Skip to content

Connectors overview

A connector defines how Datyo reaches an external data source. When a dataset is created, Datyo reads the source's schema and a sample of rows so the Agent can reason about the data. Connection credentials are encrypted at rest and are never displayed in plaintext after they are saved.

This section documents each supported connector in detail. Start from the table below, or read the shared concepts first — they apply to every connector and are not repeated in full on each page.

How queries run

Datyo executes analytical queries through an embedded query engine. Connectors fall into two execution models:

  • Federated — the source is mounted into the query engine and queried in place. Datasets from different federated connectors can be joined together in a single analysis. Most connectors use this model.
  • Native — the query is pushed down to the source's own engine and only the result is returned. This is used for sources whose dialect or protocol cannot be mounted directly (for example log services and certain cloud warehouses). Native datasets cannot be joined with datasets from other sources in the same query.

Each connector page states which model it uses under Capabilities.

Access is read-only

Connectors are provisioned for analysis, not modification. Queries that would write, alter, or delete data are rejected. Provide credentials with read-only privileges wherever the source supports them; this is sufficient for every feature and limits exposure.

Connector catalog

Relational databases

ConnectorModelSummary
MySQLFederatedMySQL and protocol-compatible engines
PostgreSQLFederatedPostgreSQL and protocol-compatible engines
SQL ServerFederatedMicrosoft SQL Server
SQLiteFederatedSQLite database files

Data warehouses

ConnectorModelSummary
SnowflakeFederatedSnowflake cloud warehouse
BigQueryFederatedGoogle BigQuery
MaxComputeNativeAlibaba Cloud MaxCompute (ODPS)
DuckDBFederatedDuckDB database files

Lakehouse & object storage

ConnectorModelSummary
S3 (and compatibles)FederatedS3-compatible object storage: AWS, Alibaba Cloud OSS, Tencent COS, Huawei OBS, Cloudflare R2, MinIO
Azure BlobFederatedMicrosoft Azure Blob Storage
IcebergFederatedApache Iceberg tables via a catalog
Delta LakeFederatedDelta Lake tables
DuckLakeFederatedDuckLake lakehouse format
AvroFederatedApache Avro files
SpatialFederatedGeoParquet, GeoJSON, Shapefile

Logs & NoSQL

ConnectorModelSummary
Alibaba Cloud SLSNativeSimple Log Service
MongoDBFederatedDocument database

Files

ConnectorModelSummary
File uploadFederatedCSV and Excel uploaded directly
Google SheetsFederatedOnline spreadsheets

Object-storage access credentials

Several connectors read their data from object storage — the S3 connector itself, and the lakehouse connectors (Iceberg, Delta Lake, DuckLake) whose data files live in a bucket. They share one credential block, described here once.

FieldMeaning
Cloud vendor presetSelecting a preset fills the endpoint, region, and addressing style for a known provider. Presets are available for AWS, Alibaba Cloud OSS, Tencent COS, Huawei OBS, Cloudflare R2, MinIO, and a generic "Other".
EndpointThe storage service host. Filled by the preset; adjust the region portion as needed.
RegionThe storage region identifier.
Access Key ID / Access Key SecretCredentials with read access to the target bucket. The secret is stored encrypted.
URL stylevhost (bucket in the hostname) or path (bucket in the path). Most providers use vhost; MinIO and Cloudflare R2 require path. Presets set this automatically.
Enable SSLWhether to use HTTPS. On for all public providers; MinIO deployments are frequently plain HTTP.

Grant the keys read-only access scoped to the specific bucket or prefix that holds the data.

Managed vs. self-hosted: local file paths

Four connectors — SQLite, DuckDB, Avro, and Spatial — can read a file either from object storage or from a filesystem path.

  • On the managed service (datyo.ai), reading a server-side filesystem path is not permitted. Use the object-storage mode (point the connector at a bucket) or, for a single file, the file upload flow.
  • In the desktop app and self-hosted deployments, a local filesystem path is available where it makes sense for that environment.

Each affected connector page notes this under Capabilities.

Data source types shown as "planned"

The console lists additional types greyed out as planned. Several have a supported path today:

Requested typeRecommended approach
ClickHouseEnable its PostgreSQL-protocol port and connect with the PostgreSQL connector.
TiDB, StarRocks, DorisConnect with the MySQL connector — these speak the MySQL protocol.
CockroachDBConnect with the PostgreSQL connector.
Alibaba Cloud OSSUse the S3 connector with the "Alibaba Cloud OSS" preset.
OracleReplicate to PostgreSQL (for example via CDC) and connect through PostgreSQL.

For any source not listed here, contact support to discuss options.

Datyo · The Agent-native intelligent data analytics platform