Docs / Snowflake, BigQuery, Databricks, Trino and other ADBC drivers

Snowflake, BigQuery, Databricks, Trino and other ADBC drivers

These databases are reached through ADBC (Arrow Database Connectivity) drivers, libraries that the app loads when you connect. Snowflake, BigQuery, Databricks and Trino have a tile each; any other database with an ADBC driver connects through Other ADBC driver.

First use: the driver download

The four named databases need their driver on this machine. When you choose one for a new connection and the driver is not installed, a popup opens over the form: Snowflake needs a driver (and so on). It says it is an ADBC driver, downloaded once for all connections of that database, and shows the package, its version, its license and where it comes from: the dbc driver registry (Columnar). Download fetches it, checks its SHA-256 (a package without one is refused) and keeps it with the app's data; Cancel leaves the form usable, and the offer returns at Test connection and at the first opening.

The offer reads only what is on this machine, so it appears offline too; the download needs the network. Settings › Drivers lists the installed drivers and removes them, checks for newer versions once a day (or on Check for updates), installs or goes back to another version with Versions…, and can install updates by itself. After a change, the saved connections that use the driver are tried (connect, version, schemas, SELECT 1), with Go back if one fails. The other-driver choice downloads nothing: you point it at a library file.

Connect

Each database has its own fields:

Database Fields

Snowflake Account (for example xy12345.eu-west-1), user, password, warehouse, role, database, schema

BigQuery Project, dataset, and a service account key file (JSON); empty means Google's Application Default Credentials. No password

Databricks Server hostname, HTTP path, catalog, schema; the password field is the access token

Trino Host, port (8080 by default), user, catalog, schema, password (optional), and TLS

Other ADBC driver the driver library file, an optional entrypoint symbol, a URI, user and password; every other option you add is passed to the driver

Database	Fields
Snowflake	Account (for example `xy12345.eu-west-1`), user, password, warehouse, role, database, schema
BigQuery	Project, dataset, and a service account key file (JSON); empty means Google's Application Default Credentials. No password
Databricks	Server hostname, HTTP path, catalog, schema; the password field is the access token
Trino	Host, port (8080 by default), user, catalog, schema, password (optional), and TLS
Other ADBC driver	the driver library file, an optional entrypoint symbol, a URI, user and password; every other option you add is passed to the driver

Trino's TLS choices are Automatic (HTTPS on port 443, HTTP otherwise), Off (HTTP), On, verify certificate and On, unverified.

Other ADBC driver: the library is the driver file you downloaded or built. The entrypoint is needed only for a library that does not export AdbcDriverInit (for DuckDB's ADBC library it is duckdb_adbc_init). The URI and the options are whatever that driver documents. A missing file, or one that is not a driver, is reported when you connect.

Privileges

Whatever the account or token has on the service. The navigator reads the catalogs, schemas, tables and columns with ADBC's metadata call.

What works

Queries, results, the navigator and autocompletion. The result streams as Arrow batches.
Cancel: the app asks the driver to cancel; Go-based drivers only cancel their own calls, so a statement that does not stop is given up with its session, and the app opens a new one.
Object DDL on Trino only (SHOW CREATE). Snowflake, BigQuery, Databricks and other drivers show no DDL.
Foreign keys that point at a table are not listed (it would mean reading every table's keys).
Exports, and imports of files (typed imports are tested on Trino).

What does not

No manual commit: every statement commits by itself, and turning autocommit off says that manual commit with an ADBC driver is not supported.
No explain, and no administration tabs.
No data editing in the grid: a save checks each change in a transaction, and these connections have none. Results stay read-only; use SQL for changes.
Terminal mode is not available.

Agents

run_query, sample_rows and the other tools work, within the checks. With no server-side read-only session, read-only agents rely on the statement check alone. Statements that act outside it are asked about before they run: Trino's query() and Snowflake's SYSTEM$ functions. See Agents and the CLI.

Known limits

A cloud warehouse sign-in other than the fields above (key pair, single sign-on) is not in the form; the generic driver takes the driver's own options.
Snowflake, BigQuery and Databricks are in the code and share the driver layer with Trino, but have no test against the real services.

Tested on

The conformance suite runs Trino 484 through its downloaded driver (types, long results, cancel, metadata, errors, statements that end with a terminator, typed imports, and rows and UUIDs) and the generic ADBC driver on the Apache SQLite ADBC driver and on DuckDB's. Snowflake, BigQuery and Databricks are not tested.