Getting started

Copy Markdown View Source

This guide walks through installing ExAtlas, configuring a provider, and spawning your first GPU pod.

1. Add the dep

# mix.exs
def deps do
  [
    {:ex_atlas, "~> 0.1"}
  ]
end

For the orchestrator and LiveDashboard features, also add:

{:phoenix_pubsub, "~> 2.1"},
{:phoenix_live_dashboard, "~> 0.8"}   # optional

Run mix deps.get.

2. Configure a provider

# config/config.exs
config :ex_atlas, default_provider: :runpod
config :ex_atlas, :runpod, api_key: System.get_env("RUNPOD_API_KEY")

# Opt-in orchestrator (one-GenServer-per-pod supervision tree)
config :ex_atlas, start_orchestrator: true

Resolution order for the API key:

  1. Per-call api_key: option.
  2. config :ex_atlas, :runpod, api_key: ....
  3. RUNPOD_API_KEY env var.

3. Spawn a pod

{:ok, compute} =
  ExAtlas.spawn_compute(
    gpu: :h100,
    image: "pytorch/pytorch:2.5.0-cuda12.1-cudnn9-runtime",
    ports: [{8000, :http}],
    cloud_type: :secure,
    auth: :bearer
  )

compute.id                # "pod_abc123"
compute.status            # :running
compute.ports             # [%{internal: 8000, external: nil, protocol: :http,
                          #    url: "https://pod_abc123-8000.proxy.runpod.net"}]
compute.auth.token        # preshared key, handed to the pod as ATLAS_PRESHARED_KEY env var

4. Terminate

:ok = ExAtlas.terminate(compute.id)

5. Next steps