Deterministic modernization platform that converts legacy stored procedures, ETL, and data orchestration logic into modern, production-ready pipelines while keeping your data and code safely inside your own environment.

Is Novuz self-service?

Yes. Novuz is software your engineers install and run inside your environment. Discovery, conversion and reconciliation run on their own. The resolver handles edge cases with engineer approval, and Novuz engineers back every engagement.

The conversion is not; it is rule-based and gives the same output every run, with the small share rules cannot resolve approved by an engineer. What is agentic is the access: agents can drive the platform through its APIs and operate the delivered pipelines through MCP.

Can our agents operate the converted pipelines?

The pipelines Novuz delivers can be exposed as MCP tools your agents call, under your own access, approvals and audit logging; the platform itself is available through APIs for your orchestrator.

Novuz is a rule-based conversion engine. A semantic parser, an inference layer and a versioned rules library convert about 95% of objects with no model calls. The resolver uses a language model only to propose fixes for the edge cases, and an engineer approves each one.

What does the reconciliation report contain?

One line per object with a pass, fail or review status, plus schema comparison, row counts and column-level value matches between the original and converted object on the same inputs. It exports as audit evidence.

Do we need to change our code before conversion?

No. Novuz reads your code as it stands. Dead code and unused objects are flagged in Discovery so you can drop them from scope.

Can we run the old and new systems side by side?

Yes. Novuz supports a parallel run before cutover, so you can compare outputs on live data before you switch.

What happens if Novuz goes away?

The converted code is native to your target, with no Novuz runtime in it, so your pipelines keep running whether Novuz does or not.

Product

Five stages from legacy code to proven pipelines

Data migration is solved. Logic migration is not. Novuz is software, not a services team: your engineers run the same five stages on every estate, inside your environment, and each stage produces an output they review before anything moves forward. Agents can run Novuz, and agents can run what Novuz builds.

Take the logic debt scorecard

Self-service platform

Your team runs it. Not a consulting project.

Your engineers run Novuz and set the pace, the scope and the sign-off. The resolver handles the edge cases rules cannot settle, and an engineer approves every proposal. Novuz engineers stand behind every engagement, so your team is never on its own.

Step 1

Install in your environment

Pull the Novuz container or VM into your cloud account or data center. No outbound connection is required.

Step 2

Point it at your code

Connect your DDL, procedures and ETL. Discovery and conversion run on their own.

Step 3

Review the evidence

The resolver proposes fixes for edge cases. Your engineers approve every proposal and sign off on each reconciliation report.

Step 4

Deploy when ready

Ship validated pipelines to your target platform on your schedule.

Discovery runs first: every object counted, dead code flagged before any conversion starts.

Stage 1

Discover: know exactly what you have

Novuz parses every schema, procedure, package, script and schedule in scope. It builds a model of how objects depend on each other and where data flows. Each object gets a complexity band. The output is a Discovery report: a workbook with a warehouse summary, an ETL summary and a line per object. This is the work teams usually do by hand in spreadsheets. Here it is a by-product of the conversion.

Full object inventory by type and source
Lineage and dependency map, upstream and downstream
Complexity band per object, used for scope and price
Dead code and unused objects flagged
Discovery report you can share with your platform team and your auditor

A sample Discovery report: object counts by type with complexity bands and dead-code flags, and the lineage of one object, upstream and downstream. Sample report; illustrative estate with invented object names, not customer data.

Stage 2

Converge: deterministic conversion

The engine converts each object using fixed, versioned rules, selected by inference over the object’s patterns, with a confidence score at every decision. It maps legacy SQL dialects, control flow and package logic to code that runs natively on your target. The same input gives the same output on every run. There are no prompts, no sampling and no model cost at this step. About 95% of objects finish here.

Repeatable output you can diff between runs
Readable target code that follows your naming standards
No external model calls during conversion
Confidence score on every decision

Stage 3

Resolve: edge cases, handled in context

Some objects use patterns that rules cannot settle alone. Novuz sends these to a resolver. The resolver draws on a knowledge base built from your own estate and kept inside your environment. It proposes a fix at 700 to 1,000 tokens per edge case. An engineer reviews and approves every proposal. Nothing deploys unapproved.

Runs on a model you approve: Claude or Gemini under your own agreement, or a local model through Ollama with no external calls. If you choose a hosted model, only the edge-case fragments go to it
Every resolved object has a human sign-off
Knowledge base never shared across customers

A sample approval queue: each edge case with its original and proposed code side by side, awaiting engineer approval. Sample screen; illustrative estate with invented object names, not customer data.

Stage 4

Validate: a reconciliation report for every object

Every converted object is checked against the original. Novuz compares schemas, row counts and column-level values on the same input data. Each object gets a reconciliation report showing what matched and what did not. These reports are what you show your auditor and your business owners.

