Your data stays put.
Read existing S3-backed Delta tables directly. No export pipeline or second copy to maintain for your evaluation.
YOUR DATA PLATFORM. ANOTHER COMPUTE OPTION.
Keep Unity Catalog.
Own your
SQL compute.
An open-source SQL engine for your existing Delta tables. Keep your Databricks identity. Run supported queries on infrastructure you control.
Self-hosted. MIT licensed. No data migration.
YOUR STACK, CONNECTED01 / 03
Your SQL clientsPython connector · JDBC
SQL over HTTP
HarborSQLSQL compute you control
Identity & table access
▤Unity CatalogYour existing catalog
Direct data reads
△Delta LakeYour tables on S3
BUILT ON OPEN FOUNDATIONSApache DataFusion+Delta Lake+Unity Catalog access
01 / KEEP WHAT WORKS
Change the compute.
Not your entire platform.
Your tables and catalog already exist. HarborSQL gives you another place to execute supported SQL, alongside your Databricks SQL Warehouse.
Your identity comes with you.
HarborSQL uses the caller’s Databricks token to request table access and temporary credentials from Unity Catalog.
Your compute is your choice.
Run a self-hosted DataFusion engine. Choose the infrastructure and move only the workloads that pass your checks.
02 / FIND YOUR FIRST WORKLOAD
Start small. Prove the fit.
Evaluate one query before moving a workload. Keep your existing warehouse available for everything else.
Application queries
Try supported SELECT queries from your Python or JDBC application against the same Delta tables.
Connect a client →Internal analytics
Compare query results and column metadata with SQL Warehouse before switching an internal workload.
Run the comparison →Your existing dbt projects
Write-query support is underway, targeting dbt-databricks without model or adapter changes. Compatibility is not yet verified.
Follow development ↗03 / EVIDENCE BEFORE PROMISES
Don’t take our word for it.
Run your own comparison.
Open code, reproducible checks, and explicit boundaries. HarborSQL is pre-1.0 software, not a complete Databricks replacement.
See where HarborSQL fits →
TRY IT ON YOUR DATA
Compare with your SQL Warehouse. ↗
Run the same query through both engines on an existing Delta table. Compare the results before moving a workload.
COMPATIBILITYTested clients. Documented limits. ↗
Check driver versions, authentication, and required settings before evaluating your application.
REPRODUCIBLE BENCHMARKSInspect the workload behind the numbers. ↗
Review hardware, methodology, datasets, and limitations. Measure your workload on your infrastructure.
FROM READING TO RUNNING
Bring a table.
Try a query.
Start a local engine, connect your client, and compare a read against your existing SQL Warehouse.
You’ll need a Databricks workspace, an S3-backed Delta table, external data access enabled, and the required Unity Catalog grants. Replace the image tag and workspace host below.
Full setup and prerequisites →export TAG="<version>"
docker run --rm \
-p 127.0.0.1:1992:1992 \
-e HARBORSQL_BIND_ADDR="0.0.0.0:1992" \
-e HARBORSQL_DATABRICKS_HOST="https://<workspace-host>" \
ghcr.io/harborsql/harborsql:$TAG
Local endpoint only. Use HTTPS for production.
A FEW IMPORTANT DETAILS
Know the boundaries.
Does HarborSQL replace Databricks?
No. It provides alternative compute for supported SQL workloads. Your Databricks workspace and Unity Catalog remain in place. It does not replace the full platform or managed warehouse operations.
Does my data move?
No data copy is needed for supported reads. HarborSQL reads your existing Delta tables on S3 using temporary table credentials requested through Unity Catalog.
Which clients and SQL features are supported?
HarborSQL supports a focused Databricks SQL connector and JDBC protocol surface. It is not blanket compatibility with all SQL, ODBC, or BI tools. Read the SQL compatibility notes and test your client.
Can I run dbt today?
Not as a verified supported workflow. Write support is in development, targeting existing dbt-databricks projects without model or adapter changes. That target is not a compatibility guarantee.
Does it enforce every Unity Catalog policy?
Table access uses your Databricks identity and Unity Catalog credentials. That does not establish full support for row filters, column masks, views, or every governance feature. Review your policy requirements before moving a workload.
What do I need to operate?
You operate the HarborSQL service and its compute, networking, TLS, monitoring, and upgrades. The software is MIT licensed; infrastructure and operational costs still apply.
YOUR NEXT QUERY, YOUR COMPUTE.