Deltix: trading engines, quantitative workflows and market-data infrastructure

Deltix supplies software to banks, brokers, exchanges, market makers, proprietary desks and investment firms. Its products separate research, execution, market data and customer-facing trading functions, while sharing underlying technology. FX is one of the supported asset classes alongside securities, derivatives and digital assets. The profile covers this trading stack rather than the wider EPAM consulting portfolio.[1]

CryptoCortex’s broker edition combines consolidated books, configurable pricing tiers, SOR, execution algorithms and account management. The exchange edition adds matching and market-data distribution, with a simulation/UAT environment. Its OTC configuration supports streaming prices or RFQ and customer-specific mark-ups; the desk using it acts as principal. These are customer-operated financial workflows, not a representation that Deltix itself supplies the trading counterparty.[6]

The connector catalogue includes FXSpotStream, Cboe FX, LMAX MTF, Gold-i and Prime XM alongside securities and digital-asset destinations. Each connector has its own supported order types, state handling and configuration. A catalogue entry establishes documented integration support; it does not provide the client’s venue account, credit arrangement, market-data entitlement or permission to trade.[10]

The LMAX MTF connector illustrates why venue-specific specifications matter. Its documentation expresses FX quantity as 10,000 units of the first currency per contract for LD4/TY3, but one unit for NY4. Market orders allow IOC/FOK; limit orders additionally allow Day. Connection setup uses credentials and a secure tunnel. The disconnect notes describe cancellation and reconstruction of open-order state differently according to sequence resets and whether LMAX initiates reconnection. Using a generic FIX template without these unit and session rules can therefore misstate order size or remaining exposure.[11]

TimeBase handles structured events such as quotes, trades and order-book changes, with historical and live access. Its published integration routes include Java, C++, C#, Python and Go libraries, REST/WebSocket, and connections with ClickHouse, Timescale, S3, Kafka and Grafana. Data collection, distribution and downstream analysis can consequently be placed in separate services. The chosen schema and retention design still determine what a particular installation can replay.[7]

CryptoCortex lists custody integrations with BitGo and Fireblocks, API integration with KYC solutions, and identity-provider connections including Auth0, Ory Hydra, Keycloak and Okta. These are external service interfaces. Custody, identity checks and the customer’s financial operating permissions remain separate from the software licence.[6]

Deltix directs buyers to sales and a demonstration. The reviewed public product pages do not give a standard production tariff, currency, minimum commitment or per-trade charge. Licence scope, deployment, implementation, connectors, support and recovery architecture therefore require a written commercial proposal. A demonstration request is not a published free production trial.[24]

The open-source TimeBase edition is not the whole commercial trading stack. Ember’s Java setup guide specifically requires Enterprise TimeBase for market prices, and places compiled venue connectors in the additional EmberPack component. Those dependencies belong in the licensed bill of materials before a project is costed.[12]

Recovery infrastructure also changes operating cost. Deltix’s local-cluster design requires leader and follower resources plus an arbiter; the remote DR design can provision some costly resources only when failover is needed. Neither is a published flat-price add-on.[15]

MarketMaker is offered on premises or cloud hosted, including a private AWS instance, with bot, asset and system-level risk controls. Its REST/WebSocket API can create and run bots and modify risk rules. Hosting this application therefore includes operating active trading logic and credentials, not merely displaying analytics.[4]

The Java quickstart specifies JDK 11 or 17, an Enterprise TimeBase connection, correctly named market-data streams and separate configuration and working directories. It first uses a simulated destination before adding a real connector. This makes market-data mapping, execution connectivity and state persistence distinct acceptance tasks. Example configuration values and old console-output version numbers are not a current supported-release matrix.[12]

Ember’s risk tutorial says checks are disabled by default and must be explicitly configured. Rules group exposure by dimensions such as account, source, symbol or currency; a position projection can exist without a limit. Position limits account for filled exposure and working orders, including the proposed order. Missing attributes and unmatched rules also affect acceptance. A visible position screen is therefore insufficient evidence that the intended order controls are active.[13]

Version and data-model migration need their own plan. TimeBase’s edition comparison warns that the listed editions are not compatible with older TimeBase versions, without supplying a universal conversion procedure. Ember’s journal tooling and the chosen data warehouse must be included when an installation changes version or moves to another platform.[8][19]

The public diagnostic material identifies concrete steps for order rejects: check the halt state and risk limits, confirm the destination and routing rules, verify that the algorithm or connector is running and connected, and identify which component generated the rejection. A failed order can originate inside Ember or at the external venue; those require different investigation paths.[22]

For an order that appears active after it should have completed, Deltix asks for the order identifier, event history and preferably an Order Debugger printout. The guide gives [email protected] as the escalation address. Support hours, response targets and restoration commitments are not specified there; TimeBase Enterprise support is explicitly governed by a custom agreement.[21][8]

The trading REST/WebSocket gateway assigns each client an API key/secret and an order-source identifier. By default clients submit orders only for that source and retrieve only their own orders and positions. This separation is a documented trading-gateway behaviour, rather than a statement that every administrative interface has the same visibility.[18]

The Administrative REST API is enabled by default but refuses unauthenticated operation. Depending on configuration it accepts validated identity-provider bearer tokens or directory-backed HTTP Basic credentials. Read-only endpoints are available to any authenticated user and position queries cover all sources; state-changing actions require explicit permissions for trading halts, risk updates, position adjustments, services or FIX sessions. Deltix instructs operators to use TLS, restrict permissions and keep this API off the public Internet. Its broader position visibility is important when allocating access to support staff or external automation.[17]

