Credentials

Credentials are the connections Data Conductor uses to reach your databases and storage. Everything else — data steps, pipelines, ingestion — runs through one of them.

Supported database types

Adding a connection

Credentials → Add Integration, then pick the type. The form changes to suit it: a Postgres connection asks for host, port, database, user and password; BigQuery asks for a project and a service-account key; Snowflake adds warehouse and role.

Give the connection a name your team will recognise — it's what appears in the Source dropdown when someone writes a data step.

What the setup step asks for

Saving a connection proves we can reach your database. It does not yet say which compute to run on, or which of your schemas the Analyst should know about — Set up on the connector card asks both, and what it asks depends on the engine.

Databricks asking which SQL warehouse to use

Snowflake asks for a warehouse, Databricks for a SQL warehouse. Databricks also shows an id next to the name, because there the name is a label and the id is what builds the connection path; on Snowflake the name is the identifier, so there is nothing to show beside it.

BigQuery going straight to datasets

BigQuery has no compute you choose per connection, so that step disappears and the wizard has two steps instead of three. The level names follow your engine as well: databases and schemas on Snowflake, projects and datasets on BigQuery, catalogs and schemas on Databricks.

Indexing scope is not a permission boundary

What you tick decides what we index and what the Analyst knows about. It does not restrict what a query can reach — the database role you connected with decides that.

Quick Setup walks through the whole step.

Getting your credentials

Every connector asks for values that live in your provider's console, not in Data Conductor. This is where to find each one.

Two habits worth keeping regardless of provider: create a dedicated user or key for Data Conductor rather than reusing a personal one, and grant it only the databases and buckets it actually needs. A credential scoped narrowly is one you can rotate without wondering what else breaks.

Databases and warehouses

Connector What you need Where to get it
PostgreSQL — Host / Port host, port, database, username, password CloudSQL · RDS / Aurora · Azure · Neon
PostgreSQL (connection string) a postgresql:// URI Same providers — most show a ready-made connection string
MySQL host, port, database, username, password RDS for MySQL
MySQL — Connection String a mysql:// URI Same
Microsoft SQL Server host, port, database, username, password Azure SQL connection details
BigQuery service-account JSON key, dataset, location Create a service-account key
Snowflake account identifier, username, private key, warehouse, role Key-pair authentication · Finding your account identifier
Databricks workspace URL, HTTP path, personal access token Connection details · Personal access tokens

Object storage

Connector What you need Where to get it
Amazon S3 access key ID, secret, region Managing AWS access keys
Google Cloud Storage HMAC key ID and secret Managing HMAC keys
Azure Blob Storage storage-account connection string Configure a connection string
Cloudflare R2 account ID, R2 access key ID, secret R2 API tokens
S3-compatible (MinIO, Backblaze B2, Wasabi) endpoint URL, access key ID, secret, region Your provider's own S3-compatibility page

Google Cloud Storage needs an HMAC key, not a service account

This is the one that catches people out. Storage access goes through DuckDB, which authenticates to GCS with an HMAC key — a pair that looks like AWS credentials. A service-account JSON key, the thing you would use for BigQuery, will not work here. HMAC keys are created under Cloud Storage → Settings → Interoperability.

AI providers — coming soon

Not available yet

These connectors exist for agent steps: a planned step type that calls a model as part of a pipeline run, the same way a SQL step runs a query. The feature is not switched on yet, so a key added here does nothing today.

This is not the Analyst. The Analyst (mAIstro) is available now, runs on Data Conductor's own model configuration, and needs no key from you — these keys would be for models your pipelines call.

Provider What you need Where to get it
Claude (Anthropic) API key Anthropic get started
Google Gemini API key Using Gemini API keys
OpenAI API key OpenAI quickstart

Supported databases

Family Notes
PostgreSQL Includes CloudSQL, Aurora/RDS, Azure Database, Neon, Yugabyte, CockroachDB, Redshift
MySQL / MariaDB Includes cloud-managed MySQL variants
Microsoft SQL Server Cloud-hosted
BigQuery Service-account authentication
Snowflake Warehouse and role required
Databricks PAT or M2M authentication
DuckDB over cloud storage S3, GCS, Azure Blob, R2, and S3-compatible endpoints

Testing

Every connection has a Test action. Run it before saving — a connection that can't be reached produces confusing failures later, at execution time, where the cause is much harder to see.

Access

Connections belong to the organization, not to the person who created one, so anyone with access to your organization can use a connection in a data step or pipeline.

If your databases restrict access by IP, see Organization Settings for the addresses to allowlist.