Back

Local NVIDIA AI Acceleration

Using NVIDIA GPUs should be simple. We make it easy, from local development to production autoscaling in the cloud.

For local work, we use Kind to run Kubernetes in Docker. This gives you a real production environment on your machine. We handle the hard parts of connecting Kubernetes to your GPU. You build. We make the hardware work.

TLDR

If you have nvidia hardware and have installed the NVIDIA Container Toolkit, then Batteries Included will automatically detect and use your GPU. Install the cli and start a local cluster with one command.

bash -c "$(curl -fsSL https://home.batteriesincl.com/api/v1/scripts/start_local)"

If you don’t have NVIDIA container toolkit installed, follow these steps:

bash -c "$(curl -fsSL https://home.batteriesincl.com/api/v1/scripts/install_bi)
sudo bash -c "$(~/.local/bin/bi gpu setup-command)"
~/.local/bin/bi start-local

Prerequisites

  • An NVIDIA GPU.
  • NVIDIA drivers installed.
  • Docker installed.

Install BI CLI

There are several ways to install the BI CLI, depending on your preferences. Below we list them in order of convenience.

Automatically

bash -c "$(curl -fsSL https://home.batteriesincl.com/api/v1/scripts/install_bi)"

Manually

  • Download the install script curl -fsSL https://home.batteriesincl.com/api/v1/scripts/install_bi -o install_bi
  • Validate the script contents and make sure it’s safe for you to run
  • Run the install script: chmod +x install_bi && bash install_bi
  • Verify ls -al ~/.local/bin

By Hand

You don’t like curl and bash being in the same breath, and that’s ok. I’ll hold your hand.

  • Go to Batteries Included’s Releases Page on GitHub here.
  • Find the checksums.txt file and the archive for your platform and operating system.
  • Check the sha256 checksum.
  • Unpack the archive into ~/.local/share/.
  • Create a symlink to the newly downloaded binary in ~/.local/bin.
  • Add ~/.local/bin to your PATH if it’s not already there.

Setup NVIDIA Container Toolkit

Next we need to configure the NVIDIA Container Toolkit. To do that Batteries Included provides a setup command. Run it:

sudo bash -c "$(~/.local/bin/bi gpu setup-command)"

Verify (Optional)

Now if you want to verify your installation, and see the results simply run:

~/.local/bin/bi gpu validate-nvidia-ctk

This validates:

  • NVIDIA GPU detection
  • Container toolkit installation
  • Docker daemon configuration
  • GPU access from containers

The validation should pass with green checkmarks.

Start Batteries Included

Finally all that is left to do is start the local cluster with NVIDIA support.

~/.local/bin/bi start-local

Use the GPU

Now by default the Batteries Included platform will configure our batteries for GPU usage. We have some example projects that you can explore. Simply create a new project and select import from snapshot to try some of our example AI projects.

If instead you would prefer to build yourself, two great places to start are:

Ollama

  1. Go to AI → Ollama.
  2. Create an instance. Select “Any NVIDIA GPU”.
  3. Deploy models. They will use the GPU.

Jupyter

  1. Go to AI → Jupyter Notebooks.
  2. Create a notebook. Select “Any NVIDIA GPU”.
  3. Your code can now access CUDA.

How It Works

The start-local command automates everything:

  1. Installs bi to ~/.local/share and links to ~/.local/bin.
  2. Finds NVIDIA GPUs. Validates your setup.
  3. Creates a Kind cluster.
  4. Configures each node to expose the GPU to Kubernetes.
  5. Installs the NVIDIA Device Plugin and Node Feature Discovery.
  6. Configures Ollama and Jupyter for GPU scheduling.

Troubleshooting

Validate

Check your setup.

bi debug validate-nvidia-ctk

This checks GPU detection, toolkit, Docker, and runtime configuration.

Common Issues

  • No GPUs found: Does nvidia-smi work? Check your driver install and make sure that everything boots cleanly with dmesg | grep -i nvidia.
  • Docker errors: Did the configuration get changed correctly?
  • Kind errors: Is your machine overloaded or under heavy load?

Next

You have local AI. With enterprise power. And full control.

Related docs