proposal after a scoping call, once we know which markets you want covered and how you want to run them.
Book a scoping callLiquidity tuned to your markets.
We plan your Vinu Liquidity deployment market by market, from trading pairs and reference price sources to inventory limits, risk controls, and hedge venues. Share your requirements, and our team will define a rollout plan and an initial ruleset tailored to your exchange.
- 01
Market-by-market rollout
Each market completes warm-up checks and goes live only when your operators explicitly enable quoting
- 02
Risk workshop
Loss, inventory, drawdown and portfolio caps set with your risk team before the first quote
- 03
Hedge venue setup
OKX, Bybit or Binance perp accounts connected and tested on testnet first
- 04
A named quant engineer
Direct access to a quant engineer supporting your deployment and quoting configuration
Next steps
- We reply with a few questions about your markets
- A 30-minute call with one of our quant engineers
- We backtest your pairs and send a written proposal