Docs/Connections

Local Files

Local file connections let you browse files, inspect metadata, query file contents with SQL, and use file-backed datasets as sources or targets in Load streams.

Supported formats

FormatSourceTargetExplorer / SQL
CSV
JSON
JSONL
Parquet

JSON files can be read as sources and queried in Explorer, but stream targets use JSONL (line-delimited JSON) instead.

When a file is the source, its column types are inferred and then mapped to the target database — see From a file source. When a file is the target, the same tables show what each database type becomes in the file.

Compressed files (.gz, .bz2) are detected and decompressed transparently. See Compressed File Support.

Connection settings

SettingRequiredDescription
NameYesConnection label
Base pathYesRoot directory for file browsing (e.g. /data/exports)

The base path determines the root of the file tree shown in Data Explorer. All file access is scoped under this directory.

Typical workflow

  1. Create a file connection with a base path.
  2. Open the connection in Data Explorer.
  3. Browse folders and files from the connection tree.
  4. Preview data or inspect file metadata.
  5. Query files with DuckDB-powered SQL in the SQL Console.
  6. Use the connection as a source or target in a Load stream.

Stream usage

Local files work as Load-mode sources and targets. They do not support CDC.

  • Source — point a stream at a file or directory with source.connections[].files. Supported formats: CSV, JSON, JSONL, Parquet. paths must be relative to basePath.
  • Target — stream output writes to the base path in the chosen format. Supported formats: CSV, JSONL, Parquet. Set fileFormat in the target configuration.

Source example:

{
  "source": {
    "connections": [
      {
        "connectionId": "conn_local_files",
        "files": {
          "basePath": "/data/imports",
          "paths": ["orders/2026-03/orders.csv"]
        }
      }
    ]
  }
}

Target example:

{
  "target": {
    "id": "conn_local_files",
    "spec": {
      "files": {
        "fileFormat": "jsonl"
      }
    }
  }
}