> ## Documentation Index
> Fetch the complete documentation index at: https://docs.efference.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Install the Efference SDK on NVIDIA Jetson

> Install the Efference SDK on an NVIDIA Jetson running JetPack

The Linux SDK builds directly on an NVIDIA Jetson running JetPack.

## Before you begin

* Install and boot NVIDIA JetPack on the Jetson
* Confirm the Jetson has internet access for apt packages
* Connect the M1 with a USB-C data cable, preferably to a direct USB 3.x port
* Ensure your user can run `sudo`

## Install

<Steps>
  <Step title="Clone the public repository">
    ```sh theme={null}
    git clone https://github.com/EfferenceAI/efference.git
    cd efference
    ```
  </Step>

  <Step title="Build on the Jetson">
    ```sh theme={null}
    ./build.sh --deps
    ```

    `--deps` installs the arm64 packages and M1 USB udev rule, then builds in
    `sdk/linux/build/`. Use `./build.sh` for later rebuilds.
  </Step>

  <Step title="Re-plug the M1">
    Disconnect and reconnect the M1 after the first build so the new udev
    permissions apply.
  </Step>

  <Step title="Verify the connection">
    ```sh theme={null}
    ./sdk/linux/build/ef-cli list
    ./sdk/linux/build/ef-cli info
    ```

    The M1 should appear in the device list, and `ef-cli info` should print its
    serial, firmware, capture modes, and sensor configuration.
  </Step>
</Steps>

## Jetson considerations

* Prefer a direct SuperSpeed port rather than a hub.
* Use an appropriate Jetson power mode for sustained decode or perception
  workloads.
* MCAP recordings can grow quickly; write them to storage sized for the
  expected session duration.
* The Efference SDK does not require CUDA; downstream applications may use it.

## Next step

Continue with [Connection & Lifecycle](/device/connection), or use the NVIDIA
Jetson as the Linux host in [Working with Wireless](/get-started/quickstart-wireless).