In Ember Monitor, halting trading rejects new orders and amendments while allowing cancellations. Cancel All is a separate action, and per-projection halts can isolate a specific source or algorithm. Likewise, the risk screen permits limits to be added, removed and edited. These controls need assigned operating authority: a halt alone should not be interpreted as confirmation that every live order has been cancelled.[16]

EPAM’s public privacy policy concerns its websites and marketing activities. It describes purpose-based personal-data retention, international processing with safeguards, and applicable access, deletion and portability rights. That policy is not a production trading-data agreement. Hosting locations, access to customer order histories, retention, export assistance and deletion after termination need to be set in the applicable service and data-processing terms.[26]

EPAM’s 2019 Form 10-K records that it acquired 100% of Deltix and its affiliates on 3 February 2020. This is the acquisition closing date, rather than the date of a later press announcement.[25]

The current Deltix website continues to identify the business as part of EPAM Systems, Inc.[1]

Deltix’s main contact page gives 41 University Drive, Suite 202, Newtown, Pennsylvania, US. The Ember site separately publishes a Boston US-headquarters address and product sales contacts. These contact details do not identify every contracting affiliate; the agreement should name the supplying entity and licensed services.[24][2]

The software has its own operational identity: the technical FAQ explains that Ember is the internal name of Deltix’s order and execution management product. TradeHub is its OEMS use case within the broader product family. Procurement should consequently reconcile the product names in the proposal with the engine, connectors, data service and documentation versions actually delivered.[23][2]

Operator

Deltix, part of EPAM Systems, Inc.

Service scope

Institutional trading software, algorithmic execution, quantitative research, market-data infrastructure and broker/exchange technology.

Customer types

Banks; Brokers; Exchanges; Market makers; Proprietary trading firms; Investment firms

Delivery

On premises; Hosted; API

Pricing publication

Public and quoted pricing

Platforms

TradeHub / Ember; QuantOffice; TimeBase; MarketMaker; SpreadTrader; CryptoCortex

Integrations

FIX; Java; REST / WebSocket; FXSpotStream; Cboe FX; LMAX MTF; Gold-i; Prime XM

Pricing scope

Commercial products are quoted. TimeBase Community uses Apache 2.0; Enterprise licensing and support are commercial.

Implementation

Product-specific deployment with explicit risk rules, venue configuration, simulation and recovery design.

Support

Public technical documentation and email escalation; Enterprise TimeBase support is under a custom agreement.

Permissions and regulatory scope

Trading technology for customer-operated financial workflows; product access does not establish venue, credit or financial-service permissions.

Provider metadata checked 20 Oct 2026Source 1Source 4Source 10Source 9Source 8Source 13Source 12Source 15Source 21

The standard SOR constructs an execution plan using available liquidity, venue fees and discounts, account balances, recent rejects, and instrument precision or minimum sizes. A rejected child order can exclude that venue for the current parent and trigger replanning; if no venue remains eligible, the parent is cancelled. Balance checks have an important limit: missing balances are treated as zero by default, while polled balances are approximate and may be consumed by activity outside Ember. An exchange can therefore still reject a request for insufficient funds. This is a configurable execution algorithm, not a guarantee of fill or final cost.[14]

QuantOffice separates historical backtesting, live-data paper trading and live execution. The development environment supports parameter runs, different calendars and sessions, and simulations ranging from bars to L2 market-by-price or market-by-order. Strategies can be debugged before they are switched to live TradeHub connectors. The realism of results depends on the selected data and simulator; a successful simulated run does not establish live execution quality.[3]

SpreadTrader adds programmatic management of synthetic instruments and algorithmic orders through REST/WebSocket. Its published workflow includes creating hedging logic, supervising live hedge orders, viewing volume and P&L, and monitoring system metrics in Grafana or Graylog. A multi-leg spread therefore has both trading parameters and an operational monitoring layer; the selected strategy defines how each leg and hedge are managed.[5]

The journal preserves trading events and can produce snapshots to accelerate restart. Snapshots represent current recoverable state, not necessarily the full historical order trail: completed orders may be removed and past fills compressed into positions. Long-term warehousing can retain the fuller history, while journal-to-JSON and reverse conversion provide a portable inspection route. The documentation warns against manually deleting journal files; the compactor coordinates removal with downstream consumers. Export planning must distinguish recoverable state, audit history and market data.[19]

Risk-limit changes are also journalled. Deltix documents export and batch modification through journal tools, preserving a separate record of how controls changed over time. This can assist migration and investigation, but it does not itself set a contractual data-return deadline or promise that another vendor can directly import the resulting records.[20]

Two further behaviours matter in production. Ember can recognize new or changed instruments but does not detect instrument deletions, and changes to the securities-stream schema require a restart. Its inbound FIX gateway expects Ember’s own symbology; translation is available when connecting outward to venues. An instrument-master migration therefore needs more than a connector name match.[23]

An emergency workaround called Discard Order only marks the order cancelled in Ember; it does not communicate a cancellation to the venue. The troubleshooting guide distinguishes this from the normal Cancel action. Operators must reconcile venue state before treating the displayed status as the final financial exposure.[21]