Schema, row count and value-level checks
Pass, fail or review status per object
Exportable evidence for audit and sign-off

A sample reconciliation report: schema, row counts, aggregates and sampled rows for one object, each with a pass status, above an estate view listing every object with a pass, review or fail status. Sample report; illustrative estate with invented object names, not customer data.

Stage 5

Deploy: production pipelines inside your perimeter

Validated code is deployed to your target with orchestration rebuilt to match the original schedule and dependencies, and the running system is replicated there. Code assistants hand back files; Novuz deploys and replicates. Novuz runs as a container or VM in your cloud account or data center, and no outbound connection is required.

Orchestration rebuilt with the original dependencies
Converted code is native to your target, with no Novuz runtime in it
Docker or VM deployment in your environment
Supports a parallel run before cutover
The delivered pipelines can be exposed as MCP tools your agents call

Coverage

What Novuz converts, by source

Sources: SQL Server (T-SQL), SSIS, Teradata including BTEQ, Hadoop (Hive, Impala) and Snowflake. Targets: Databricks, Microsoft Fabric, Snowflake, Apache Iceberg and Delta Lake. Not today: Oracle, mainframe, Informatica, DataStage, Netezza and Redshift. Coverage is confirmed per estate during the assessment.

Object type	SQL Server	SSIS	Teradata	Hadoop (Hive, Impala)	Snowflake
Schemas and tables	Yes	n/a	Yes	Yes	Yes
Views	Yes	n/a	Yes	Yes	Yes
Functions	Yes	n/a	Yes	Yes	Yes
Stored procedures / macros	Yes	n/a	Yes	n/a	Yes
Scripts (BTEQ, HiveQL)	n/a	n/a	Yes	Yes	n/a
ETL packages and data flows	n/a	Yes	n/a	n/a	n/a
Orchestration and schedules	Yes	Yes	Yes	Yes	Yes
Data migration	Yes	n/a	Yes	Yes	Yes

Walkthrough

See a conversion from start to finish

A short walkthrough of one estate going through all five stages, from inventory to reconciliation report.

Why deterministic

Why deterministic conversion

There are three common ways to convert legacy logic. They differ in what you can prove afterwards.

	Manual rewrite	Probabilistic (LLM-only) conversion	Deterministic conversion (Novuz)
Same input, same output	No	No	Yes
Elapsed time	Months to years	Quick to draft, slow to prove	Typically 4 to 8 weeks on 1,000 to 1,600 objects
Explains each change	Depends on the engineer	Rarely	Yes, rule by rule
Evidence per object	Manual testing	Manual testing	Reconciliation report
Cost driver	Engineer hours	Open-ended tokens and rework	Fixed price per scope, set before work starts
Code leaves your environment	Depends on vendor	Often	No
Deploys and replicates the running system	Depends on the team	No; hands back files	Yes

Novuz converts by fixed, versioned rules. A language model is used only at the edge, by the resolver, with an engineer approving every proposal, because auditors do not accept probabilities.

Agentically primed

Agents can run Novuz. Agents can run what Novuz builds.

Both the Novuz platform and the pipelines it delivers can be run by agents: the platform through its APIs, the pipelines through MCP. The conversion itself stays deterministic.

The platform is headless

Everything Novuz does through its UI it also does through APIs. Your orchestrator, your CI/CD pipeline, your SI’s tooling or an AI agent can trigger discovery, conversion, validation and deployment, inspect results, and gate the next step on a human approval.

The pipelines are MCP-ready

Through the Model Context Protocol, each converted pipeline can expose run, check status, retrieve lineage, inspect outputs and rerun a failed step as tools your agents discover and call, under your access, approvals and audit logging, on Databricks, Snowflake or Microsoft Fabric.

What stays the same: the conversion is deterministic. Agents decide what to ask for and when; the platform gives the same output for the same input, so every result can be tested, audited and repeated.

How agentically primed works

	Headless platform	MCP-ready pipelines
Applies to	The Novuz modernization engine	The pipelines Novuz delivers
Used by	Delivery teams, partners, orchestrators, agents	Operations and data agents, downstream applications
Purpose	Run modernization programmatically and continuously	Run and manage the modernized estate through agents
Outcome	Faster, repeatable migration	Pipelines that are agent-ready from day one

An agent gets only what you expose, under your own access, approvals and audit logging.

Honest fit

Who should not choose Novuz

Novuz is not the right choice for estates under about 300 objects; for Oracle-only or mainframe estates; for teams with no target and no date inside 18 months; for buyers who want staff augmentation rather than software; or for organizations that cannot run software inside their environment or share code for an assessment.

FAQ

Frequently asked questions

How we measure: Methodology

Last reviewed: October 29, 2026

Find out what your logic debt is costing you.

Every object counted, lineage, dead code and a fixed price in three business days. $3,000, credited.