Databricks migration accelerators: what the free tools cover, and what still needs people

Databricks offers a set of free tools for moving a data warehouse onto its platform:

Lakebridge, for assessment, code conversion and reconciliation.
The agentic code converter, which runs in Genie Code.
Lakeflow Connect, for loading data.
Lakehouse Federation, for querying sources where they are.
UCX, for upgrading workspaces that still use the Hive metastore.

These tools measure the source estate, convert syntax, move data and check results. They leave four decisions open: the scope of the migration, the order of the waves, the acceptance criteria, and which data may be processed on a regional platform. Those decisions are made by people, and they account for most of the cost of a project.

Below, each tool is placed in the phase it serves, with its support status and the limits stated in its documentation. The article closes with five questions to ask about any accelerator.

What counts as a migration accelerator

A migration accelerator is a tool that shortens part of the work: measuring the source system, converting code, moving data, or showing that the result matches the original.

They fall into three groups, and the support terms differ between them.

Databricks product features. The agentic code converter, Lakeflow Connect and Lakehouse Federation are documented on docs.databricks.com and released through Databricks’ usual stages: Beta, Public Preview, then general availability.

Databricks Labs projects. Lakebridge and UCX are open source, published under the databrickslabs account on GitHub, and free to use. The UCX documentation states that Labs projects are provided for exploration and are not formally supported by Databricks with service-level agreements.

Accelerators from implementation firms. These are usually scripts, templates or converters that come with a services contract. They are not covered here, but the five questions at the end apply to them as well.

Phase 1, assessment: Lakebridge Profiler and Analyzer

Databricks introduced Lakebridge in June 2025 as a free migration toolset. Assessment uses two of its components.

Profiler collects structure and workload data from the source system.
Analyzer scans the exported SQL and ETL code. It lists the objects, gives each one a complexity score, and shows which jobs, tables and reports depend on one another.

An optional SQL Splitter breaks large SQL files into one file per object before assessment or conversion.

What a scan measures, and which questions it leaves for interviews, is covered in what a one-week migration scan decides.

Phase 2, code conversion: three transpilers and an agentic converter

Lakebridge comes with three transpilation plugins.

BladeBridge and Morpheus are rule-based. The same input always gives the same output, they run locally without calling external services, and they handle large batches of files.
Switch uses a language model and is marked experimental. It suits logic where intent matters more than syntax, and sources the rule-based transpilers do not cover.

Supported sources, as listed in the transpile guide on 26 October 2026:

Source	BladeBridge	Morpheus	Switch (experimental)
SQL Server, Azure SQL	Databricks SQL	Databricks SQL	Spark SQL
Azure Synapse (dedicated SQL pools)	Databricks SQL	Databricks SQL	Spark SQL
Oracle	Databricks SQL	—	Spark SQL
Teradata	Databricks SQL	—	Spark SQL
IBM Netezza	Databricks SQL	—	Spark SQL
Amazon Redshift	Databricks SQL (experimental)	—	Spark SQL
Snowflake	—	Databricks SQL	Spark SQL
MySQL, PostgreSQL	—	—	Spark SQL
IBM DataStage	Spark SQL, PySpark	—	Spark Declarative Pipelines
SSIS	Spark SQL (experimental)	—	Spark Declarative Pipelines

Switch also converts Airflow to Databricks workflows, and Python or Scala code to Python notebooks.

The agentic code converter

The agentic code converter is a product feature, separate from Lakebridge. It runs inside a Databricks workspace and uses Genie Code to convert SQL written for T-SQL, Snowflake, Redshift, Oracle, BigQuery and Teradata.

How it works, according to the documentation:

Work is organised in a migration project, which tracks the status of each file.
Output is either Databricks SQL or notebooks that use the DataFrame API. The notebooks option is in Beta.
A batch holds up to 300 files.
Teradata BTEQ scripts are not yet supported.
A project can load the JSON output of the Lakebridge Analyzer to show complexity and lineage.

Databricks announced the converter in Beta on 30 July 2026. In the example in that post, 6 of 8 T-SQL files converted successfully. The other 2 needed review, because the stored procedures they called had to be referenced by their three-part Unity Catalog name.

The documentation also recommends running the original and converted queries on the same test data and comparing the results. Code that runs is not necessarily code that returns the same answer.

Phase 3, data movement: Lakeflow Connect and Lakehouse Federation

Lakeflow Connect provides managed connectors. For databases there are two kinds:

CDC connectors capture changes continuously, using an ingestion gateway and staging storage.
Query-based connectors query the source on a schedule and need no CDC configuration.

