# Load assets

> Copy data from a database, bucket, or file into your pipeline without writing code.

A Load asset replicates data between two connections — from a source database, storage bucket, or file into the warehouse your pipeline runs on. It’s how raw data usually enters a pipeline: no code, just a source and a target.

![A Load asset with inline source connection and file controls, and General properties for its target and materialization](https://getrenart.com/docs-media/load-asset.webp)

## Create one

On the canvas, add an asset and choose **Load**. The main pane presents inline source controls; target settings live in the properties panel:

- **Source connection** and **source table or object** — where the data comes from. The connection picker is grouped into databases, storage, and files; once a connection is chosen, Renart browses available tables or bounded S3-compatible/GCS objects while still allowing an explicit path.
- **General → Connection** selects the destination. For a database target, the asset name is the target relation, such as `raw.exchange_rates`. Storage and file targets use an explicit destination path.
- **General → Materialization** controls how each run writes.

Local files are a first-class source: choose the **local** entry and point the source at a file path (for example `data/exchange_rates.csv`) to load a CSV that lives in your repository.

## Start from the Data Browser

S3-compatible and SFTP connections can be browsed by prefix or folder in **Data Browser**. Configure an S3 bucket first; an optional connection path acts as its browsing prefix. SFTP lists folders from the server root using the existing Sling connection, without Ingestr or a separate FTP integration.

Drag a file or prefix onto the canvas, or choose **Use … in canvas** on mobile or with the keyboard. The top target opens a Load source draft. Drop a writable prefix or connection after a database asset to create a Load destination draft. Review the connection, object path and write mode before saving. Cancelling does not create an asset, and saving does not transfer any data.

Storage browsing lists metadata only, up to 500 entries per folder; it does not offer SQL previews for objects. Paths containing wildcard or selector characters are not shown because Sling would treat them as patterns. A source prefix prefills a wildcard; a destination prefix suggests a Parquet filename. Review both explicitly, especially when a destination already exists.

## Write behavior

A **Read-only** connection can be the source, but not the destination. From the Data Browser, add an existing table as a **Source** asset, then add a Load asset downstream and select a writable connection in **General → Connection**. The Source stays externally managed; the Load asset owns the copied target.

Load copies data without deleting it from the source. **Full refresh** changes the destination only. Add downstream SQL assets on that destination connection to transform the copy.

| Choice               | Each run…                                                                                                                            |
| -------------------- | ------------------------------------------------------------------------------------------------------------------------------------ |
| **Table (replace)**  | Loads into staging and replaces the target table. This is the atomic, schema-refreshing default where the destination supports it.   |
| **Table (truncate)** | Keeps the existing table, empties it, then loads all source rows. Use this when views, grants, or other objects depend on the table. |
| **Append rows**      | Adds source rows without removing existing rows. An optional update key can limit the source read.                                   |
| **Merge by key**     | Updates or inserts rows using the declared primary-key columns. An optional update key can limit the source read.                    |

On PostgreSQL, replacing a table is blocked when an existing view depends on it. Choose **Table (truncate)** for that shape. It preserves the relation and its dependents, but it is not an atomic swap and keeps the table’s existing schema. If the schema needs to change, update the dependent objects before using **Table (replace)** again.

## Tune a large transfer

For a large transfer, open the Load asset and set **Load parallelism** to allow more transfer workers. Start with a small increase and compare run duration and resource use. Higher values consume more memory and connections; destinations may limit the effective parallelism. This does not split a source SQL query or change how many pipeline assets run at once. Leave it empty to keep the automatic behavior.

## Columns and lineage

The upstream dependency is inferred from the source when you save, so the Load asset takes its place on the canvas automatically. Use **Sync schema** to read the source’s schema into the asset’s column list — after that, downstream SQL assets get completions and type checking against the loaded table like any other upstream.

## Related

- [HTTP API assets](https://getrenart.com/docs/asset-types/http-api-assets.md) — when the source is a JSON API rather than a database or file.
- [Managing connections](https://getrenart.com/docs/connections-environments/managing-connections.md) — defining the source and destination connections.