Skip to content
Products/Build layer/ZITRANX AI
Build layerLegacy modernisation

ZITRANX AI

Understand the system nobody left documentation for, then move it without breaking it.

Most modernisation programmes fail at the first step: nobody can say with confidence what the current system actually does. ZITRANX reads the estate — COBOL, PL/SQL, JCL, stored procedures, the monolith everyone is afraid of — and produces behaviour documentation, a dependency map and a characterisation test suite before a single line is rewritten.

COBOLPL/IJCLPL/SQLT-SQLNatural / ADABASRPGVB6Java monolithsDelphi
Estate scanPARSING
2.1M
Lines under analysis
1,847
Programs mapped
312
Dead paths identified
94%
Behaviour coverage
Copybook FIN-ACCT-REC referenced by 214 programs
canonical record identified · shared across 4 subsystems
Undocumented business rule extracted
interest accrual differs for accounts opened pre-1998
Cyclic dependency between BILLRUN and ADJPOST
flagged for architectural review before translation
Characterisation suite generated
3,140 cases from 90 days of production traffic
2 weeks
From code drop to dependency map and verdict
90%+
Behaviour coverage before any translation begins
Zero
Cutovers without a passing equivalence run
Read-only
Assessment phase never touches your systems
How it works

Understand first. Translate second. Prove it throughout.

Every stage produces an artefact you can review independently. You can stop after any of them and still be better off than when you started.

Stage 01
Read the estate

Static analysis across source, copybooks, JCL, schedules and database schemas builds a complete call graph and data-lineage map — including the paths that have not executed in a decade.

Dependency map and dead-code report
Stage 02
Recover the behaviour

Agents extract the business rules the code actually implements and write them in prose your domain experts can dispute. Every rule cites the lines it came from.

Reviewed behaviour specification
Stage 03
Build the safety net

Production traffic is sampled into a characterisation test suite that pins current behaviour, including the bugs downstream systems now depend on.

Executable equivalence suite
Stage 04
Translate and prove

Target services are generated subsystem by subsystem, run in parallel against live traffic, and only promoted when the equivalence suite passes clean.

Tested services, cut over in slices
Why these programmes fail

The rewrite is never the hard part.

What goes wrongWhat ZITRANX does about it
Nobody knows what it doesThe people who wrote it retired and the documentation describes a version from three mergers ago. ZITRANX recovers behaviour from the code itself and makes your domain experts confirm or correct each rule.
The bugs are load-bearingDownstream systems have quietly adapted to defects for twenty years. A characterisation suite built from real traffic pins current behaviour — bugs included — so you decide deliberately which ones to keep.
Big bang cutoversProgrammes that go dark for two years get cancelled in month eighteen. ZITRANX works subsystem by subsystem, with each slice provably equivalent and independently releasable.
Generated code nobody ownsOutput is idiomatic in the target language with the recovered business rules as its documentation and the equivalence suite as its tests. Your engineers can maintain it without ZITRANX in the loop.
Unbounded model exposureThe estate is the most sensitive code you own. ZITRANX runs inside your boundary, and every agent step is evaluated by ZNYX against your policy before it touches source.
Equivalence, not resemblance

Nothing cuts over until the behaviour matches.

Translated code is run in parallel against production traffic, and every divergence is a ticket. The characterisation suite generated during the understanding phase is the contract — not a developer's reading of what the COBOL was probably for.

How ZNYX governs the agents
Parallel runLegacy and translated services process the same production traffic; divergences are logged with the input that caused them.
Rule provenanceEvery recovered business rule cites the source lines it was derived from and the expert who signed it off.
DeploymentRuns in your cloud, on-premise or air-gapped. Source never leaves your perimeter and is never used to train shared models.
Model choiceAny model through GuideLite routing, including self-hosted open weights where source sensitivity requires it.
Audit trailEvery translation decision, override and promotion is recorded against the subsystem for programme governance.
Questions

What architects ask first.

Most estates start with a two-week assessment on a single subsystem. You get the dependency map and the honest verdict on feasibility whether or not you continue.

Book an estate assessment
Do you need our full source to start?

No. A single subsystem with its copybooks and job control is enough for the two-week assessment. Analysis is read-only and can run inside your environment.

What target stacks do you translate to?

Most commonly Java or C# services, and Python where the workload suits it. The target is a decision made after the assessment, not before — sometimes the honest answer is to leave a subsystem where it is.

Is the output maintainable, or generated soup?

It is idiomatic code structured around the recovered domain rules, with the characterisation suite as its regression tests. If your engineers cannot read it, we have not finished.

What if translation is the wrong answer?

We will say so. Some subsystems should be encapsulated behind an API and left alone, and some should be retired rather than moved. The assessment tells you which.

How does this relate to ZNYX and GuideLite?

ZITRANX runs as a set of GuideLite agents, and every step those agents take is evaluated by ZNYX against your policy. The governance is the same one the rest of the suite inherits.

Point us at the subsystem nobody wants to touch.

Two weeks, one subsystem, read-only. You get the dependency map, the behaviour documentation and an honest verdict on whether translation is the right answer at all.

Book an estate assessment