The choice depends on what the source supports. SQL Server, for instance, offers both change tracking and CDC.

Lakehouse Federation lets Unity Catalog query data that has not been moved yet. This helps during the period when part of the workload still runs on the source system. It works in two ways:

Query federation pushes the query down to the external database over JDBC. The documentation states that the query runs both in Databricks and on the remote system.
Catalog federation reads tables directly from object storage, using Databricks compute.

What this means for organisations in Vietnam

Databricks has no region in Vietnam; the closest is Asia Pacific (Singapore). With federation, the source tables stay where they are. The Databricks side of the processing, and the results it receives, are in the region of the workspace.

Federation is therefore not grounds for saying that data stays in Vietnam. Which data flows need which procedures is a question for your compliance team and legal counsel. Two related articles:

Phase 4, reconciliation: Lakebridge Reconcile

Reconcile compares the source system with the target on Databricks. It produces four types of report:

schema: column names and data types, to check the DDL migration.
row: a hash of each row. No join key is needed, so it works on tables without a primary key.
data: row and column values matched on join columns, with mismatches shown per column.
all: data and schema together.

Supported sources are Oracle, Snowflake, SQL Server, Redshift, Teradata, BigQuery and Databricks. Two caveats from the documentation:

Teradata has no portable cryptographic hash in pure SQL. The hash-based reports need a hash UDF installed on the source.
Some BigQuery types have no exact equivalent in Databricks and may show as mismatches even when the data migrated correctly.

By default the reconcile job runs on serverless compute.

The tool produces the report. What counts as a match, and how long the two systems run in parallel, still has to be agreed. How those choices affect the estimate is covered in estimating a data warehouse migration.

Phase 5, governance: UCX for Hive metastore workspaces

When the source is a Databricks workspace on the Hive metastore, rather than another warehouse, the tool is UCX.

It assesses the workspace and runs repeatable workflows for groups and tables.
It is a Databricks Labs project, with the support terms described above.
Its documentation states that the code migration workflow is still under development.

How to use it, and the mistakes that cost time in this kind of upgrade, are covered in common Unity Catalog migration mistakes.

What accelerators do not decide

Each tool above reduces manual work within one phase. None of them settles the following:

Scope and goal. Which workloads move, which are retired, and what problem the migration is meant to solve.
Wave order. Which tables, jobs and reports move together, based on the dependencies the Analyzer finds and on business priorities.
Acceptance criteria. Exact match, match on totals, or match within a tolerance, agreed for each table before conversion begins.
Parallel run. Whether both systems run side by side for 30, 60 or 90 days, and the cost of doing so.
Data placement. Which data may be processed on a regional platform. This is a legal decision before it is a technical one.

Five questions to ask about any accelerator

These apply to free tools and to tools bundled with a services contract alike.

Which of your source systems are supported at a stable level, and which are experimental or in Beta?
Where do your source code and sample data go when the tool runs: a local machine, your own workspace, or an external service?
Who owns the converted output, and can you keep using it if you change implementation firm?
Was the automatic conversion rate measured on your own code, or on a generic sample?
Who fixes what the tool cannot convert, and how is that work estimated?

How Sancire uses these tools

The Migration Scan runs the Lakebridge Analyzer on your SQL, scripts and ETL code. Interviews then settle what the tools leave open: scope, wave order, acceptance criteria and the parallel run.

Duration: one week, from $2,500.
Output: a plan in waves, a risk register, and a fixed-price quote for each wave.
The fee is credited in full against a migration contract signed within 90 days.

The Migration itself runs in waves of 4 to 6 weeks, from $12,000 per wave, with a reconciliation record for each migrated table.

For banks and insurers, the Data Zoning & Compliance Assessment is the opening step. The decision on which data may go to a regional platform has to come before the choice of tools. See also the reference architecture for banks.

Sources, all public Databricks pages, read on 26 October 2026: Lakebridge Assessment Guide, Lakebridge Transpile Guide and Lakebridge Reconcile Guide (no date shown); Convert SQL with the agentic code converter (updated 16 October 2026); Convert proprietary code to open ANSI SQL with Genie Code, Databricks blog (30 July 2026); Introducing Lakebridge, Databricks blog (12 June 2025); Lakeflow Connect connector concepts (9 October 2026); Connect to external databases and catalogs (29 September 2026); Use the UCX utilities to upgrade your workspace to Unity Catalog (6 October 2026); Supported regions (21 October 2026). Lakebridge and the agentic code converter change quickly; check supported sources and release status in the current documentation before planning.