Find a driver for a database
Use the dbc command line tool to find available drivers for a database. It's possible no driver may exist for the user's database of choice as not all databases have ADBC drivers.
Install dbc
If the user does not have dbc available, try to install it for them.
Prefer installing it with these commands, in order of preference, if the tool is available:
- If
uvis available:uv tool install dbc - If
pipxis available:pipx install dbc - If
brewis available:brew install columnar-tech/tap/dbc - On Windows, if
wingetis available:winget install dbc - Otherwise, direct the user to the installation docs: https://docs.columnar.tech/dbc/getting_started/installation/
Search for a driver
shdbc search
Install a driver
Install a driver by running dbc install <DRIVER>. This is idempotent — if the driver is already installed, it is a no-op that simply reports the existing installation — so there is no need to check first before running it.
Prefer to install drivers using dbc over installing driver packages from PyPI or Conda Forge.
For anything beyond a one-shot dbc install <DRIVER> — reproducible dbc.toml / dbc sync workflows, version pinning, or any other dbc subcommand — invoke the dbc skill rather than guessing at commands. In particular, note that there is no dbc list command.
Referring to drivers
Refer to drivers by their dbc short name and avoid specifying drivers with absolute paths.
Example: After running dbc install sqlite, refer to that driver as sqlite. For example:
pythonfrom adbc_driver_manager import dbapi dbapi.connect(driver="sqlite", db_kwargs={"uri": "foo.db"})
Don't use dbc info to find where drivers are installed and find their absolute paths on disk.
Programming Language
See the resources below depending on which language or languages the user wants to use:
- C++:
resources/languages/cpp.md - Go:
resources/languages/go.md - JavaScript:
resources/languages/javascript.md - Python:
resources/languages/python.md - R:
resources/languages/r.md - Rust:
resources/languages/rust.md
Prefer any language the user has already said they're using or can use. Each of these examples use the "sqlite" driver, connect to an in-memory database, and load a "penguins.parquet" file. This is done just for explanatory purposes and the code should be adapted to the user's problem.
Using a Driver
See the resources below depending on which database the user wants to use:
- DuckDB:
resources/drivers/duckdb.md - FlightSQL:
resources/drivers/flightsql.md - MySQL:
resources/drivers/mysql.md - PostgreSQL:
resources/drivers/postgresql.md - Snowflake:
resources/drivers/snowflake.md - SQLite:
resources/drivers/sqlite.md
Connection Profiles
Connection profiles are TOML files that store a driver name and connection options, referenced by a profile://profile_name URI. They decouple credentials and environment-specific config from application code.
Suggest connection profiles when the user:
- Has multiple environments (dev/staging/prod) to switch between
- Wants to keep credentials out of source code
- Wants a shared, reusable connection config
All language bindings that wrap the C++ or Rust driver manager support connection profiles, including Python, Go, R, Java, GLib/Ruby, C++, and Rust. JavaScript (@apache-arrow/adbc-driver-manager) also supports connection profiles via the profileSearchPaths option.
See resources/connection-profiles.md for the TOML format, file locations, and environment variable substitution syntax. See the relevant language resource for the binding-specific API.
More Resources
- Official Apache Arrow ADBC documentation: https://arrow.apache.org/adbc/current/index.html
- ADBC Quickstarts. Minimal but well-documented examples for each language and many databases: https://github.com/columnar-tech/adbc-quickstarts
- Documentation for many dbc-installable drivers: https://docs.adbc-drivers.org
