-   [](https://platform.exaloop.io/)
-   Get started
-   Your First Native Application

# Your First Native Application

Build a native application that writes and reads a Parquet file. This tutorial covers project creation, SDK selection, and executable builds using a small, self-contained dataset. You will need the Exaloop CLI, its matching Codon compiler, and the Data SDK; no external services are required.

**Before you begin:** install the [all-SDK distribution](https://platform.exaloop.io/start/distribution/). It includes the CLI and automatically registers the Data SDK used here.

Use an existing Codon installation

Set up [individual SDKs](https://platform.exaloop.io/start/install/) and the [CLI](https://platform.exaloop.io/start/cli/#install), then register the Data SDK once using its installed path:

```sh title="Terminal"
exaloop plugin add data --path /path/to/sdk/data
```

## 1\. Check the Toolchain

```sh title="Terminal"
exaloop doctor
exaloop plugin list
```

Confirm that the selected compiler is the one you intend to use and that `data` is available. If a check fails, resolve it with [troubleshooting](https://platform.exaloop.io/development/troubleshooting/) before continuing.

## 2\. Create and Run a Project

Choose a new directory; `init` does not overwrite existing project files:

```sh title="Terminal"
exaloop init analytics-demo
exaloop --project analytics-demo --target app run
```

The starter prints `Hello from Codon`. It contains `app.codon` and an `exaloop.toml` manifest with a named native target. `--project` selects the directory; `--target app` selects what to build and run.

## 3\. Add a Data SDK Example

Replace `analytics-demo/app.codon` with:

```python title="app.codon"
import exaloop.data as data

with data.table({"value": data.array([2.5, 4.0, 6.5])}) as table:
    table.write("values.parquet")

with data.read_parquet("values.parquet") as table:
    values = table["value"].to_numpy(float)

assert values.sum() == 13.0
print("Parquet round trip passed")
```

Update `analytics-demo/exaloop.toml` to select the Data SDK:

```toml title="exaloop.toml"
[targets.app]
kind = "native"
source = "app.codon"
plugins = ["data"]
codon_args = ["-release"]
```

Run the target again:

```sh title="Terminal"
exaloop --project analytics-demo --target app run
```

Expected output:

```text title="Output"
Parquet round trip passed
```

It also writes `analytics-demo/values.parquet`. Target commands run from the project directory, so relative file paths are resolved there. The NumPy array returned by `to_numpy` owns its data and remains usable after the table closes.

## 4\. Build an Executable

```sh title="Terminal"
exaloop --project analytics-demo --target app build
```

The executable is `analytics-demo/dist/app/app`. It does not need the CLI or a Python interpreter to run, but it still needs its Codon, Data SDK, and transitive native libraries. Before moving it to another machine, read about [runtime packages](https://platform.exaloop.io/start/distribution/#runtime-only-packages) and host requirements.

## Next Steps

-   [Tables and files](https://platform.exaloop.io/data/data/): schemas, NULLs, conversions, and ownership.
-   [CLI and projects](https://platform.exaloop.io/start/cli/#projects): multiple targets, tests, and packaging.
-   [Profiling](https://platform.exaloop.io/start/profiling/): inspect a representative application's runtime costs.
-   [PGO](https://platform.exaloop.io/opt/): improve a build using separate training and evaluation inputs.

For profiling and PGO, use a longer-running program with representative inputs so that startup time does not dominate the measurements.

Source: https://platform.exaloop.io/start/quickstart/